Introduction – August 2025

These days, our competitors are producing AI features, and since AI is the cool thing that all corporations want to promote alongside the claim they have “cutting-edge innovation”; we want to follow suit and we want to rush an AI product out.

If we make it too advanced though, it will meet the requirements for a 6 month (or more) legal approval process, so we want to keep it basic.

In recent years, even though we have struggled to get software out in a reasonable time-frame, they want this feature to be out after 3 months at most. The initial implication was that they wanted it finished by the end of year so would release straight after Christmas.

The lead developer reckoned the initial set up, which includes all the DevOps tasks, would take 3 weeks. Creating a repository is a few clicks, then there’s the pipelines to create, and all the “tooling” like installing and configuring Jest and Linting on the application side. I know Terraform is fiddly and the approval process for the AWS user permissions etc takes a while due to the process, but 3 weeks seemed a bit extreme to me.

The flow of the planned AI application was initially as follows:

We have an old product SystemNow, and its (eventual) replacement ProjectFutures. ProjectFutures is currently in use as a “Companion App” with minimal features. ProjectFutures runs as a browser based app. They don’t communicate effectively with each other in my opinion. So the old product SystemNow opens a new screen which then triggers opening a new page on the new product ProjectFutures. Then the user can hit record on ProjectFutures, upload the transcript to the 3rd-Party, who then sends back the transcribed audio in an annotated text format. Then we transform that data to then display inside SystemNow.

Reading through the architecture, it seems crazily complex, that the speech-to-text process involves sending the data to a 3rd party API, it then comes back with a processed transcript and we then send that to the cloud to send back to a different program on the same machine (this part was later changed to send a local message to the WebView component). There can be some kind of delay as well though; so the user could be waiting, but has to keep the program open.

Then what happens if there’s errors at each of those stages? How easily can the user recover if they have recorded a 10 minute clip but haven’t filed the data due to an error? I’d imagine once that happens a single time, then the user will stop using the feature.

I’m skeptical that this would save the user time, but might make their note-taking richer as long as the 3rd party actually sent detailed notes back.

To keep the company and product anonymous, I don’t want to disclose the actual product. To imagine the use, picture a customer having a consultation, in person or telephone, with a house designer. The designer will create a “consultation” which can have multiple pages, one for each room. Most consultations could just have one page, but multiple pages consultations are also used. Once they fill in text for each section in a page, they need to switch the free text entries to actual products. So if they have typed “white toaster”, they see some product suggestions, and then select an actual product in the system, which replaces the free text with a “code badge” eg [Russell Hobbs 27370 Eden 2 Slice Toaster] which makes it read-only. They can delete the “code badge” or swap it for another product, but cannot modify part of the text. This has a product ID associated with it, and can have “qualifiers” which are slightly different product variations without requiring a different code e.g. colour options for white and black.

So our AI speech to text feature aka “Scribe” will replace all (or most) of the typing. It listens to what the consultant and customer discussed, then generates the consultation pages and sections, and suggests product codes for each section. The consultant then just needs to verify each section looks correct and do the product coding.

Week 1 – 18th-22nd-August

Day 1: It seems Dave (Lead Developer) and Liam (Team Manager) have already created quite a few User Stories (aka “Work items” or “tickets”; where the requirements are supposed to be described as user-focussed) in meetings they had separately from the team. 

The idea was that the team could go into a Refinement meeting and focus on Story Pointing (estimating) but the disadvantage is that Jonny fully understood the requirements and the rest of the team didn’t.

Story Pointing is where we assign a score based on the predicted size using the fibonacci numbers 1,2,3,5,8,13; where 1 point is the smallest task. It’s a team vote so the most popular or average score is supposed to be chosen. My frustration is that we often scored too high, influenced each other’s votes, and didn’t consistently choose the average vote; so estimating always seemed like a waste of time in this team.

The initial User Stories were mainly the basic set-up, the prerequisites for development, so have nothing to do with user features. It’s setting up the repositories, pipelines, and configuring SonarCloud. I often wonder if these should be classed as User Stories.

For one of the set-up stories, I stated “it is a 1 point and we don’t need to discuss it, just move on to the next ones”. The team decides to vote anyway. Lucy chooses a 3. Then chickens out and comes down to a 1 without explaining why she thought it was far more complicated.

James then drops off planning.

Lucy chooses 2 for the next one, but Dave chooses 3, so Lucy then makes out she meant to choose 3 all along. Now we have two votes for 3 and my vote for a 2, so now I am outnumbered. I managed to get them to change to a 2 though.

For the next one, we all choose 2 apart from Lucy who chooses 3. She comes down to a 2. So that is 3 chickens in a row.

James comes back after 45 mins; what was he doing?

Week 2 – 25th-29th August

We now have the two code repositories created; one for the Client-side (UI) and one for the Server-Side (API) along with all the PR checks via GitHub Actions, and CodeQL.

There’s a debate of which 3rd-party provider we are using. Surely we shouldn’t be starting the project without API keys licensed?

We get assigned another 3 developers who are contractors. One contractor will join the business later. Two are currently employed with us but were originally assigned to another team but their project got cancelled. Is there a chance our project will get cancelled? It’s slow progress and the requirements aren’t locked down.

Everyone seemed to take one day’s leave at short notice but there’s not too much going on so not a big impact.

James declares he doesn’t want to do development work, but wants to focus on testing. Testing what though? we hadn’t got the repos fully set up, and hadn’t written any code to test. It will be a while until there’s any real testing work.

Week 3 – 1-5th September

The two contractors, Marek, and Andrei are sitting around reading documentation. They do ask for work but Liam (Team Manager) is unsure what to suggest as he didn’t expect to get them so soon.

I told Marek that our ReadMe in the API repo was a template and can be ignored. There’s literally no features so nothing to build or configure despite what the ReadMe says. Later he declares he doesn’t have the AWS access required as stated in the ReadMe. I tell him again “it’s a template”.

The next day Marek says he can’t install something as stated in the ReadMe. I tell him again “it’s a template”.

Andrei keeps starting meetings early when Outlook reminds him there’s a meeting in 5 mins. One of our meetings overruns so the manager declares a 10 minute break before we start our next meeting. Andrei instantly starts the next meeting.

We had a meeting with Sam, the UX Designer. The user flow is absolutely massive with a few questions over confidentiality and security etc. Also what happens if there’s connection problems during each phase. Does it carry on recording, and how long for? The user can’t really wait around, so can they quickly revert to manually entering the data instead?

Since we only planned work for 4 people, we don’t even have work to give the contractors. We need more refinement meetings but this means the entire team aren’t working.

In the estimation session, Dave (Lead Developer) shows how easy it is to set up Release Please which generates release notes. I say, “isn’t it quicker just to check it in now than to write the User Story for it?”. Dave agrees, but he creates the User Story and asks to story point it anyway. To my surprise, the team votes were 1 2 2 3 3 5. So we literally saw Dave do the work within seconds and yet the only person voting 1 was me! Lucy said she “went high just to be safe” (voted 3), and Marek said he “went 5 because I’ve never done it before“. Mate, Dave showed the config file from another repo. It’s like 10 lines to copy and you change nothing.

I suppose different teams vote differently, but from past experience 5 usually means 5-8 days.

What is the point even of Story Pointing when people are voting 2,3,5 for 5 minutes of work? Can you trust any estimates? I feel like it’s a complete waste of time.

By the end of the week we have added Sonar, and set-up Playwright testing framework on the UI side.

Week 4 – 8-12th September

We had meetings all day Monday. One was a demo about the prototype that was written by AI from Chris (Framework Lead) and Mark (Architect). We had another refinement session.

Marek asks “What is ProjectFutures“? It’s the start of week 3 for him but surely he should have been aware what product he is working on. I thought Liam (Team Manager) would have given them an Induction; it would have been the sensible thing to do.

Later in the week, we had a few more long refinement meetings. The contractors are still pointing high because “they’ve never done it before”. For one User Story, James gave a link to another repository which showed the exact method we had to add for authentication: 1 2 2 3 5 5 5 were the votes. Surely it is a 1!

Week 5 – 15-19th September

I start working on adding the Automated Testing framework; Refit to the API, although we don’t have an API, so what am I even adding!?

There was a major NPM security incident where malware was infecting several packages. On investigation, it didn’t affect us because we didn’t use any of those packages, but NPM was disabled by our security team “as a precaution” and we were told to just do some “personal development“.

Week 6 – 22-26 September

Another contractor, Ivan joins us. The NPM incident is still ongoing so we kept on doing “personal development”.

Week 7 – 29th Sept-3rd October

“You guys fly through work so this incident has come at a bad time”

Jean (Product Owner)

We are in Week 7 and don’t have any features complete. I suppose Dave (Lead Developer) was correct about it taking a long time to set up, but it really shouldn’t. I think it’s because we keep splitting tasks down and spreading them throughout the team, so one person does an hour-long task but then drags it out to fill the day. Yet one person focussed on it could just whizz through.

We also spend too much time planning which is also because we have broken tasks down too much. So we end up discussing what the scope is; where one task ends and where the next one begins, but the second one is instantly blocked by the first as a dependency. So it’s all just stupid. If you can’t get two developers working on different items at the same time, then what’s the point in splitting the feature down like that? 

I’m still trying to get Refit working in our repository but I still don’t really know what the objective is since we don’t have an API. I just want the boilerplate code I guess, but our internal package APIAutomation was showing an error about missing data.

Week 8 – 6-10 October

We had another bizarre planning session. One item had the following scores: 2 3 3 5 5 5 

James, who voted 3, says it’s most likely a 1. Why did he vote two scores higher? 5 is the majority, but then people admit it’s probably simple to implement a config flag, it’s just we haven’t got experience creating config flags for ProjectFutures. What is the point of voting if everyone doesn’t believe in their vote apart from me. 2 shows it’s easy but allows more scope if we have trouble. I suppose I should have voted 1, but voted 2 to attempt to be more in sync with their numbering.

