Jason Gorman

Live training workshop: Code Craft (3 days) - July 7-9

Strengthen your software engineering foundations to accelerate release cycles, shrink lead times and improve reliability - with and without AI

https://www.codemanship.co.uk

Public Key

npub1x9e8f27nz76l4hndqm69ntk47jzujxj6rv20qnt9khm8jpl98szsyy23ah

NIP-05 Address

jasongorman@mastodon-cloud.mostr.pub

Profile Code

nprofile1qqsrzun540f30d06meksdaze4m2lfpwfrfdpk98sf4jmtaneqljncpgpz3mhxue69uhhyetvv9ujuerpd46hxtnfduqs6amnwvaz7tmwdaejumr0dsdxrs7u

Publishing to

relay.damus.io

nos.lol

Author Public Key

npub1x9e8f27nz76l4hndqm69ntk47jzujxj6rv20qnt9khm8jpl98szsyy23ah

Show more details

Last Notes

2026-04-02 10:37:29 UTC

The longer sequences of autonomous "agentic" software development get, the stronger quality gates have to be.
The reason why fully autonomous execution will never be a thing is because it doesn't take that many steps in a row before the gates have to be effectively *perfect*.
That's a Fool's Errand.

2026-04-01 04:11:20 UTC

CircleCI's analysis of 28 million CI workflows confirms the same picture the DORA data shows. While feature branch activity's up significantly, the median impact on *main* (i.e. release) branch activity's net-negative 7%.
Only the top 5% of teams saw significant gains. The top 10% flatlined at 1%.
For the average team, AI slows them down overall.
Told ya!
https://www.linkedin.com/pulse/what-28-million-workflows-reveal-ai-codings-biggest-risk-circleci-j9syc/

2026-03-31 11:54:08 UTC

I'm pretty satisfied at this point that the true extent of "AI" use in software development has been massively exaggerated.
Sure, lots of devs are using LLMs. But there seems to be very little advanced use. It's mostly chat window stuff and occasional inline completion, to keep the boss happy.
It is, however, a massive distraction.

2026-03-30 10:21:46 UTC

I described to a new friend yesterday evening what I did for a living.
She asked what the value proposition was (she works in marketing).
I gave her the spiel about the impact on lead times, release reliability and long-term cost of change. And then I showed her some stats from clients about those outcomes.
She said I'm sitting on a goldmine.
You'd think so, I replied.

2026-03-24 07:37:19 UTC

Imagine Acme Mega-Oil invested in a taxi service, and the drivers working for that taxi service were measured not on profits, or revenue, or journeys completed or passenger satisfaction, but on gasoline burned. That's basically Big Tech right now.

2026-03-22 07:58:04 UTC

- reply

Part of the obstacle is that if we were to set the bar at "basically competent", 90% of the profession would have to significantly up their game.

2026-03-22 07:36:38 UTC

Ever since I read Steve McConnell's After The Goldrush, I've been listening to people tell me why software engineering should be exempt from the kind of controls that e.g. electrical engineering or medicine or plumbing are subject to.
The reason is almost always "Their work can do real harm".
Yeah? Look around you, dude.

2026-03-20 07:39:15 UTC

New rule: if your for-profit company is "too big to fail", you get nationalised, not bailed out.

2026-03-18 19:27:00 UTC

"All the optimism about using AI for coding, Smiley argues, comes from measuring the wrong things."
https://www.theregister.com/2026/03/17/ai_businesses_faking_it_reckoning_coming_codestrap/

2026-03-10 00:32:25 UTC

Trying to establish if it's true that Tony Hoare passed away on Thursday.

2026-03-08 10:19:09 UTC

I'm expected to believe now that businesses who are using "AI" extensively in their software development processes are keeping it a secret?

2026-03-05 06:32:07 UTC

Another day, another "Well, sure, that's what the data says. But in my experience..." take.

