Software engineering teacher, mentor and consultant available for clients who care about creating real value for users, businesses, communities and society. Yeah. I didn’t mention shareholders.
What makes LLMs work isn't deep neural networks or attention mechanisms or vector databases or anything like that.
What makes LLMs work is our tendency to see faces on toast.
The biggest data sets (DORA, Faros, CircleCI) on the impact of AI code gen on software teams show that, for the majority, changes to code are taking longer to get into production, and the software that ships is less reliable.
Worse software, later.
Now there's a marketing strapline for you.
"We’re moving from writing code… to describing it."
Instant water. Just add water.
Another day, another data set telling us what we already knew.
In their AI Engineering Report, Faros detail what they call an “acceleration whiplash” effect caused by AI code gen.
It's the same “downstream chaos” that DORA and CircleCI saw in their data.
https://codemanship.wordpress.com/2026/05/10/slow-the-fk-down/
I feel bad for the folks being laid off by Meta. Imagine having to admit you'd have stayed given the choice.
It's really interesting how many posts about AI-assisted and agentic coding don't mention users or user feedback. Like, at all. It's rarely cited as either a goal or a constraint.
Sigh. Growing talk of "software factories" again. They're not factories, they're laboratories.
The lifecycle of ideas in software development goes something like:
Problem -> Solution -> Book -> Cargo Cult -> Problem
"Design user experiences in minutes, not days"
Wow. The users you're working with must talk really fast.
This morning's TDD training workshop turned into a coaching session, as it was just the 3 of us in the end. Good fun!
The final Test-Driven Development workshop for self-funding learners is on Tuesday June 2nd at 18:45 BST (GMT+1), still spaces left.
Going AI-first without solid engineering foundations is a bad idea.
It should be foundations first, AI last.
Otherwise you'll never find the time to build them, and the engineers who could help won't look at you twice.
After oodles of research & experimentation, I've got a pretty good handle on what kind of input tends to produce the best results in AI coding.
The hard part was coming up with a catchy mnemonic.
Introducing C.R.E.S.S principles for context engineering.
https://codemanship.wordpress.com/2026/05/04/c-r-e-s-s-principles-for-context-engineering/
The difference between an aspiring scientist and an aspiring philosopher? "How are we going to test this?"
While I can still afford the tokens, I'm running a closed-loop experiment this morning to measure the difference in results between 2 approaches to prompting:
1. Explicit negation ("Don't do this")
2. Leading by example (show an example of what you want)
2 is winning by a country mile
This is the training your team *really* needs on AI-assisted development.
Engineering foundations first, *then* AI.
Or let the code-generating firehose overwhelm the bottlenecks in your process and turn production into a perpetual dumpster fire.
Find out more -> https://codemanship.co.uk/codecraft.html
When I say "Continuous Delivery", it puts people in mind of continuous integration, build and deployment automation, pipelines, and all that DevOps stuff.
But at the core of CD is our ability to keep code shippable at any time, and that's very much a "you" thing.
"The problem with test-first is that you end up with lots of tests"
Okay. So how many tests would you end up with if you wrote them *after* the implementation? Fewer?
Hmmm. Interesting.
The problem with test-after is [insert your own punchline here]
It's no idle boast that TDD accelerates release cycles while improving release stability. It really does.
But you do have to actually DO IT
9/10 devs who claim they do TDD have never done it.
Join me May 19 or 23, and become one of the 8% who have
Great news in an email I found in my spam folder. I've been nominated for an Innovation & Excellence award by a totally legitimate awards body.
It's probably for that excellent innovation I did. Yeah, that'll be it.
Still space left on June 2nd evening's Test-Driven Development workshop for self-funding learners.
Accelerate release cycles and shrink lead times without sacrificing quality.
No hype. Just stuff that works.
Just £99 + VAT. Last time at this price
Register -> https://www.tickettailor.com/events/codemanship/2131015
RE: @jasongorman@mastodon.cloud
This is the course your teams need on AI-assisted software development - engineering foundations first, *then* AI.
Join me on July 7-9 for an intensive online 3-day Code Craft training workshop.
Learn foundational technical practices that accelerate feedback loops, shrink lead times and improve release reliability - with and without AI.
* Specification By Example
* Test-Driven Development
* Refactoring
* Design Principles
* Continuous Delivery
* Code Craft & AI
I see so many people obsessing over token efficiency barking up the wrong tree with bervity and compression and wotnot.
They're missing The Big Factor. Inference works *recursively*. Every token in the context is fed back through to predict the next token.
The big savings are in context lifetimes.
Essential Code Craft workshops coming up.
Specification By Example - Sat May 16th 09:45 BST
Test-Driven Development - Sat May 23rd 09:45 BST
SBE - What tests should drive the design of our software?
TDD - How do we drive the design of our software from tests?
"Software engineering has been contaminated for a while, because it deviated too much from the tenets of computer science"
How is that even possible?
Okay, so Tues May 19th's TDD workshop is sold out - but you can join the waitlist in case somebody drops out.
Plenty of places on Sat May 23rd's TDD workshop, though.
Register -> https://www.tickettailor.com/events/codemanship/2131015
7 days to go and just 4 places left on my Specification By Example evening workshop for self-funding learners on May 12.
Non-programming product managers, UX designers, testers etc welcome. It's not a workshop that involves programming.
We've got more available spaces on Saturday May 16th's, too.
Register -> https://www.tickettailor.com/events/codemanship/2162096
CRESS Principles for Context Engineering - S is for Specific
LLMs produce more accurate results when contexts are about one specific problem, and contain the minimum amount of pertinent information needed to solve it.
Move forward by putting one sure foot in front of the other.
Watching another episode of A-list Celebrities Eating Road Kill with Bear Grylls.