I’m still struggling with the APIAutomation package error.

There seems to be an idea that we won’t actually open our Scribe in the “Companion” mode. It will now be launched as an embedded Webview2 control inside SystemNow. I’m not sure what the reason is because this surely means you can’t place it on a separate monitor and we will be really restricted in terms of space.

Week 9 – 13-17th October

I took a week off. Perfectly timed really since not much was happening.

The lead automation tester at the company admits that there was a bug in the APIAutomation package and I was the first to see it.

The rest of the team are adding basic UI components and test tools like Storybook and Playwright.

Week 10 – 20-24th October

I had to reset my laptop at the request of Group IT, which would then have all the new security policies. However, the update failed to reinstall, so I was completely screwed.

I called IT Tuesday morning, and after investigating if anyone else had the same problem, they asked me to come in tomorrow to reset it. At the office, they have a couple of attempts and don’t succeed, so I head home. They later fix it so come and deliver the laptop to me.

There’s some weird restriction about having to wait 24 hours before we can download applications from the Internal App Store. With meetings like the End of Sprint review and Friday’s Retrospective, I’ve gone the full week without doing any work again.

When I think back to the original plan for the project, I suppose we have missed the 2 month project deadline that the managers wanted. It’s an absolute joke that we don’t have any real functionality to demo. Liam (Team Manager) keeps saying it’s because of the NPM major incident that prevented us from doing work, but we weren’t making much progress before that.

Week 11 – 27th-31st Oct

Liam (Team Manager) keeps mentioning “version 0.5” which I get the impression was an outcome from a meeting during my week off.

So the first thing to start the week is yet another planning meeting to discuss the items in scope for version 0.5. Not only do we have extra meetings about descoping work, there’s extra work for temporary config flags. Due to the fact we are behind, we want to rush out a demo to a small group of users to get feedback which actually comes at the cost of delaying the main project by a couple of weeks.

One of the User Stories is to install the Fluent Validation package. We wasted more time talking about it than it would to take to install. I think items like that are dumb because a User Story should deliver functionality to the user, and this is just Step 1 of delivering some value, and it’s an implementation detail. It’s to check that part of the message content is valid, then we will send a Rejection if not.

I spent most of the week reconfiguring my laptop after the reset. All kinds of things hindered me setting up my laptop. The AppStore triggers installations off in the background so it is hard to know if and when they are running. There can be hour delays, and required restarts. Having to set up things locally and on WSL adds to the pain, and restoring the Linux image took ages.

The team collectively are working on Launch Darkly (Configuration Flags), and some Message Validators.

Week 12 – 3-7th November

Andrei kept on talking about adding a DateTimeValidator even though for his story, he just needed Date (not DateTime). When we asked him why he was doing it, he kept on saying he had “only ever used DateTime before and wanted consistency through the application”. We are like “no mate, some things need Date, and some need DateTime” so we eventually need both types of validations.

We have our End of Year Reviews to fill in at short notice so this is another distraction.

We completed a high number of Story Points this week, I believe around 45 when it was more like 20 in previous weeks.

Liam (Team Manager) asked what caused this increase. Then Andrei joked “GitHub Copilot“. Although he did increase productivity, it definitely took longer to review his code; it had loads of scope-creep and complexity. I keep thinking he gets ideas in his head like wanting DateTime then begins doing more, or the wrong thing – to what was asked.

I wasn’t sure why our velocity actually increased. It might just be that all the basic set-up is complete and we were doing work which was small and the Story Points were inflated, so we were getting 3 or 5 points for items that were really 1 or 2s.

Week 13 – 10th-14th November

The week started with an incredibly long planning session. We spent way too long debating how many Story Points to bring in. Do we go for 45, or try to use an average, so plan around 33 points?

I saw this quote from one of the managers. I think they were talking about costs. I hate all this business jargon:

“We’ll design a good enough path for both pppy and pupm taking into considerations user facing Flow and internal ops constraints that can be ready by Q2 2026”

Andrei and I have been adding Validators. Marek added one as well which is only his 3rd PR in 11 weeks.

James is still working on hosting issues.

Liam (Team Manager) announced that we were reducing the number of Microsoft Teams chats associated with the project. This was the first time I had heard that another team had started their work. I knew my mate Adam was going to be working on the SystemNow side of things at some point because he had told me. I didn’t know that he had started, and didn’t know what his team was called. When I asked him about it, he said he wasn’t aware his team had their own Microsoft Teams chat.

There was a demo arranged Friday to show the project so far. On the call, there were Software Architects, Safety, Marketing, Data Scientists, all kinds of managers at various levels in the hierarchy. During the introduction, the managers were constantly hyping up the features and overpromising the project. The Marketing staff seemed really excited since it is a new product  to sell. After all that hype, the development team were basically like “here is a button“.

These demos seem to be expected to be weekly but how much progress are we going to make in a week? and it’s another distraction that we have to plan for the demo. It’s like Thursday afternoon will be wasted just to put a demo together for Friday morning.

The 3rd Party that will provide us with the embedded application that handles the recording and transcription is called VerbatimEcho. They are a small company but claim to be a market leader in other English-speaking countries.

Week 14 – 17th-21st November

Dave (Lead Developer) reckons we should make sure the “elements” in the JSON are ordered, but the requirements in the user story were written with the implication that unordered elements were invalid and should be rejected, rather than our code correcting it and accepting it.

I asked the question on Microsoft Teams, mainly for the Architect to confirm, I also asked which User Story we can pick up because most of the items we have seem to be blocked. Both questions were ignored.

I started looking at a User Story that involved rejecting the message if there were extra elements that weren’t in the schema.

I started writing a unit test and the GitHub Copilot added some schema parsing code using Newtonsoft. I wasn’t sure if we could use Newtonsoft with the new security changes. I searched for it in Nuget Package Manager and couldn’t find it. I looked on JFrog directly and found something similar named but couldn’t tell what it was since there was no link to the actual package. I noticed that Netwonsoft now have moved schema validation to a new package and they charge for it. So I was kinda stuck not knowing what to use.

I’m finding it progressively frustrating that the only reason we have our Daily Stand Up at 10AM is because of James and he keeps turning up late. It’s 3 hours into the day for the Contractors so I think sometimes they sit around waiting to be told what to do. It would be very beneficial for us to have the meeting earlier at 9.

A big turnout on the Friday demo; 36 people. Adam said they have nothing to show but then Jack, the tester in their team, showed a JSON generator tool to use rather than actually using the VerbatimEcho app. Our team demoed a tab control, and some error messages. I’m sure all these people join this demo wanting to see some revolutionary AI, and we are showing them basic UI controls.

Week 15 – 24th-29th November

On Friday at 6pm, James decides to tell the team he has annual leave this week, and he hasn’t handed over the work that he was assigned, so now that is probably blocked.

Mark confirms the schema story I spent most of last week on was another AI hallucination of ours. Jean (Product Owner) has been using Microsoft Copilot to generate the User Stories.

With Andrei changing how error messages work, he has taken the initiative to update all the old messages and remove the old code. Some of the helper methods that were suspiciously AI have now been deleted.

There was one message that said 

“needs to be in one of the following formats: <formats go here>” 

but we only ever specified a single format it can be. I told him to create a separate message for when it is single.

He changed the message so it would read like it will specify one format, but we still allow an array to be passed in so it looks like it will display multiple. I asked him why he did this, and he said “although there’s no requirement to support multiple formats, the code is there if we do“. This makes no sense because if we do have the need for multiple formats, we would have to change the message like I suggested. However, since there is only a requirement for singular format, then why don’t we just code for that, and support multiple later? (but I don’t anticipate it ever being needed anyway).

On the side, and on a similar note: Andrei has made some custom error logging framework. He allows multiple errors with an array, yet we only produce 1 error, then the validation stops processing. He also has a property for each individual error to hide it from the user, then an overall property to hide all errors in the array from the user. When asked why he had done this, he said that “we might change the validation to allow multiple errors” and “It might be useful to hide errors from the user“.

So there is no User Story for it; he has just invented some work and showed off. For both validation and error logging; he is saying he has implemented them in a flexible way but both will require more work to get them to work that way. They look like they support multiple items but they don’t.

I don’t know enough about API development, but I can’t imagine something like this wouldn’t exist in the framework natively. So he is doing what he wants and creating extra work instead of just doing the planned work. It’s getting so frustrating.

We are quickly getting through our assigned work though. I don’t have much experience with UI but with GitHub Copilot, I quickly got my item done which was worth 5 Story Points. Although I thought I’d got it done; but it did need several minor tweaks which then delayed it.

The architect tells us that the schema will change. So the work we did to validate the schema will need tweaking.

Chris (Framework Lead) is apparently more involved now, although I think it is for framework changes. Although the work is driven by our use case, it will allow any app to work in this embedded way. Chris is due to finish 16th January.

There was talk of a March deadline for the project.

Week 16 – 1-5th December

James is back from his week off. He turns up at 10:05 though in classic James fashion. I ask him a question at 10:56 and he responds at 12:13.

I wanted to ask Sam (UX) about a few minor tweaks to the UI but he was off.

We were talking about Version 0.5, and James started kicking off, which I thought he was talking himself out of a pay rise during this End of Year Review period. “I wasn’t aware of that, so there’s not much vision and direction to the team“. Liam (Team Manager) was a bit annoyed.

Liam was talking about getting more people on the project to speed things up.  It’s the classic management question, thinking you can speed progress by throwing more people at it. Liam sounded aware it wasn’t a great idea if it would be disruptive to train people up, and split more work across more people. However, maybe we can delegate certain tasks like the DevOps side?

