Jason Gorman
@jasongorman
Software engineering teacher, mentor and consultant available for clients who care about creating real value for users, businesses, communities and society. www.codemanship.co.uk
A handy visual aid to understanding why your product's usage stats might not be telling you what you think they're telling you.
The odds are stacked against error-free long-horizon agentic coding. But even if we *could* create models reliable enough and quality gates fast enough and strong enough to achieve it, there's a strong argument for why we *shouldn't* - REALITY GATES Reality, annoyingly, runs at its own pace.
Do the maths - error-free long-horizon agentic programming requires quality gates so good that I seriously doubt the capability of anybody to even get close. Fols who claim to have cracked it do what I call "silent failing". Errors propagate + compound, and agents just keep building on top of them.
We know from the largest datasets that AI coding slows the majority of dev teams down and makes their software less reliable. And we can explain why. We can even explain why individual developers appear more productive at the same time, and why many teams don't even realise it's happening.
AI coding tools lay many traps for us, but the Illusion of Fluency is easily one of the most dangerous
Prepping for the last (ever?) Code Craft training workshop that starts tomorrow. Going out with a bang, I hope. Aiming to make it the best so far. Some nice touches, e.g. giving teams a chance to experience a customer bottleneck in CD - followed directly by an exercise in AI coding.
It's a shame I didn't get to run the remaining Essential Code Craft workshops for self-funders. Much left to cover, including modular design, putting it all together in Continuous Delivery, and code craft & AI. But there's one last chance to do all of them codemanship.co.uk/codecraft.html
Amazon openly flouting UK law. "Yay! You're getting free delivery", "Delivery fee: £1.99" on checkout
The median salary of a software engineer at Tesla is ~£220K. Employment costs roughly 2x the salary in the US, so $200/week is about 2.5% of the total cost of an engineer. AI, the bean counters just put a value on you.
Tonight's the last out-of-hours Specification By Example workshop. So much more than a Gherkin tutorial! Driving designs from end goals, MVPs, and rapidly iterating designs in thin vertical slices. And there's *one more chance* to join us for the extended version of *all* the workshops next week.
It's not about how fast we can generate guesses. It's about how much each guess collapses uncertainty. Before the 1st guess, both teams have a 1:10,000 chance of getting all 4 digits right. After the 1st guess, Team A has a 1:9,999 chance. Team B has a 1:9,000 chance.
Mistake #1 is believing feedback latency is how much time has passed before we get feedback on a decision. It's how many consequent decisions build on it before we get feedback. How far can errors propagate and compound before being spotted? We forget that it doesn't end at getting feedback.
Just 2 spots left on my 3-day Code Craft workshop, July 7-9. Will there be vibe coding? No. Will there be Spec-Driven Development? No. Will there be time-tested techniques we know accelerate delivery cycles, shrink lead times & improve release stability? Could be. See profile for details.
If you can do Java, Kotlin, Python, JavaScript, TypeScript or C#, the (possibly) last (ever) Codemanship refactoring workshop has places available on Saturday morning. www.tickettailor.com/events/codem...
What developers think of LLM-generated code roughly places them within the training data distribution of the model
June 30th at 18:45 GMT+1, Specification By Example. There's no coding in this one, so testers, PMs, UX folks etc are very welcome to join us. Again, until midnight tonight, just £75 + VAT. Bargain! www.tickettailor.com/events/codem...
Spec-Driven Development's all the rage these days, but a lot of teams are walking into a trap we call "Waterfall". We've known of a better way for decades. Join me on June 30th to learn how to be specific in your specs while maximimising feedback and learning. www.tickettailor.com/events/codem...
The skill that made me a much better software architect? Refactoring. Too many "senior" software folks are like the guy who says "I'm a great husband. I must be - I've been married 9 times." Spend a few years living with and fixing poor design choices, and you'll tend to make better ones.
400+ features shipped in a week. That totally happened, and those users are definitely real.
What pisses me off is that if a COO asked me to justify the ROI of teaching teams to work in smaller batches with faster feedback loops, I can easily point to real, measured impact on release cycle times and change lead times. More throws of the dice for the same investment, basically.