The issue is his base are motivated and the Conservative party is a disaster.
09.02.2026 19:11 β π 0 π 0 π¬ 0 π 0@pdj.bsky.social
Multi CTO, Climate, Data, Governance, Government, NED, advisor, strategist, ex AWS and GDS Chair of Milton Keynes Community Energy Chair of Milton Keynes Green Party Other: ADHD/Leadership/Christianity M: @pauldjohnston@mastodon.green X: @pauldjohnston
The issue is his base are motivated and the Conservative party is a disaster.
09.02.2026 19:11 β π 0 π 0 π¬ 0 π 0Bonkers isn't it
09.02.2026 19:10 β π 0 π 0 π¬ 0 π 0I have 2 kids who have gone to uni recently.
One who graduated with a BSc, then an MSc, and is now doing a PhD.
The other has completed 3 years at uni but has chosen to not complete the degree yet.
Both have significant debt. It's a "went to uni" debt, not a graduate debt.
Economically if Starmer goes, then Reeves goes, and while I have no love for Reeves approach, I don't see anyone else coming through there either.
This govt is the Starmer/Reeves govt and there are significant risks of the wrong parties getting in if the leadership process descends into chaos.
I'm genuinely not sure I want Keir Starmer to resign.
There is no person waiting in the wings (although names have been mentioend) and I wonder if the country will be put into a much deeper crisis.
Might it be better to retain him as a weakened PM and push him towards better policies?
It's more a recognition of the reality we're living in and the future we're likely moving towards.
Lots of things are about to change.
I'm seeing a lot of divergence of usage with this kind of thing. Lots of people focusing on very specific elements e.g. "add a function that does X with tests" and then others, like Simon who are throwing complex problems, but crucially with examples to build on, at these agents.
It's a skill.
While I do think that LLMs are not going to be able to do anything innovative and creative, there is *so much* that they are aware of that they can be incredibly useful for.
Most people don't know many things (even clever ones) and LLMs have a lot of knowledge that can be used in lots of contexts.
Can someone explain how to do live sport video streaming...
Not to me, to ITVX because it seems to buffer and fail all the time.
Ugh
Nice. Will take a look
07.02.2026 14:28 β π 1 π 0 π¬ 0 π 0Options in my mind are:
* GitHub Pages might be just the easiest way to do it
* Medium again because... some history
* Something something serverless
* LinkedIn
* and... ?
I need to have a blog again, just for me.
I *REALLY* don't want to host my own blog, and I *REALLY* don't want to rely on services that are there to monetise above all else e.g. medium
My old blog is on medium, before that it was elsewhere, and I see a lot of "old style blogging" on linkedin.
I get the point.
But the impact of the built environment is not just the small bit of land it's on.
It's water use, vehicles, noise, some types of habitat loss and so on.
But I do agree that building regulations are not as important as reforming farming and nature.
But planning needs reform to
This is a good take.
I see something slightly different. I see a generation of developers who are going to be using regeneration as the norm, not bug fixing.
It's not now, but it's coming
Oh... I'm writing a talk for senior tech leaders on exactly that
05.02.2026 22:38 β π 1 π 0 π¬ 1 π 0The problem here isn't that it didn't come up with a workable solution.
It is that the solution that it seemed to prefer was not wrong, but was less easy to maintain over time.
Again, not wrong, but not really what I want.
Specifically asking for minimal maintenance doesn't really help.
It got stuck on me saying I wanted to have a set of routes on /api and lambda functions behind them, and then all other routes outside /api are served from s3. It was asking me whether I wanted a lambda function to do the proxy or cloudfront or a number of other options.
05.02.2026 17:09 β π 0 π 0 π¬ 1 π 0Claude code specifically had a number of mini fails at trying to understand how to setup Cognito with Google as the identity provider.
Once it figure out what I wanted though it got it mostly right. The scope was wrong, as it's a web app, but when I pointed it out it was fine.
No. I've got a lot of people around me using Claude Code and other similar tools, and I wanted to do a non-AWS tooling test to see what happens.
I might try it for comparison though.
I did a number of things, including providing a URL with a reference implementation for deploying cognito + api gateway and it still did it's own thing.
I know how to do most of this on my own, so this is very much a "what happens if a non-expert tries something like this?".
Good question.
05.02.2026 16:44 β π 2 π 0 π¬ 0 π 0This Labour Govt has royally screwed up
05.02.2026 16:44 β π 1 π 0 π¬ 0 π 0I think at some point these coding GenAI tools will be able to create really good code that is also readable by humans.
At this point, there's a lot of effort that goes into making the code produced "good" in my view.
Does it work? Often yes.
Is it good? Not often.
It's very frustrating because I would have thought there was enough information (e.g. github, serverlessland.com and others) that would give the LLMs enough context to be able to build minimal solutions that do what is needed.
But maybe I'm expecting too much.
Although I don't think so.
It is frustrating to watch an LLM tool not be able to deploy serverless solutions without a lot of tinkering and direction.
I could have given it a base template and started there, but I wanted to try.
I've not done code like this for a while (a few years) and this is showing me LLM limitations.
One of the massive advantages of AWS SAM and serverless deployments is the minimal configuration for deployment.
A lot of work is done for you, and while sometimes what you want is detail and explicit, 90% of the time, the shorthand works for what you need.
The problem here is that as a serverless expert, I know at least to a high level, how to deploy things properly.
It's not that the code did not deploy the things that I asked for.
It's that it ignored some of my requirements (and I had defined them clearly) and created overly complex deployments.
And early on I said I wanted to use openapi definitions for the api gateway.
That's not in there at all.
I can live with that... for now.
It also took a while for it to understand how to do the cognito aspect of the deployment.
Admittedly it's a bit more complicated than just using cognito directly, but I was hoping it had enough data to be able to recognise it.
That took some work too.
It took several attempts to prompt it to use serverless/SAM base for deployment.
The API Gateway deployment was very complex and overly messy.
It annoys me because I specifically said SAM and specifically serverless approach.
And yes, of course I can do this by hand.