Liam mentions that Alan (Head Architect) is working closer with Adam. Sounds like he’s not happy with his estimate. Alan basically vibe-coded a prototype in a week. Adam reckons another 4 months’ work is required on top of the 2 months Adam has done so far.

Sam (UX) is going on about mono tertiary buttons. I don’t know what that means. Turns out the design has changed a bit and Dave (Lead Developer) doesn’t like certain things so I have to rewrite my control I made last week… Again

Christmas Party removed a day from the week.

Managers have demoed a UX Prototype to some users to get them hyped. Probably end up getting them too hyped when we deliver something different 6 months later!

Week 17 – 8-12 December

We had a User Story to create a database, but no one had come up with a design. Dave (Lead Developer) lists the columns we need then begins trying to rearrange the tables so it’s easier if we then “partially migrate” a table to DynamoDB or something daft.

I don’t understand this idea, but assuming we did go ahead with it; I don’t think it will be any easier because you’d still be dropping 1 table and modifying the other table to add a URL column to point to the new location. So sacrificing the design now means we have a potential overhead until we do the migration, but then we might not ever do this migration.

Then he wanted to add audit fields in there like Update/Created By/On. So 4 new columns. I didn’t think you’d need audit fields in both tables because the initial write would add entries to both tables, then additional rows for that entry would just add to the second table; so you can infer who added the entry by only looking at the first entry in the second table.

Regardless, the design would come from Mark, our Architect and database expert, so why are we even discussing this in a Sprint Planning meeting? The User Story should just be “Create database” and the exact design gets added later. Despite my protests, we carried on talking about the design ourselves.

Nearly an hour went by in this Sprint Planning meeting and we hadn’t even planned 1 item, and we needed 7 to keep people busy for now.

Andrei said it made sense to create a 4 column composite key. Wouldn’t that mean that you have to add 4 duplicate columns into the second table that it foreign keys onto? That sounds bonkers. Andrei claims to have experience in database design.

Andrei then suggested using a GUID for the primary key rather than an incrementing Integer. I don’t see why a GUID has any advantage over an Int, but Ints will have lower storage memory cost and be easier to read.

After 1.5 hours, we had 13 points planned in (we wanted about 30).

On the next item, it was Marek’s turn to chicken out, who voted 5 “I am happy to go for 3 or a 2“.

I noticed James was late to every Daily Stand Up by at least 5 mins. Then he’d disappear for long periods of time throughout the day.

I pick up an item to do with authentication, but we need some Azure Client ID and Secrets which only managers can create. I requested one from Liam (Team Manager) but he was busy. The next day he was off. The next day he was busy again until the afternoon but I managed to get some of his time. Dave (Lead Developer) then decides we can reuse some credentials for now so Liam wasn’t even needed.

Week 18 – 15th-19th December

We’ve had many Town Hall Meetings that have caused disruption. 1 hour long presentations of business guff.

James still keeps being late. We moved a Daily Stand Up to 10:30 because of a Town Hall meeting clash but he was still late. The next Town Hall a few days later meant we moved it to 11:00 and he was late again. It illustrates that Liam (Team Manager) does a bad job if he isn’t having words with James.

I don’t recall the last time I saw Dave (Lead Developer) actually work on something on the board. I never understand how we get random issues like a single unit test will be failing, and Dave is the one to notice it. He will then fix it and only tell people when he has the PR for it. Then there was a package he replaced due to some security issues, then another package he replaced because it had a bug that sometimes affected us so took the initiative to find a better one. It’s almost like he is inventing problems.

For my authentication work, I showed Dave what I had done so far. I created a temporary method to create a bearer token and Terraform to store the secrets. After looking at the current authentication package and API we need to call; Dave says the package actually handles the authentication for us so my User Story can just be binned.

James is taking ages to create the Transcript UI. I thought he was good with UI stuff but I suppose he is really a Tester and not a Developer. He also keeps being late so maybe he isn’t working. He keeps claiming GitHub Copilot is generating bad code, and he is struggling to sort the CSS out himself. I am very doubtful of this unless he isn’t asking the correct questions.

Marek and Andrei are doing some database work but it sounds like there’s a bit of crossover or dependencies with their work. They also discuss how to do unit tests which shouldn’t have a dependency on a database. Using EntityFramework you can use the InMemoryDatabase if you want some simple Integration tests, but then Andrei claims it doesn’t work with DynamoDB or something. I thought we were using SQL anyway, so I’ve no idea what is going on.

Marek asked about if he should check in the DB schema changes but I said something needs to be checked in to create the tables on the different environments (INT/PRD) because they aren’t gonna create themselves. No one in the team responded to me; an awkward silence as if I had said something stupid.

They keep using the term “Migration” as if everyone knows what that means. It does seem like it is a EntityFramework name, but I hadn’t heard of it before. It sounds weird to talk about Database Migrations when we haven’t created the initial database – there’s nothing to “migrate”.

Week 19 – 5th-9th January

Most of the team were away over the Christmas period. The 2nd was a Friday so pretty much everyone had that off as well. So a full 2 weeks are basically skipped.

The 5th was the real start of 2026. Liam (Team Manager) said James was off so “everyone was here on the call“. I pointed out Lucy was missing. Lucy then joins. James turns up 15 mins late. Great start from Liam; he is in charge of approving annual leave and has no idea who is supposed to show up to work.

7 items are still waiting for reviews to complete.

4 are Ivan’s and he should have had them ready over the last couple of weeks but they actually still have outstanding comments to resolve on them.

The Sprint Planning session was the usual debating the UX for a control so we planned nothing in by 11pm. We agreed to focus on clearing down the current items on the board at Liam’s request who seemed angry at the delay. But it’s really all on Ivan, so we are sitting around not doing anything. Dave (Lead Developer) said there will be a proper planning meeting before 2pm. It didn’t happen.

On Tuesday’s Story Pointing session, there was one item where everyone voted a 5 except me who voted an 8. Since the team is in agreement, it makes sense to go with 5. However, Andrei then says it could be an 8. It still makes sense to go with 5 because it’s still the majority vote. Dave (Lead Developer) then comes up with the idea of creating a “Spike” to investigate the scope of the item, then we can more accurately point. Absolutely nonsensical idea.

There was another item where we discussed for a while. The majority vote was 2, with a couple of votes for a 3. Dave votes 2 too. However, as Team Lead, Dave overrules and gives it a 3. There’s no real consistency in the process.

Another story involves clicking a Text Field and if the transcript panel is open, it then scrolls to the evidence in the Transcript and highlights the appropriate text evidence. I thought it was pretty standard, like 5 points. However, we then split it up, and score the scrolling 5 points, then the manual controls (next/back buttons) was a 3. So we are splitting work up too small, pointing really high, and we cannot often share the work out because there’s a dependency on each other.

In today’s jargon news, there’s a call for the “Provenance Derivation and Semantic Match Criteria within class 1 criteria“

I had my User Story mostly complete by Monday evening but then I made some changes suggested by Copilot on the PR but they didn’t work. I struggled to understand the Jest output and I couldn’t see a way of debugging the tests. GitHub Copilot stopped working with some random network issue but I could connect to everything else.

On Thursday, I had it completed, so I picked up a new item and got it done within an hour. I encountered a random test failure but couldn’t understand why my changes had any impact on it at all. A bit of a weird week for me.

Week 20 – 12-16th January

James is on annual leave and hasn’t finished his story so Dave (Lead Developer) agreed to look at it Monday.

I saw a few PRs from Dave which were fixing something to do with a build (code coverage reports) – completely unrelated to what James was working on.

I picked up a related story which was to create a navigation control. It was really simple, just a coloured border around some text with 2 chevron icons to interact with. So I got it done within a couple of hours. It had a dependency on the item that James/Dave was meant to do so I think some of the functionality is blocked.

On my other item, only Lucy reviewed it, and pointed out some changes which were a good suggestion, but we could have just fixed it on the next related User Story rather than delaying this one.

On Tuesday, I pushed up another change for Lucy but I was mainly idling. Daily Stand Up was meant to be at 10 but we had a “Timesheet meeting” so moved it to 11:15 which seemed a bit excessive especially when the timesheet meeting didn’t last as long as scheduled, so I idled a bit until the Daily Stand Up. Then people were talking about all kinds of things so it ended up lasting around 35 minutes. So by 12pm no one had done anything productive.

We seem to organise the work in the least efficient way, have too many meetings about it, at inconvenient times, then people just idle between these events.

On Wednesday, there was a long call, estimating/refining work for next week.

As a team, despite the distractions, we ended up making quite a few changes throughout the last 2 weeks and got quite a few items checked in on Thursday ready for the Friday demo. These were some updates to API Schema changes, some controls to scroll through the Transcript evidence, some database changes, and a few tweaks to the build pipelines. Enough to talk about but still not exciting for the managers.

Week 21 – 19-23 January

In contrast to last week, not much seemed to happen.

I worked on the Transcript Navigator control.

Dave (Lead Developer) made a Mock site improvement.

Lucy finished off Ivan’s work.

In Week 16, Liam (Team Manager) asked about increasing the team size. We actually went with the idea of delegating some DevOps tasks to someone. So Richard was now in charge of tweaking or configuring deployment. That’s one less distraction for Dave.

Week 22 – 26-30 January

I created a drop-down menu. For some reason, the usual “fire event” in the unit test doesn’t work and I had to “use user event”, so that was a day wasted in the confusion. Lucy added a few comments on my review, then after I addressed them, she added some more for something completely different. Valid points but frustrating she keeps moving the goalposts recently.

On Wednesday, we were told that the Senior Managers wanted the release one month early. Do they know how software works? It’s done when it’s done unless we cut out features that aren’t Minimum Viable Product (MVP), but A, we already cut the scope, and B, we will release something more underwhelming so it’s a harder sell.

Maybe there was some pushback, because Liam (Team Manager) declared the date as 27th March so the old claim of a March deadline is still accurate.