2026-03-04 05:35:25 UTC

"the fastest and most successful adopters are abandoning caring about the structure of the code it produces or even what language it uses"
Looking forward to seeing that data

2026-03-03 14:11:18 UTC

You measured the impact, right? And you did it a bunch of times to establish a trend, yes?
Or is it just "I did this once and it felt like it worked"?

2026-02-25 08:45:25 UTC

It's becoming ever clearer that, far from leading to "super-intelligence", scaling LLMs is more likely to end at "super-mediocrity" - producing outputs that are closer to the average of their training data.
For software developers, this has implications. The average quality of publicly-available code examples is *not* good.
https://codemanship.wordpress.com/2026/02/25/super-mediocrity/

2026-02-24 20:15:00 UTC

A pincer movement will finish "agentic" coding:
* Producing code faster than we can understand it paints us into a corner
* The fastest way to understand code is to write it
LLM-based tools will ultimately finish up as rubber ducks and inline complete
Useful? Sure.
$multi-trillion useful? LOL.

2026-02-24 08:48:02 UTC

Because they have essentially the same causes, the buggier a code base is, the longer it usually takes to change (because changes keep breaking it).
The most popular "AI" coding CLIs seem to be *very* buggy - especially Claude Code. Thousands of open issues.
It's also probably very insecure.

2026-02-22 21:47:21 UTC

Blade Runner is still thought-provoking and exquisitely beautiful and easily my favourite film. Ridley Scott duped Hollywood into making a $30 million art film.
Fun fact: director of photography Jordan Cronenweth was nominated for an Oscar... for Peggy Sue Got Married.
Not for for Blade Runner.

2026-02-21 14:37:15 UTC

"Code doesn't need to be clean as long as it's easy to change"
This is why, when I mean "easy to change", I say "easy to change".
So then folks would have to say "Code doesn't need to be easy to change as long as it's easy to change".

2026-02-19 10:40:04 UTC

The more autonomously you want "AI" coding agents to work, the better your automated workflows, coordination and quality gates will need to be.
As you approach 100%, cost rises to infinity. It's an asymptote.
Walk things back to where the value > the cost
https://codemanship.wordpress.com/2026/02/18/100-autonomous-agentic-coding-is-a-fools-errand/

2026-02-08 01:01:17 UTC

Watching a very disappointing video about AI-assisted coding.

2026-02-05 14:54:39 UTC

It's going to be interesting when hyper-scale LLM usage plans go up to $thousands per month.

2026-02-01 12:09:26 UTC

- reply

@nprofile…acpk There's a wall confronting hyperscale LLMs, and the writing's on it

2026-02-01 11:59:23 UTC

I'm calling it. This age of "hyperscale AI" will end. For so many reasons, it *has* to.
Smaller, more specialised models - perhaps fine-tuned on our data (e.g., our code) - running locally on high-end consumer hardware.
That's where I see things eventually heading.

2025-12-25 07:37:07 UTC

Merry Just Like Any Other Day Except Everything's Closed to all who celebrate!

2025-12-20 20:27:13 UTC

I was ████████ the ████████ files, and then ████████ ████████, but what kind of ████████ would ████████ for that? I mean, they'd have to be total ████████.

2025-12-10 15:28:06 UTC

Consider the possibility that communicating is the work, and coding is the interruption.

2025-12-06 08:50:40 UTC

"Why should I train my developers? They'll just leave."
"Why is it so hard to find developers with the skills I need?"
One question answers the other.
https://codemanship.wordpress.com/2025/12/06/training-mentoring-is-a-common-good/

2025-12-02 07:18:44 UTC

Marathons are easy to start.

2025-12-01 09:29:56 UTC

- reply

@nprofile…sxvw But he was feeding them back in.

2025-11-22 10:26:47 UTC

If I was an employer of software developers, I'd be hiring now while I can have my pick.

2025-11-22 00:13:56 UTC

I've done quite well this week at not mentioning you-know-what.

2025-11-19 10:57:34 UTC

What if... Now, hear me out on this... What if we made it so that the lion's share of a multi-trillion industry all relied on one specific version control host owned by a company not known for their exemplary commitment to reliability?

