Rob Bowley
@robbowley.net
2.2K followers 1.1K following 5K posts
Product & Tech Leadership Advisor, Consultant, Coach & Mentor Tech, Software Development, Science, History, Economics, Politics https://blog.robbowley.net https://www.linkedin.com/in/robertbowley https://pragmaticpartners.co.uk Manchester, UK
Posts Media Videos Starter Packs
Pinned
robbowley.net
Lots of new followers so thought I'd do an intro 🙂

This is me, without hat, in my garden, on a rainy morning in Manchester 🐝, UK

Been working in tech for 25 years - software engineer then various leadership roles. Nowadays I'm a product & tech leadership advisor, coach, consultant.

1/4
Selfi of Rob Bowley on a rainy morning in Manchester with his garden in the background
robbowley.net
I have seen just as awful code written by humans
robbowley.net
The thing about having e.g.a rules.md file and when it decides to actually follow it...
robbowley.net
Not weird, me too! When I was more hands on, refactoring messy code was always the bit I enjoyed the most
robbowley.net
What's galling is hearing AI leaders/boosters brush off overbuild or bubble risks as if they don’t matter

The BoE warns a burst could dry up finance for households and businesses.

When people with power say ‘don’t worry’, what they really mean is ‘we’ll be fine’
Bank of England warns of growing risk that AI bubble could burst
Possibility of ‘sharp market correction has increased’, says Bank’s financial policy committee
www.theguardian.com
robbowley.net
Refactoring, I love it
robbowley.net
I quite enjoy cleaning up crappy code
robbowley.net
Some reflections on @duncanjbrown.com's recent article "Team dynamics after AI", which is one of the best/most important pieces of writing I've read on AI in a long time
Reposted by Rob Bowley
carlquintanilla.bsky.social
NVIDIA and OpenAi:

Concerns that their “increasingly complex and interconnected web of business transactions is artificially propping up the trillion-dollar AI boom.“

@bloomberg.com $NVDA 👀
www.bloomberg.com/news/feature...
robbowley.net
From a CTO WhatsApp group I'm in
Message in CTO WhatsApp group:

I did a code and architecture review for a company yesterday who had vibe coded their MVP - there were no common classes, modules, functions, etc. every "feature" was a separate file with literally a main method full of all the code required for that one function - no reuse anywhere and loads of bits of code that were declared but never used, variable names that didn't make sense, the same variables defined everywhere, etc. Because they were quite far down this path the tidy-up process is going to be a long and difficult road
robbowley.net
This is now the favourite thing I've read on AI.

Why the idealistic AI maximalist dreams just do not stack up.

Can't recommend it enough. Long read, so grab yourself 15 mins without distraction.

Sure I'll be coming back to it many times
robbowley.net
Had me with Ian Mckaye and skateboarding, stayed for to A++ article
robbowley.net
When my brother bought his house in the U.S, it took *2 weeks*
robbowley.net
Long overdue reform.

Sellers should be responsible for surveys etc through and offers should be binding commitments.

Confounded why not been addressed sooner. Is it because means less incentive to sell?

www.bbc.co.uk/news/article...
Buyers could save hundreds in new house buying shakeup
Under new proposals, sellers and estate agents will have to provide key information about a property up front.
www.bbc.co.uk
robbowley.net
I have some failures #cachestampede
robbowley.net
I have not gone further than very very rough calcs so may be well put either way
robbowley.net
Very roughly: a 500k LOC system is on the order of 5m–10m tokens.

Current LLMs generate 30–100 tokens/sec. That’s ~14–90 hours of straight generation, before retries, validation, or tests. At today’s API pricing ($0.002–0.01 per 1k tokens), that’s easily in the thousands per full build.
robbowley.net
Sure, thats not the same thing as AI becoming an abstraction layer though (that means you dont have to deal and manage with the inherent complexity and understand code)
Reposted by Rob Bowley
robbowley.net
The idea of AI as a new abstraction for software keeps coming back – prompts as the source of truth, code as an artefact.

Why it won't work any time soon, even if things like non-determinsim and hallucination become solved problems.
Why AI won’t work as a software development abstraction | Rob Bowley
blog.robbowley.net
robbowley.net
What happens when new requirements force those “tiny agents” to change roles or cross boundaries?

Just back with the same problem
robbowley.net
Good article. I wanted to explore the what if of non-determinism being solved and the interesting thing was even then, it doesn't work, at least not without a ridiculous amount more compute than we currently have or will have anytime soon.
robbowley.net
The idea of AI as a new abstraction for software keeps coming back – prompts as the source of truth, code as an artefact.

Why it won't work any time soon, even if things like non-determinsim and hallucination become solved problems.
Why AI won’t work as a software development abstraction | Rob Bowley
blog.robbowley.net