In the Story Pointing session, for one item the votes were:

5 5 8 8 13 13

So a complete tie, but 8 would be a good average. Lucy tries chickening from 8 to a 13 because Dave (Lead Developer) had voted 13. However, Dave says we should average to an 8.

Week 23 – 2-6 February

On Wednesday, James said he’s been waiting a week for VerbatimEcho to answer a question. Liam (Team Manager) said it was only sent Monday and VerbatimEcho had the day off. Sounds like the typical communication problem where we have to tell a manager like Liam, and hope that he then asks VerbatimEcho. It would be better if we could just message them directly.

I keep finding that User Stories are vague and not sure of the actual scope. I picked up one that involved sending messages to SystemNow but we don’t have the integration yet to test it so it’s all theoretical.

Week 24 – 9-13th Feb

No real progress from me. One User Story kept on running into merge conflicts as the team checked in their items, then GitHub Copilot was re-reviewing it, and coming up with new comments on code it had already seen.

Meanwhile I ran into problems with WSL, with networking failures and Extensions failing. I think part of it was that my current version of DotNet was then on the block list for my employer with their new strict policies.

On a call with Andrei, who claimed he could help with WSL problems, he insisted part of it was the “obj directory”, but he is wrong because that’s for the compiled DLLs and this was an error on VS Code load, and nothing to do with launching our app or compiling it.

He then said he writes code on Visual Studio on Windows, and then copies the folder into Ubuntu to run via VS Code. I asked him repeatedly why he does this and it was just that he is “unfamiliar with VS Code”. Yet he is still using VS Code to build and run the app isn’t he? and this is an incredibly cumbersome and tedious process he is using.

With Dave (Lead Developer)’s changes to a Mock API, we now need a different command to launch our app. We also need to specify Database Config on the API side even when a database isn’t needed (I think this was Andrei’s fault). It annoys me when colleagues change the process but then there’s no clear guide to getting everyone set up.

The first pilot customer is live in pilot with the “Partner API Scribe version”. No one in the team seems to know what that means. They definitely aren’t using our API and app and we thought there were only 2 teams on our project, and the other team doesn’t know what it is either.

Week 25 – 16th-20th February

I’m trying to remember how our Refit Automation even ran. I added some ENV variables the GitHub workflow used but it was still erroring. It was calling out a specific variable so I added it and tried again. Repeat about 5 times. Then it moans that the INT URL isn’t available but I want to run this locally, but how do I do that?

I tried to test the API with the JSON files we had on Confluence which were examples that matched the schema. One of the objects cannot be parsed. I manipulate it to match our code. I get loads of errors about invalid date formats and missing attributes. 

This is a great example that if you mock everything in tests, you can end up creating nonsense. Why aren’t we using the JSON from Confluence in our tests? The JSON format is a key part of our project and has been massively wrong all this time. Ridiculous.

The next day Dave (Lead Developer) writes almost a blog post on the issues and the team reacts surprised like they didn’t know about it. Dave seems to be getting credit for finding this mess but it was me that spotted them all.

In today’s “Andrei being weird” story:

Lucy creates a PR with some hardcoded test data in a control she is working on. She declares it in the description. Copilot Code Review comments on it, stating that despite her declaring that it is temporary data in the description, it is better to be declared in an inline comment in the code to state it is temporary i.e. a TODO comment. Fair enough. 

Andrei then comments saying “add these strings to the Translation file” as if at some point in the future we will translate it to German. It’s temporary code mate. He hasn’t paid any attention has he? He is just looking for text strings to write that comment on. A few weeks ago he correctly flagged up some hardcoded button text that we needed to move to the translation file, but here he looks like an idiot for saying it.

In Week 17 we had the idea of adding 4 audit columns which seemed excessive to me. This week, Andrei has added initiatedByUser column to the Database even though we have “CreatedBy”. Dave (Lead Developer) points out it’s redundant and asks where the requirement came from. Andrei says we need to “stay flexible and consistent with the database design. Relying on CreatedBy might create additional work later if it can be triggered by 3rd parties“. What is he on about? He’s gone mad.

On another PR, I noticed his database code doesn’t always update the UpdatedOn timestamp so I asked “if that was intentional because I didn’t know what scenarios the code was executed”.

He replies “you’re absolutely right!” then explains the exact scenario back to me, clearly an AI response. He then asks “can you confirm this is not expected?“

It’s frustrating because I’m asking him that exact question.

He then screenshots it and sends it on Microsoft Teams which angered me further. If he gives me a link, I can get to the review to have a look, but a screenshot means I have to navigate there myself. He then says “please resolve if it’s not relevant, or we can have a call“. How do I know it’s relevant? He hasn’t given me more info. He loves Teams calls. So I tell him if the code can be called when the same user edits, then their new UPDATED ON isn’t logged. The line needs to be moved out of the IF statement. I also asked why he is pasting AI responses to me when he hasn’t understood what it has written.

He then explains it’s only really auditing if the user’s display name has changed and we update their name. So that sounds fine then, my comment was irrelevant after all. However he then says “it needs more thought because the change tracker object is excluding them“. So I say “I’m happy to mark the comment as irrelevant, but if he’s spotted a problem then I’ll wait for him to fix it before I approve the change”. He immediately says that he “can’t see any issues with it”.

However 30 mins later he sends me some code on Teams which implies there is a problem but most of it is commented out so it’s quite unclear what the fix is. I ask him where the timestamp is updated because I don’t see it in the code he pasted, and he gives me some line numbers but there are no line numbers in his snippet, so I don’t know what he is on about. He asks for a call, I think it’s a wind up.

Week 26 – 23rd-27th February

I pick up a User Story that was something like

“as a user, I would like customer details in the API call”

The user doesn’t want that. The user knows nothing about API calls. This isn’t a User Story in my opinion but most of our User Stories seem to be like this. Also, it has no detail to be clear what the scope of this work is. What customer details do we need? Which API call is it included in? I stared at the code for a while trying to make sense of what it could mean. I eventually wrote a single line of code but is it correct?

I kept on running into environment issues. Sometimes I cannot log in, sometimes it’s slow, and sometimes says ApplicationID is missing so the app doesn’t load. Then the data I wanted didn’t exist, but did the next day but I hadn’t changed anything.

Dave (Lead Developer) keeps doing what he wants. Multiple surprise PRs each day for things not on the board.

Ivan is going on about Redux but we aren’t using Redux. Andrei is going on about “caching enums” but I don’t know what that means.

VerbatimEcho are still violating our schema. You’d think they would have implemented it pretty quickly and got feedback on it, but we’ve been coding against a mock app so we’ve rarely been testing it.

VerbatimEcho changed their URL so now we need to add that to the firewall and security aspects like CORS.

Adam says his team can’t merge into Main until next month and only then can we see if our project works. 26 weeks in and we have no idea if it’s going to integrate together. Also, surely we need to Integration Test before they put their code in Main? When it is in Main, the code will then be in the next release. We really have no idea how to release software anymore.

Week 27 – 2-6 March

On the 4th we found that VerbatimEcho have said they won’t be complete until 13th March. Adam’s team will be ready by the 17th March.

Our app doesn’t work in the Staging for some reason.

The Project Manager suggested we should have had Daily Stand Ups with VerbatimEcho much earlier on. Isn’t that her job to approve each of their milestones? You’d think there would be a contract and she would be in charge of managing the project.

Marketing Manager – “We are changing the game when it comes to AI – end users watch this space“

I keep running into environment issues which are a massive inconvenience to developing; app not loading, product search giving me 400 errors.

I picked up a bug to fix which needed debugging but no one in the team actually knows how to bug a Typescript app. They all just do “console.logs” like noobs. I managed to get it working after a bit of trial-and-error but annoyingly it launched a browser window without the Keeper extension so I couldn’t use the saved credentials to log in. Instead, I have to keep a different browser window open with Keeper logged in and copy each username/password/2FA across. Since the bug actually happens in another repository, I tried changing the package reference to point to the local files. However their repo refers to many other packages using this “workspace” keyword which then doesn’t build, presumably because they aren’t in the main repo’s workspace. Nothing is ever easy is it?

I later realised there were similarly named packages and I was actually updating the wrong reference. With the correct one in place as a file reference, I then got it working. Although I couldn’t get the debugger to work with it so had to resort to “console.logs” like the rest of the team.

Week 28 – 9-13 March

I realised that the issue was caused by one of our changes many months ago and seems specific to certain Product Codes and only when this property we added was set to True. Therefore it would only impact our project which is why no one found it earlier.

I then noticed that GitHub Copilot had added a comment on the original change which the developer ignored. I added the suggestion in and it worked. Although some unit tests were now failing and it looked important. With some refactoring, I made the change simpler and the unit tests passed.

The functionality in general seemed a bit janky though and the method was triggered multiple times when it didn’t need to. The main problem was a hardcoded 2 second timer which I think was to basically ignore some of these refreshes. I asked Copilot to fix it and it added even more timers! I asked a Premium AI model and it came up with a fix which was a bit copy and paste, but it works.

However, Ivan has been making large changes to this Add Product functionality. From what I understood, the current other issue is that you can select a Product Code, then go back to edit it; it hasn’t updated with your changes. Some Product Codes have more fields to fill in, and these weren’t set to your current selections.

I thought Ivan was fixing this “save state”. I tried our changes together and it wasn’t retaining the data or updating the data. Also when you clicked a new row, it also added the same selected product code there too which was obviously wrong. I told Ivan about it and he claimed it works. Then we ran into environment issues where the Customer Find wouldn’t work, or the Add Product picker wouldn’t work, or you just couldn’t log in. So that was a day written off.

We had a meeting to discuss outstanding issues and it seemed the other team members had found various issues. Some we needed to fix now and others maybe could be delayed until after the initial pilot release. Managers were raising concerns that we have never seen it integrated all together; it’s always been tested with mock data and we have interactions between SystemNow->ProjectFutures->Third Party, so there is a massive scope of things to go wrong between.