2025-11-12 21:30:27 UTC

I think we can probably forgive Tony Hoare for nulls. It was his Jaws 4: The Revenge. He was in some much better stuff that more than makes up for it.

2025-11-12 09:45:45 UTC

- reply

@nprofile…sxvw Most scientists see humans as highly improbably products of billions of years of evolutionary processes

2025-11-12 09:30:09 UTC

LLM users tend to fall into two camps: those who see a mind behind the screen that reasons, plans and understands, and those who see LLMs for what they really are.
Simple, deterministic games can reveal their true nature.
Can you see the tiger?
https://codemanship.wordpress.com/2025/11/12/the-tiger-in-the-magic-eye-picture/
https://media.mastodon.cloud/media_attachments/files/115/536/021/562/836/380/original/0afb6fc78c57e654.jpg

2025-11-05 20:07:11 UTC

We're software developers. We can say "No" to the bad tech thing. And then it doesn't happen.
Because the people we'd be saying "No" to would have nowhere else to turn.
Well, I suppose they could try vibe-coding it.
LOL.

2025-11-04 06:51:23 UTC

I'm beginning to wonder if having most of the media owned by a handful of billionaires is such a good idea after all

2025-10-28 10:31:57 UTC

A lot of professional services companies' expertise seems to essentially boil down to "We read the instructions".

2025-10-20 08:49:37 UTC

We apologise for the current eggs outage. It appears the basket is down.

2025-10-13 06:17:50 UTC

Humans innovate, LLMs imitate.
It's really important to remember this if you're building a business with or around the technology.
https://codemanship.wordpress.com/2025/10/13/why-you-probably-wont-get-rich-building-with-llms/

2025-09-10 09:23:23 UTC

Serious question, are version control systems too heavyweight for the frequency of commits some of us are doing today?

2025-09-06 07:43:08 UTC

Stanley Kubrick: "Man, filmmaking is so hard. Today, we did 99 takes for a single shot. And I'm still not entirely happy."
Ed Wood: "99 takes? Wow. I usually just need 1. You must be using the camera wrong."

2025-08-20 11:43:52 UTC

Some folks think I was joking that "agentic" "AI" coding tools like Claude were allowing the BDUF Big Process command-and-control guys relive the glory days.
I'm afraid not. These people are out there.
https://media.mastodon.cloud/media_attachments/files/115/060/911/816/181/871/original/af972a943d570659.png

2025-08-15 05:25:12 UTC

- reply

@nprofile…9zg9 Star Wars just doesn't interest me.

2025-08-14 21:36:15 UTC

Some people seem genuinely offended that Alien: Earth is the only thing on Disney+ I want to watch that I can only watch on Disney+.
Not the Star Wars' TV series? Nope.
Marvel? Oh, please.

2025-07-25 07:10:40 UTC

- reply

@nprofile…9rp6 An engineering manager actually said that at a conference to me once. "We've hired this new developer, and he's so good, nobody can understand his code."

2025-07-25 06:53:43 UTC

"While some devs are writing unit tests and committing 15 times a day, the best devs are just shipping".
The devs who are committing tested code 15 times a day can ship any time they want.
That's kind of the point.

2025-07-04 09:22:10 UTC

- reply

@nprofile…plfp There's just one teeny problem with that analogy: LLMs don't make dev teams more productive. Like AT ALL.

2025-07-04 08:07:30 UTC

Anyone else hearing from devs that they're being "strongly encouraged" to use LLMs by management - who are also raising expectations about "productivity gains" and "cost savings" - and they're not at all happy about it?
Is "A.I." just the latest stick to beat teams with?

2025-05-23 09:58:01 UTC

- reply

Assuming the same business model as Airbnb, Uber, Deliveroo etc, they'll "disrupt" software development until businesses have very few alternative choices.
The takeaway delivery platforms, for example, ran at a loss for years and years, subsidised by investors to effectively eliminate direct deliveries and give restaurants and customers few other choices.
*Then* the prices started going up.
If you're not hiring entry-level because "AI can do that", you're the mark.

