Week 888 & 889

Weeks beginning Monday, 19th January 2026. We’ve spent the last couple of weeks working for the Government Digital Service via our friends at dxw. James and I had a couple of days wiped out by lurgy this week, but we’re all back fighting fit as I type.

Government Digital Service

The broad aim of our current work is to try and understand the way that content published on GOV.UK can be “tagged” and whether there are improvements that can be made to the publishing tools to improve this tagging.

Tagging is a catch-all term for attaching metadata to documents that live in the content store. In particular over the last two weeks we’ve mostly been focussing on a particular kind of tag - a taxon taken from the Topic Taxonomy.

As a team we’ve been given a steer that topic taxonomy tagging is sometimes missing, or incorrectly/poorly applied. The hypothesis is that this makes it harder for both human and machine users to find the content they need.

We’ve been analysing the documents in the content store (there’s almost 1 million of them at this point) and looking at the source code of the publishing apps to understand and summarise the extent of the problem.

At a high level we’ve identified some publishing apps that do not make it easy to add topic taxonomy tags, and a few inconsistencies between how validation is applied. We’ve also, and helped by the parallel user research work, spotted some non-technical/workflow reasons why certain document types are not being tagged.

I think we’ve done a pretty thorough job of this part of the work and I’m looking forward to turning what we’ve done into a plan for how we can make things better. I’m also please that although we tend to brought into projects as “developers” we’re happy and able to turn out hand to this kind of business analysis/data/research-y type of work.

Other things

We’ve not had a ton of time to look at much else over the last two weeks, but here’s a few things:

• Chris has been experimenting with opencode and one of its free models to fully vibe-code a terminal based TODO app. He claims to have not read the code, nor to really care what language it’s written in. I think it’s fair to say we’re all pretty sceptical of gen-AI and unimpressed by the collateral damage its having on our work, but it pays to base that scepticism on experience.
• I’ve been discussing gen-AI album artwork on jam.coop with our friends at Mirlo and Ampwall. I think we’d all like to enforce a blanket ban, but it’s time consuming and difficult to do that. I’ve had a good discussion about collaborating on a blog post to show how easy it is to make some beautiful artwork by hand even if you’re not a visual artist.
• We were pretty frustrated to read this account of how £4.1m was spent on PwC to build the AI Skills Hub website. It’s seemed obvious to us for a while now that small UK-based web agencies (let alone co-operatively owned ones that you’d hope a Labour government would be keen to support) find it almost impossible to win Government procurement opportunities through things like Digital Outcomes and Specialists. But we know for a fact that there’s plenty out there that could (and have in the past) done much better work for a lot less money. This is a case we need to start making (again) through organisations like CoTech.

Until next time.

– Chris