We had a demo where Alan (Head Architect) said our implementation was wrong. When the user selects a Product Code, then the text should be replaced. There was a huge message thread back and forth where part of it was debating about the behavior and others were debating about the word “replaced”. Alan was insisting that we never “replace” but I didn’t understand his point.

I can’t really picture how his requirements work in all scenarios. If you select a code then replace it, then what text does it then show? does it then revert to the original text? Then what about a standard example where the Product name isn’t as descriptive. So start with the text:

"High chair"

then they select

[Chair]

do we then display

[Chair]

then it's up to the user to type the "high" back in, or select the appropriate code.

Jean (Product Owner) asks how to log into our test system and says she has never seen any of the functionality herself. Week 28 and the Product Owner hasn’t seen anything.

Week 29 – 16th-20th March

We had Sprint Planning at 10 so that’s basically 1 hour written off. After 30 mins, we had an ad-hoc department meeting so we had to attend that. Then not everyone was free to resume planning straight away.

Rather than simply Sprint Planning, we ended up Story Pointing and/or prioritising items. We were going to drop one item from scope after discussing it for 10 mins. The discussion was that maybe it wasn’t needed for the Pilot release but was valuable later. Then Andrei said it was ready to review, he just hasn’t assigned it to himself and was working on it last Friday. Why didn’t he say that straight away?

Andrei keeps going on about the order of an enum and pasted a whole essay from AI about it. When looking at his changes, he has defined an enum to represent Gender with values F M I U but for U (unknown) should be 0 as that’s the default value. So instead of putting U first, he has then just specified values 1 2 3 0. When Marek asked why he couldn’t just put it first, he says that there’s some “migration problem with Postgres“. My thought was that if Postgres has its own enum with F M I U then now we have a problem because the database uses F as 0. Turns out we actually don’t store the number value, we only save a single character. So it seems fine, but then does the order of the enum actually matter then?

We then talk about another item Marek was currently looking at. We debated 3 or 5 points then shortly after the meeting, Marek sent it for review. So he could have just put some points on it without discussion because it was also already complete.

We were about to be ready to integrate SystemNow launching our Scribe app containing VerbatimEcho.

James: “With stakeholders wanting to access the app now via SystemNow and us yet to actually see it working ourselves, it might end up being a bit of a disaster. Is it worth mentioning to avoid any disappointment when it doesn’t work yet?”

Dave (Lead Developer): “we’re currently in the period of integration testing so it should be expected that we are finding issues.”

We had a meeting with all the teams to talk about the remaining work. Or that’s what I thought it was for. It turns out there was an issue where the other team need to refresh the browser 3 times for VerbatimEcho to load properly.

Alan (Head Architect): “why do you have to refresh the browser 3 times for it to appear?”

Jack: “Sometimes it is more!”

It completely cracked me up when he said that. He said it like it’s just an accepted thing. I mean, I suppose he has put up with it for a bit without previously flagging it as an issue but it’s a serious issue if that gets released like that.

Then we were told our App wouldn’t work at all in the Staging environment which was due to some missing config from the Users team so it wasn’t added to the list of “accepted products”. Then there was another intermittent networking issue.

Chris (Framework Lead) pointed out the communication flow was

SystemNow->Companion->B2C Auth->Companion-CA->VerbatimEcho->VerbatimEcho(Auth)->ProjectFutures-Users->VerbatimEcho

Sounds horrendously complex.

We did more of our classic story pointing. Four votes for 2, Three for 3, so you’d think the majority wins, but Dave (Lead Developer) went with 3. There was one later on which was even more where it was five votes vs two votes, but two won.

Dave is surprised that replacing a Product Code doesn’t work. I literally told the team this morning on the Daily Stand Up, and everyone apart from Lucy seemed surprised at it. Ivan kept quiet so I didn’t know if he was actually there, but I told him to fix it because he didn’t have anything to do and I needed to finish off another item by addressing the code review feedback.

Week 30 – 23-27 March

Dave (Lead Developer) is rushing quick wins out to appease Alan (Head Architect) and Liam (Team Manager).

Tuesday was a write off. I kept on running into different problems; couldn’t log in, or I could but the app wouldn’t load, or horrendously slow to load customers etc.

We are starting to integrate with the real VerbatimEcho app now. Some bugs we are finding now because they are blatantly obvious should have been spotted earlier.

If we add new Headings in our app, they error when sending to our API due to missing fields.

We aren’t populating the “Main Product ID” of the section, and not populating Product IDs for headers that aren’t “Main Product” which contain Product IDs. Or that’s what Andrei claims.

We aren’t setting the Provenance for new items. Andrei reckons we should just remove the validation on the API rather than passing correct data through the app. His reasoning that it is quicker. It’s going to be roughly the same effort in my opinion. If it was a complicated object to set, then it’s only going to be an extra day or so. If we do his suggestion, we have to come back to it in the future, undo the deletion and fix it properly anyway.

When AI fills in a row, there is a button you can click to “Show Source” which should open the Transcript and highlight the appropriate sentence(s). We have the button but it doesn’t do anything. I added the implementation to make this Show Source button work. Lucy is working on removing the button. Classic. This is the classic thing that happens when you do your own thing rather than what’s planned on the board. My item was on the board.

In the Daily Stand Up on Wednesday, Dave (Lead Developer) was attending another meeting so I spoke up. I tell Lucy where the prioritised list of items is. She wanted to work on the bug James logged about not populating values automatically. I said “we cannot do it without having a guarantee that the units are the same”. For example, you could have a size listed in cm or metres which is the same metric but a different scale (as in 1 metre is 100 cm). We could do a simple text match but it might not work in all scenarios. (e.g. match “cm” but the unit is sent as “centimetres”).

An example in another context is “M” could be “metres” or “miles”. There’s thousands of possibilities across all types of products so unless we verify all of them and have an understanding of the metric then we cannot do it. If we get the 3rd party to use a standard ID associated with the measurement then we can guarantee it matches.

The next day she insists on working on it. At first, Dave (Lead Developer) seems fine with it but then realises we had recently said it wasn’t safe. He then explains similar scenarios I had explained yesterday when Dave wasn’t on the call. Despite that:

Lucy : “Shall I carry on working on it? I’m not sure if I can continue“

James : “We can populate the values at least, then we know its unsafe“

What is their obsession with doing this work? We have probably wasted 30 mins at least explaining why it’s not safe to do it.

Week 31 – 30-3rd April

Adam said some of our events aren’t working correctly.

VerbatimEcho say they changed start_finalization event to camel case to match the rest of the events. Dave (Lead Developer) says we’ve been using startingFinalizationProcess which used to work. So he doesn’t know what they are talking about since that is already camelcase and has different wording.

Jean (Product Owner) is seeing it integrated for the first time. The side panel is small and when extra warning banners are shown it squashes it vertically, cutting off the buttons and therefore unusable.

VerbatimEcho recording is currently not working. It intermittently records what you are saying so either misses the entire sentence or has several missing words in each sentence. It’s extremely frustrating to have to say loads of statements to hope you can fill the minimum words to get to the app. We really need to go back to our mock app but it needs some extra config changes or something to work again.

In Story Pointing, there was one item where everyone voted 5 apart from Lucy:

Lucy: “I was thinking 5. I’m not sure why I voted 3”

I don’t understand why she has to lie or chicken out of her vote every single time.

We did tell VerbatimEcho yesterday that their audio recording suddenly barely worked. They didn’t seem too fussed about sorting it. But since that is step 1 in the process, it should have been high-priority.

After chasing it up, VerbatimEcho then said it was only present on the Developer environment which is “vulnerable to instabilities“, and implying that contractually they had no obligation to fix it as a priority. I think “vulnerable to instabilities” just means it gets new features first, and therefore the bugs first; so he is just saying “these things happen”. But surely the entire feature not working 99% of the time is somewhat urgent, at least so we don’t think a company that specialises in speech-to-text software aren’t a bunch of cowboys when the recording feature doesn’t work. You would think they would have automated tests which would catch such a significant problem so it’s impossible to release like that.

Regardless, he was lying, or misinformed. It also affected their Production environment so it’s a shambles of a process if a broken feature can be rolled out like that.

Another problem being spoken about is that there is roughly a 125 consultations per hour limit due to a Third-party API rate limit that VerbatimEcho are using. VerbatimEcho say they are making around 40 calls per consultation on average. Our Software Architects are quite concerned that this is a major flaw that we need a solution for.

Week 32 – 6-10th April

James says he needs help, then was gone 11:30 to 12:30 when he finally replies to me. I asked him a follow-up question then he was gone till 1pm. I was gonna ask Lucy but she was gone for 1 hour. Another question, another 45 minute wait. At 4:15, I asked him another question. I noticed he’d been gone since 3:30.

On Wednesday, UsersAPI made a change which was reported at 10am but might have occurred earlier. It showed BLOCKED when attempting to log in. Lucy and James only noticed at 12:15 which makes me wonder what they have been doing all morning. UsersAPI team reverted it but it didn’t build.

People are spamming the channel with bug reports. We’ve got Jean (Product Owner), Safety Officer and UX designer testing it out. If only they viewed it earlier rather than when we were supposed to be finished the project.

On Thursday, we had official confirmation that the “Autopopulate units” idea (which Lucy keeps insisting she works on) cannot be done safely.

The “End of Development” demo was scheduled today. We now have loads of bugs to fix so we delayed the project once more.

The Config team added a new value without telling anyone which resulted in no apps when logging in. So that was quickly reverted Friday morning but blocked most teams for a few hours.

The Friday Retro was cancelled for the 3rd time in a row. (We didn’t have a retro again in subsequent weeks either).