2025-05-23 09:52:14 UTC

#1 - The future looks like more and more code generated by LLMs, and fewer and fewer programmers who can fix it. The "hallucination" problem's never going away, and there's no reason to believe they'll ever be capable of fixing all the bugs they generate.
#2 - Meanwhile, price plans for the most capable models are massively subsidised. At some point, prices will *have* to go up, and by at least an order of magnitude.
The race for AI companies is to make #1 a chronic problem before #2 happens.

2025-05-04 06:26:12 UTC

We're 2.5 years into this gold rush, and I still haven't seen any gold. I've seen people selling picks & shovels. I've seen "gold experts" selling maps to the gold. I've seen CEOs announce they're going "gold-first". I've seen people selling land where they claim there's gold. But no actual gold.

2025-04-03 06:06:06 UTC

'When most developers say “I am refactoring my code”, they’re not performing a sequence of atomic rewrites with continuous testing and frequent working commits. They’re just hacking away at the code, wandering further and further from the path into the deep dark forest.'
https://codemanship.wordpress.com/2025/04/03/the-a-z-of-code-craft-r-is-for-refactoring/

2025-03-30 21:34:54 UTC

I can remember when the British press reported that Kate Bush was mentally ill because she rarely agreed to talk to them.
See also Stanley Kubrick, the "notorious recluse" who was regularly seen chatting happily to people in his local Waitrose.

2025-03-27 15:10:35 UTC

Ask yourself this: will fares in self-driving cabs be cheaper? Will subscriptions to vibe-coded SaaS cost less? Will tickets to see model-generated movies be a fraction of the current price?
Or do all these technologies just solve the problem of having to pay people, with no benefit to anyone else except shareholders?
And then ask yourself this: if nobody's getting paid, who's hailing the taxis, using the SaaS and watching the movies?

2025-03-23 21:56:13 UTC

The United States of America is falling to fascism. Eastern European states and Scandinavia are actively preparing for Russian invasion.
But, yeah, tell me again about how autocomplete knocked together a shitty game that worked for a whole 15 minutes.

2025-03-22 21:31:08 UTC

Pro estimating? It's very simple. When they ask you "How long?", you say "Twelve". The units are their problem.

2025-03-20 08:04:10 UTC

- reply

@nprofile…yxwg I've had founders say the quiet part out loud to me: "It only needs to last until we exit".
The irony is that we have a mountain of data going back decades that shows that it would be cheaper and quicker to do a better job.

2025-03-20 06:36:59 UTC

A team of 10 will almost always get more work done than a team of 50 or 100. This is like wearing a hat and taking the handbrake off and then claiming the hat made your car go faster.
https://media.mastodon.cloud/media_attachments/files/114/193/370/735/244/555/original/0a3dcf90b788b510.png

2025-03-12 09:49:03 UTC

When I play guitar, I rarely look at what my picking hand's doing or give the coordination between picking and fretting much thought. After decades of regular practice, it just happens. This frees my mind up to focus on the music.
When I write programs (as opposed to teaching it, which is different), I don't give much though to the mechanics of TDD, refactoring and CI. It all just happens. This frees my mind up to focus on things like the customer and the architecture.

2025-03-09 08:02:30 UTC

I guess we're just going to keeping repeating the lie that AI coding assistants speed up software delivery until it becomes truth.

2025-03-07 07:48:46 UTC

In a similar way that bacon, lettuce and tomato between two slices of bread is "akin" to a sandwich
https://www.theguardian.com/commentisfree/2025/mar/07/i-have-been-an-ai-researcher-for-40-years-what-tech-giants-are-doing-to-book-publishing-is-akin-to-theft

2025-02-26 12:51:17 UTC

Businesses whose dev teams are heavily relying on "AI" coding assistants are still in the FA stage of this hype cycle. The FO stage is going to be... interesting.