Managers are very concerned with the slow performance from VerbatimEcho. VerbatimEcho claimed their production environment was fast, but we say otherwise. I saw Mark’s claims of “7 seconds app load” and “3 seconds customer load”.

Week 33 – 13th-17th April

It was Bank Holidays for the Contractors so some days we had fewer team members.

Usually we had a process of:

1. Developer makes a fix, 
2. team members review,
3. a dedicated Tester tests it works, 
4. then demo to the Product Owner for the final sign off.

Currently we have: 

1. Developer makes a fix, 
2. team members review, then that is it. 

So that is why the PO, UX designer etc keep finding bugs or missing requirements.

We were discussing making a change to our process but then Liam says:

“There’s absolutely no reason that the person who fixed it cannot verify that it works”

Liam (Team Manager)

Which is just completely against the usual software development process. Any experienced person in software development knows that.

Liam goes on about how we’d have to track metrics on how long it takes external reviewers to approve. His aversion was also that it would hurt our velocity if things got delayed or reopened. This is also dumb. Software Quality is about “delivering the right thing”, not “delivering the right thing fast”. He is just chasing “velocity” which results in the wrong thing fast. I don’t think we have gone fast though, we are massively delayed.

It sounds like our managers have had strong words with VerbatimEcho about their shoddy performance. It wasn’t just on app load, but the time it took to process the recording.

They came back with a detailed PDF hyping up their technology:

“Following extensive benchmarking across six models and multiple commercial providers we can confirm that the current implementation is performing as close to state of the art as possible.”

VerbatimEcho

Thursday morning, I had the “no headers yet” control ready for review and told the team in the Daily Stand Up. On Friday I made it clear that I still need 2 people to review. By Friday 5pm, no one had reviewed it, and I had my new item completed to change the Enter/delete/backspace behaviour in the editing controls.

Week 34 – 20-24 April

Lucy has the week off. Andrei logged a bug about needing to double-click for the caret to appear in the editing controls. I didn’t notice that problem on my branch, or on Main, so I got his branch that was “Ready for review” and saw it happen every time. I wrote that on his PR.

Wednesday; Pilot release is delayed again; pushed back to the 5th May. Ivan is ill so took a day off. He did a Classic developer blag update “Rebased… begun investigation”; so you did nothing then.

Adam wanted 2 weeks for regression but has been told they have 3 days. Liam (Team Manager) says we need more people in the team, which we definitely don’t. Our current size is the problem, since it means we split the work up too small, and we end up dragging items out with the usual Parkinson’s Law.

Andrei still hasn’t fixed his bug but he’s still saying it needs a review when in the Daily Stand Up.

I noticed a loading spinner on the VerbatimEcho finish button. It was positioned slightly off the control which was a really weird choice.

On Thursday, Andrei says he “tried to sync” with me about the issue I pointed out on his PR Monday morning. I never noticed the message and he had never chased up. So he has been blocked on it for 3 days but only saying that now.

Our contractors began having problems with our VPN which meant they could no longer even connect to GitHub so a couple of days were a write-off for them.

We discussed in the Daily Stand Up that the new VerbatimEcho spinner is due to the “Real Time Fact Extraction”. Since we complained about how long it took to summarise the recording, they had changed it so it starts working as you speak. It’s just a waste of processing if you cancel, or accidentally leave the recording running. If there’s a slight delay with the final words, there is a delay before you are allowed to click finish. The UX designer absolutely despised this. I suppose it doesn’t look good and can seem frustrating. The alternative is just to let them click finish and just have the main loading screen like they were showing previously. Suppose that would probably be the most logical implementation.

Some sites are already trying out VerbatimEcho without the integration with our software. I’m not sure what version is on live because they are constantly changing the text it generates, yet claim to have loads of customers around the world. So what does their currently released software do?

We’ve had some feedback on the VerbatimEcho feature in live compared to their biggest rival. VerbatimEcho:

• Doesn’t pick up technical terms as accurately
• Doesn’t understand other languages
• Often duplicates entries
• Negation statements not understood

Liam (Team Manager) says the negation is a massive issue and declares it UNRELEASABLE

Week 35 – 27-1st May

Our contractors are still having problems with our VPN. Liam (Team Manager) asks to describe the problem in detail so he can escalate it to Group IT.

Andrei says he is following steps to add “New Company”. Dave (Lead Developer) points out his screenshot says “Previous Company”, so he is using the wrong credentials.

Andrei says he cannot add “Previous Company” and shares a screenshot showing he has typed “Previous Company”. No error message or anything. Just showing he has begun to do something dumb; using “Previous Company” rather than “New Company” credentials.

Marek then acts surprised that deleting “Previous Company” allows him access to GitHub.

There did legitimately seem to be a problem because they couldn’t sign-in with “New Company” credentials either, but removing “Previous Company” from the application allowed them to connect to GitHub at least.

I don’t know why the contractors were set up differently. They all claimed they had always been set up with Previous Company credentials in the VPN and hadn’t been told to switch at any point.

There were a few other instances where they asked how to get access to things like Dynatrace. We told them it used “New Company” credentials but they clearly tried to use “Previous Company” then posted a screenshot showing as such.

We had a surprisingly long refinement session. We are supposed to have finished the project; but then there’s all this work I haven’t heard about. It all seemed like the most boring work. Some UI changes or Background jobs to delete data when past the retention policy.

Andrei finally fixes his “focussing on first row when you switch tabs” issue. It’s full of hacks. I tell Andrei that 150 lines of hacks probably isn’t a good idea. It’s messy and going to cause more bugs.

On Monday Alan (Head Architect) starts kicking off that the “associated text” bug isn’t fixed. It has to be fixed by tomorrow! Tuesday morning it was! Were VerbatimEcho just holding onto it all this time? Or was it that easy to fix?

After the Tuesday Daily Stand Up, Dave (Lead Developer) calls a meeting about Andrei’s change. I say that it could most likely be fixed in the “Add-Product” package. Dave agrees that’s most likely the case. We agree to check in the simplest fix for Andrei’s item. A new one will be logged for the edge cases.

Ivan has picked up the bug Andrei logged about double-clicking. You know, the one I told the team was only in Andrei’s branch. They now agree that the reason why Ivan cannot recreate it is – that I am correct.

Apparently, we’ve given VerbatimEcho a deadline of tomorrow but I don’t understand how we expect them to get it all perfect.

• Inconsistent splitting
• Qualifier duplication
• Hallucinations
• Not using preferred product codes
• Duplication of comments
• Some comments not descriptive enough
• Wrong product codes used for certain items

Check out this confusing series of statements from Alan (Head Architect)

Alan says it crashes when adding value codes to the Main Product Heading. You shouldn’t be able to add value codes there, but he specifically says “we must not SEND values”. The logical way of phrasing this would be to say we must not allow the user to ENTER values (therefore we wouldn’t send them). Saying SEND implies it’s fine for the user to enter them.

He wants it fixing, hopefully by tomorrow.

Dave (Lead Developer) says it should be handled natively by the Product control we used but we got told to set the property to MainProduct=false because we didn’t want all the other qualifier fields which aren’t shown when MainProduct=false. As a side-effect, it also means it allows entering values as an unintended behaviour.

Alan asks when we do send the values, does SystemNow ignore or crash? But he just saw it crash so why is he asking?

However Adam reckons he added validation code to ignore values in Main Product Heading; implying the crash is a different cause.

Alan declares everything is fine then: “no change necessary“.

We are all like “wait…what?“

Dave says we shouldn’t just “stop sending” values, or ignoring values. The actual fix is a third option – to stop users entering them.

Jack confirms it does crash. Alan says that it’s not possible to add a value to a Main Product Heading in ProjectFutures (contrary to his initial claim where he said it crashes when you do that).

James shows a screenshot of the Main Product Heading having values in ProjectFutures.

Dave repeats that the cause is that Main Product Heading allows values to be entered due to us setting MainProduct=false. It needs more thought to be able to hide both qualifier fields and values fields.

Alan repeats, (probably in an aggressive tone) “you cannot add values for Main Product Headings, full stop.”

Alan then uploads a video of him adding values to Main Product Heading codes and it crashing on submit.

On Friday, Liam (Team Manager) questions why Andrei’s item has been thrown back in the New column. Dave says it will be fixed as part of the additional bug we logged. I interject because that makes no sense, we created the bug to fix what wouldn’t be fixed as part of the main change. Andrei should have checked in a simple fix, and not abandon it. So I take over and complete the work instead.

Week 36 – 4-8th May

Ages ago, Andrei was talking about “caching enums” which he had a Draft PR for, then just abandoned it. This week Andrei told Marek to pick up the work.

A new bug was found. It turns out we allow any type of Product Code in each heading but SystemNow actually has a restriction in some sections. If we send data that violates that restriction, then it is rejected. I thought we agreed early on that any Product Code can go in all sections; the exception of course is Main Product Heading like we discussed last week.

James is saying the API throws validation errors if you save a blank page or a page contains a blank section. Shall we add validation client side then ask the user to remove blank sections before trying to save, or just strip the data client or API side for the save process?

“Sometimes the user will forget to add data and won’t want the section to be removed. What do you think about that?”.

Andrei

I think you are chatting nonsense. Why would they want a blank section to show after saving? They can always edit it and add new sections if they legitimately forgot to add data.

On a Town Hall, one of the Senior Managers stated to the department:

“Scribe is far superior to anything else on the market”

Senior Manager

Ironically, a day later, VerbatimEcho went wild and became even worse.

Me: “Weight 12 stone four”

output:”Work off stone farm”

The inconsistency that you can do the same consultation on the same day and get different results is worrying. But then it has days where the quality drops significantly.

I also noticed “centimetres” was always “cm” so the Transcript isn’t verbatim like we asked in the requirements. It made me wonder if this was why the end result was often so inaccurate – that it changes what you said, then AI can’t make sense of the modified transcription.