2025-02-17 20:50:51 UTC

In a nutshell
https://media.mastodon.cloud/media_attachments/files/114/021/197/841/280/217/original/537973a5bfc68e93.png

2025-02-16 21:49:55 UTC

Just a heads-up: I can usually tell when your post or reply has been generated by an LLM.
Most of us can.

2025-02-15 08:07:14 UTC

In 2005, devs might have worried about things in our careers like "Am I specialising too much?", "Do I really want to move into management?" or "Is this technology going to be in demand in the future?"
In 2025, "Am I working for fascists?" is a real thing.

2025-02-15 07:40:10 UTC

Code generation tools fix the problem of software development being slow and expensive in much the same way that faster cars fix traffic congestion.

2025-02-14 08:51:04 UTC

"Wow. It generated a 10,000 line app in just a few hours!"
You managed to pin down that many customer requirements in just a few hours?
Ah. I see. Silly Jason!
Well, in that case, I can generate about 5,000 LOC a day.

2025-02-12 23:13:53 UTC

I'm betting a shiny penny that in a few years' time, organisations will be paying a premium for developers who *haven't* been relying on LLMs
https://www.404media.co/microsoft-study-finds-ai-makes-human-cognition-atrophied-and-unprepared-3/

2025-02-12 09:26:12 UTC

Turns out the solution to the Fermi Paradox is that technological civilisations use all their resources trying to build a marginally better sales bot.

2025-02-10 21:38:36 UTC

One of the most influential albums in metal. Half this lineup no longer with us, sadly. On drums, the brilliant Sean Reinert (RIP).
https://www.youtube.com/watch?v=EzvtfbqJeIY

2025-02-08 07:58:21 UTC

- reply

@nprofile…9zg9 That would imply that they used to train software engineers 🙂

2025-02-08 07:44:45 UTC

Why is it so hard to find good software engineers?
There's really no mystery. It's because you didn't train any.

2025-01-23 12:28:48 UTC

Once you've figured out that if you don't write the code, it ain't happening, you start to look at the managers who exert "control" over you a little differently.

2024-03-12 04:31:21 UTC

Interesting reading Glassdoor reviews of TPP after this article in The Guardian. I've seen inside a few tech fiefdoms in my time. I'm also reminded of things GPs have told me about SystmOne. Fascinating how you can often see the culture in the code. https://www.theguardian.com/society/2024/mar/11/frank-hester-computer-programmer-who-made-fortune-from-public-sector-contracts

2023-11-22 12:50:38 UTC

Hot take: some programming languages dating back more than 50 years had pretty much all the ingredients you need to write reliable, testable, readable, modular code. We've just been very good at finding reasons not to, and a new language won't fix that.

2023-09-01 13:35:41 UTC

In the early years of my career, most of the books I bought on programming/sw development were about specific technologies. I don't have a single one of those left.
And there's maybe a bit of clue there about the skills and knowledge that last in this job.

2023-08-15 11:03:40 UTC

While your developers are focusing on the 0.1% possibility the software will need to scale to millions of users, see if they can spare a little time to think about the 99.9% probability that the software will need to change many times.

2023-08-06 12:03:20 UTC

It's been over 20 years, and I'm still waiting to hear about this mysterious "other way" of providing clarity to requirements and creating reliable software that's easy to change and can be shipped at any time.

2023-06-15 12:43:43 UTC

The opposite of Y.A.G.N.I. ("You Aren't Gonna Need It") is A.B.W.I. ("Ah, But What If...?"), which is very much a rabbit hole you don't want to go down.
Design for the problem you've got, and leave the door open to change by keeping your solution as simple as possible.

2023-05-13 15:53:20 UTC

- reply

@npub16gl…syzj Have you used it to write software?

2023-05-13 14:57:34 UTC

I've managed to simplify ChatGPT user demographics down to 2 distinct groups: people who think it's amazing, and people who understand the responses.