Another theory was that I just wasn’t speaking clearly so I put my phone next to the headset and recorded at the same time. I then generated the transcription with AI on my phone. My phone was at least 95% accurate and the words it did get wrong was a logical possibility when you listened to the recording (my pronunciation was ambiguous). Turns out great speech to text technology does exist after all.

VerbatimEcho reckon their solution is “state of the art” yet it’s usually 60% accurate to me on a good day, and completely unusable on others.

Week 37 – 11th-15th May

Liam (Team Manager) says that Pilot should start 18th May. However, he also said VerbatimEcho are rewriting their implementation due to the poor quality. Weird, I thought it was state of the art last week?

Marek and Andrei are still going on about caching enums. I still don’t quite understand what it is. Their PR described it as follows:

“Pre-cashing enums at startup removes one time initialization costs from the hot path keeping performance test clean and consistent “

and

“Isolated into a non parallel collection because some tests mutate or poison the process wide static caches. Eg warming up assemblies containing duplicate keys”.

What the hell does that mean? It sounds AI, but the spelling mistake (Cashing) makes me think they partially wrote it at least.

Ivan is supposed to fix the focus issue, but I test it and don’t see any difference. The main method that sets the focus is unchanged and there’s several places in that same file that can also set the focus, so he can’t have tested it, or even made much attempt to investigate the problem. 

The “New VerbatimEcho” date is 19th May. I can’t expect much changes if they think they can turn the problems around in a week. They plan to drop more enhancements in “phase 2” which is 22nd May which sounds even more bonkers. Why would it take several months for a terrible solution then they can fix it in a week just because we are complaining more?

I spent ages writing a handover document so someone could investigate the bug I was working on. I had a fix that worked but was a few hacks, plus some good changes just to reduce component re-renders.

Week 38 – 18th-22 May

I took the week off. I’ve been waiting for a slow period in the project, and was fed up with how bad VerbatimEcho is. I hoped they would live up to their promise and have it fixed when I return.

The team switched to a different JFrog server, which ended up being several small PRs in classic Dave (Lead Developer) style.

Another bug was discovered where: if you apply a Product Code suggestion, then uncode it, the original suggestions no longer appear.

Idempotency solution doesn’t work, or so Mark claimed. He pasted loads of claims from Copilot which Dave said was incorrect. Dave countered with his own Copilot. I think Dave was right in the end but there was a lot of back and forth and it was hard to follow. Just ridiculous that we debate using different outputs of Copilot.

On Thursday, a new firewall rule blocked authentication so no work could be done because people could not log in.

VerbatimEcho have pushed out an update that has drastically increased the description detail. Where you have said loads of content, the consultations contain massive amounts of text. It’s way more verbose than we are used to. We are not sure what caused them to change this and why it wasn’t possible before. It was soon rolled back.

There was a discussion about what to do with silences. Do they show a warning so users can consider microphone issues? What about if it’s not a silence but lots of audio interference which makes the audio unclear and unprocessable? The concern people had is showing a warning when the user intentionally pauses can be very annoying or alarming to the user.

What I don’t understand is this: surely these are “solved problems” by VerbatimEcho with all their other customers. Unless they don’t actually have any. This idea seems more and more likely doesn’t it?

Mark pointed out that 6% of recorded consultations get wrongly rejected for not having Product-related content. He also sees the old processing version numbers in the JSON, so it’s either wrong, or we aren’t actually using VerbatimEcho’s new updates.

Week 39 – 25th-29th

Monday was a bank holiday. Dave (Lead Developer) has the whole week off. Lucy and James had the day off Tuesday.

I checked my User Story that I had written the handover document for. No one had looked at it. I was fuming. I spent hours writing that document, and I really didn’t want to look at this bug again.

Andrei said he would pick up Dave’s work that Dave had told the team to finish whilst he was off. It really just needed testing. However, Andrei then said he didn’t want to complete one of Dave’s PRs for a bug fix because I had already got a similar fix in the branch for my bug fix (that I may or may not include in my actual fix), yet Liam (Team Manager) had said Dave’s item was urgent to get in ASAP, so Andrei should just check it in.

Also, I had some updates so rebooted my computer, then WSL wouldn’t work. I rebooted and had more updates, so I rebooted and had more updates. Restarting took ages, then it was the end of the day. I definitely need Andrei to check in that fix if it is needed.

On Wednesday, Andrei asked to talk after the Daily Stand Up. He said he thinks it is important to get the bug fix in as soon as possible, so it shouldn’t be checked in with my changes. I tell him that’s exactly what I told him yesterday. He then asks me to do it, even though the work is assigned to him. Hours later he then says he hasn’t got the “Add-Product” repo building due to an PNPM error. He was supposed to have been working on this Monday, what has he been doing?

I ended up completing it by Wednesday lunchtime, then when Andrei was ready to check in the PR on our repo, I told him we needed to update the version number of the “Add-Product” package to actually get the bug fix that I checked in. He ended up typing loads of words about it being irrelevant or something. I noticed that Ivan also had a fix to check in for “Add-Product”, so I thought it would just get updated when Ivan completes his work. I don’t get why Andrei volunteered to finish Dave’s work then has spent more time typing meaningless words to not do the work.

Marek is working with Andrei to remove those CreatedBy and UpdatedBy fields. They keep talking about “Orphaned Tables” and “Self Referential Foreign Keys”. I didn’t know what they were talking about at first, but I think there’s a UserTable which has these CreatedBy/UpdatedBy columns and it references a UserId. So I think when the User uses the feature for the first time, then they get a User created plus these audit columns then refer to itself. Then if you deleted the row, it sounds like it throws a foreign key violation. Not sure if that is true though; I’ve never heard of a row referring to itself. You could only create that by inserting a row but with NULL in the CreatedBy column, then instantly update it to then populate it with its own ID. 

Regardless, this is extra work due to Andrei and Dave (Lead Developer) insisting we have these columns because “they might be useful“, even though the Architect did the DB design and designed it to be useful to his Monitoring tool, and he didn’t have a need for these columns.

VerbatimEcho still seems inconsistent. Sometimes missing words, sometimes abbreviating words, but nearly always misunderstanding at least one word in each sentence. What happened to this new version we were promised?

Week 40 – 1st – 5th June

One of our managers had complained to VerbatimEcho that the transcript was randomly missing out words, and completely mishearing words, which is what I’d been saying for a while but no one else seemed to have noticed it, unless they have just been chilling, or ignoring it. 

The manager said he could record the audio, then reuse the same audio and expect the same result. However, he gets different transcripts and gets different consultations when processed; even in the cases where the transcript was very similar.

VerbatimEcho’s response was:

“Just to clarify, were these the exact same audios? If the audio was recreated slightly differently, then this would make sense which is the only assessment we’re able to come to”

I’ve got an idea, why don’t you test it out yourself? Record on your phone then compare the transcript. It took me a few minutes to complete my recreation. I checked my phone recorded and transcribed it perfectly, and then compared to VerbatimEcho which had various inaccuracies on every line. Then I used the same recording for 2 more consultations and got different results. 

Now, you could say each time I made the recording by playing my phone’s recording – that the recording wasn’t EXACTLY the same because of various background noises like cars driving past the house. However, you would think the audio processing would strip out background noise and would be analysing the words and tone which is EXACTLY the same because it’s a recording. They can probably just send in a pre-recorded file anyway to get an EXACT test.

It winds me up so much that they instantly deny our reported bugs, claim it doesn’t happen, or can’t happen, then later have to admit it does happen when they finally check it. They clearly haven’t bothered investigating or completing an “assessment”. They just lie all the time.

We’ve been discussing rolling back some of the UI changes. The removal of the delete button on each line has annoyed Alan (Head Architect). Then we were talking about rolling back the change we made to group together the text areas into one larger one per section. I said we shouldn’t make changes without approval of the UX designer, otherwise what is the point of having him assigned to the project? He had spent hours designing this. Anyway, the UX designer has been temporarily reassigned to one of those stupid ideas we have occasionally; where it’s a project to discuss a new process and becomes top priority and everyone is completely ringfenced. Under no circumstances can we contact him.

I realised the Customer API was broken, but couldn’t see anything posted in the Environment’s Team’s channels. This would affect everyone working on ProjectFutures. I asked my team about it. 30 mins later, Andrei says the integration tests are failing on PRs. I thought it makes sense if the Customer API is down because that is a prerequisite. 30 mins later, Dave (Lead Developer) says he’s re-run the tests and hoping for the best. But James says it’s most likely that an API is down. It’s almost like my entire team didn’t see my message.

The next day Dave says he’s noticed the API is down. Then other teams begin to complain about it too. The Customer API team admits they pushed out an update. 30 mins later it “is fixed” because they rolled it back. Except that it wasn’t fixed because you couldn’t SELECT a customer (only SEARCH for them).

What I’d like to know is – how people do so much work without using the app? Or is it that they are chilling so it’s always me that finds the issue first? Even though I’ve been massively chilling the entire project. I’ll call this the “Chill Paradox”.

Liam (Team Manager) said that they tried to set-up a new environment and although the app loads, it doesn’t transcribe. The animation plays like it is recording but no words show at all. I said I had seen the issue but then it started working later on. He had thought it was a problem with their Staging environment but I said I was using the Development environment. Liam informs VerbatimEcho of the problem but never mentioned I had seen it.

We had some kind of Sign-Off meeting for the app to go to the initial Pilot sites. What happened to the 18th May deadline!?

Just before the meeting, Alan (Head Architect) posts a recording. Everything he said was transcribed wrong. Words and sentences were missed out, numbers he said were transcribed as a different number. He didn’t complain about any of that though. He was angry that he has a weight in Stone and our app is supposed to convert to Kilograms but it showed as Stone so failed to file into SystemNow.

In the meeting, I find it weird how we are all pretending it is good enough to go live. It’s been a disaster. 

“How do we know VerbatimEcho won’t update something and make the solution worse”

Liam (Team Manager)

Me (to myself): “It can’t get worse than this“

I get annoyed how I find problems with VerbatimEcho and I am the first one to rant about them but then others are like “yeah I have seen this before”. WHY IS EVERYONE PRETENDING IT IS OKAY? I also pointed out on the recording screen, the Transcript shows the accurate times. When the processed Consultation loads, the Transcript shows completely different times, presumably because it has trimmed the silences and then showing the new times. But the user does not have access to replay the audio. The only way they could do that is by recording it separately but now the times will not match.

Week 41 – 8th-12th June

Dave (Lead Developer) and Liam (Team Manager) log in at 7am or something daft for the site activation. But it’s actually turned on at 12:30. No users use it that day, or even the day after.

We have another Story Pointing session, mainly about reverting some of the UI changes to the text areas. Some of the stories for reverting are 3 points. Some tweaks to behaviour are 3 points. Similar tweaks to behaviour that could probably be fixed together, all at 3 points. No one really cares about my protests that these are 1 point stories, or a 2 at a push, and we don’t need to split them up so much.

Mark is trying to process files for his Monitoring tool. I still don’t know what it does or how it works but he claims that a 4 min file is stuck on processing status for 45 mins. VerbatimEcho say they have only just activated scaling on that environment. Mark reckons he is talking nonsense because it’s a single request, nothing to do with scale.

Andrei fixes one of the deletion stories but then I realise he has removed the behaviour for deleting the Product Code Badge. I point it out and he sends me links to the User Story. He messages me telling me to resolve the comment “as irrelevant“. I told him I understand the main issue and confirmed he fixed it, but he had also wrongly removed functionality for the Product Code Badge. He said “it’s in the user story” and pastes the entire Acceptance Criteria to me, but states “further improvements can be made in other user stories“. That doesn’t make sense to me because these “improvements” should be done now, or shown to actually be requirements in other User Stories; otherwise it won’t get done.

After reading the Acceptance Criteria again, I then noticed a requirement I had overlooked. It specifically calls out that “deletion of Product Code Badge should remain unchanged”. Andrei backs down and says I’m correct. Funny that he said “it’s in the user story” to do it, but it actually specifically says not to change the behaviour.

Dave (Lead Developer) runs out of GitHub Copilot credits. Andrei says “we should either reduce usage or request a higher limit.” Cheers Captain Obvious what would we ever do without you?

Andrei fixes quite a few small items. A mix of 2 and 3 Story Points but clearly just took a few hours like I thought they would. So my protests during the Story Pointing session were proved correct.

Week 42 – 15th-19th June

The upper management are annoyed at the metrics because they seem inaccurate and there’s not much data. Liam (Team Manager) had an example in Dynatrace where he looked at the last 7 days and there were 4 entries. He manually selects the 10th-15th which is just the previous 5 days; there were 80 entries! I had problems like this when we used Datadog. When it shows data that makes no sense, then I don’t know why we spend an insane amount of money using it.

Dave (Lead Developer) and James were trying to add extra info to get more logs, and of better quality. Seemed quite trial and error, but eventually they got some improvements in.

We seem to have basically run out of work but no one seems to be bothered. Marek and I were asking for work, and the rest of the team had nearly finished with theirs.

Marek was upgrading the DotNet version to 10. There were some code analysis errors that appeared, so he suppressed them. Andrei challenged why he didn’t fix them, and he seemed to make out it was a lot of work. It wasn’t a lot of work, but just needed to make small changes in maybe 20 places which would clutter the PR. So it did make sense to split it out across two PRs, but then he started asking to create a new User Story for it, but it was unnecessary. I challenged him on it, and the team agreed with me.

We eventually had a long refinement meeting but we were mainly saying how the old bugs people logged were contradictory and that’s why we had basically blocked them. Lucy says she has some new bugs to log. Why didn’t she just log them? They were quite simple things that were similar to previous issues we had raised like interacting with controls whilst the Product Coding Panel is open (it then closes and ignores what you clicked on).

They will be straightforward fixes but we probably spent longer talking about them. The main problem was that Jean (Product Owner) didn’t understand them and kept on trying to recreate them as we talked her through it over and over again. Then when the fix was debatable, we asked her to make a decision but she either didn’t understand it or was incapable of making the decision.

We had a couple of User Stories to change how the Product Coding Panel works on small resolutions. It usually opens when you click the Add Product icon, or just click into a row. On small resolutions, we will just open it when you click the button. Another User Story was to  overlay rather than squish the text fields. We will also move the Product Coding Panel to the right so it will cover the end of sentences rather than the start. I didn’t think that would make too much difference overall but that’s what the UX designer wants.

Lucy disables adding codes on small resolutions completely. I told her it was just to disable opening the Product Coding Panel when clicking the text field; but you can still open it by specifically clicking the icon. She then says: “the intent here was not to break small screen coding – it doesn’t seem to be readable”.

Well, you have broken adding codes, and the new design is still unreadable in my opinion because we still have limited space and are now masking the rest with an overlay. So I think we should have questioned the new design, rather than actually just doing it.

Lucy says Ivan will fix what she has intentionally broken in another User Story. Andrei says he agrees with Lucy’s approach. I don’t understand because what Ivan will have to do is just restore the logic she removed, but then add a condition to block specifically when clicking the text field. But that describes the exact requirements from Lucy’s item. So in my opinion there is no advantage to temporarily break the functionality, and it just moves the required work to another team member.

I finish the week with 5 items waiting to be reviewed. I don’t know why people are reluctant to do reviews and it frustrates me when I keep saying each day that they need reviewing and it still doesn’t get done. It’s more frustrating that people were saying they had nothing to do, and still didn’t review my changes.

There was some User Feedback shared from the pilot sites: “it’s a bit rubbish“. They also mention it often “fails to generate the summary so you lose all your work“. In order to keep using it, he “Doesn’t have the confidence so will have to record using another AI product at the same time as a backup”.

Remember my initial prediction at the start of the project?

“How easily can the user recover if they have recorded a 10 minute clip but haven’t filed the data due to an error? I’d imagine once that happens a single time, then the user will stop using the feature”.

Me

Week 43 – 22-26 June

There was a call arranged to discuss the possibility of releasing to 4 more Pilot sites. The meeting was cancelled and I think it was decided to not release.

On Tuesday, we had a meeting to discuss the remaining work and a new item about displaying a warning to the user that certain warnings won’t be displayed when saving when a consultation is created via the Scribe.

Jean (Product Owner) said that pilot users have been reluctant to use our software because they have seen crashes. It’s exactly what I feared. As soon as they see a failed consultation then their trust in it will be zero and they will no longer use it. If you have spoken for 10 minutes then it fails, it is maximum frustration.

As the discussion progressed, there was a suggestion that users did not like the Product Coding Panel which had caused us loads of problems to get working as it does.

Then there was talk about scrapping our data entry completely in favour of the existing Consultations module that has also been drastically improved recently. So that will mean the UI side could be completely replaced.

Then there was talk of another team attempting to recreate our own AI product so we can bin off VerbatimEcho.

So it currently sounds like the product we have worked on is going to be completely replaced.

James claimed he had been secretly investigating making his own Scribe. No one had asked him to do this, and it wasn’t clear what he had done because it would probably involve signing up to various tools/services to make it work.

Week 44 – 29 June-3rd July

We were shown the new Scribe, it was made by Alan (Head Architect) and Mark so I suppose that was no surprise.

They showed a complex consultation and it seemed 100% accurate with transcription but also generated the consultation notes in real time, and was suggesting actual Product Codes for each statement. It seemed very fast as well. So it seemed better in every aspect.

No idea if it was using any hacks or would scale well, but they have proved the idea is possible regardless. It seems like they got this working in 3 weeks too which is incredibly impressive when you compare to VerbatimEcho who are supposed to specialise in this.

Tuesday, Liam (Team Manager) confirmed that the VerbatimEcho contract will be cancelled. At the end of the day, it was suggested that the plan really was to get rid of our UI work too.

I’d love to be a “fly on the wall” at VerbatimEcho. This contract would have been absolutely massive for them, and to lose it could be potentially crippling to them.

Conclusion

Another project cancellation, and it was supposed to be the most exciting project for the company in recent years. 44 Weeks to decide it wasn’t good enough for users, and that’s after around 6 months of delays and internally declaring “”Scribe is far superior to anything else on the market”.

I think the following would be a good summary of the main failures. Some directly contributed to the cancellation, whereas others just contributed to the joke.

1. Microsoft Copilot used to create the User Stories; sometimes invented irrelevant requirements.
2. Splitting up work so it’s too small. Makes the refinement meetings too long, and creates blocking dependencies.
3. Team size slightly too big which meant we needed to spend longer in planning meetings to get enough items for everyone.
4. Planning sessions ended up including some refining and estimating which should have been separate meetings.
5. No conviction when estimating: People instantly backing down on their vote. Choosing minority vote as the final score made voting often feeling pointless.
6. Daily Stand Up delayed till 10AM so contractors are sat around for up to 3 hours.
7. Liam, Team Manager not clamping down on James’ constant lateness.
8. Liam or Dave not holding people (mainly Ivan) to account when they’d claim to fix items but were still obviously broken.
9. Andrei often made up requirements.
10. Lead developer, Dave, not doing work that is on the board.
11. Project Manager wasn’t meeting with VerbatimEcho. Not holding them to account when their delivery was sub-par.
12. Early product demos were a waste of time. Marketing don’t need a demo of a button or a tab control.
13. Team not informed of other teams related to the project. The sudden declaration of Adam’s team, or the “Partner API” AI Scribe.
14. Took 30 Weeks to see the application integrated, then we had a massive influx of discovered bugs.