Jason Gorman
@jasongorman.bsky.social
Trains and mentors software developers in... well... software development, come to think of it. If you're serious about your investment in your dev teams, visit https://www.codemanship.co.uk
Excited to be running the 2-day Refactoring deep dive for a new UK client next week. Easily one of the most valuable and undervalued skills in software development.
Also, one of the most misunderstood.
codemanship.co.uk/refactoring....
Also, one of the most misunderstood.
codemanship.co.uk/refactoring....
December 1, 2025 at 12:49 PM
Excited to be running the 2-day Refactoring deep dive for a new UK client next week. Easily one of the most valuable and undervalued skills in software development.
Also, one of the most misunderstood.
codemanship.co.uk/refactoring....
Also, one of the most misunderstood.
codemanship.co.uk/refactoring....
The interesting thing about videos of people using "AI" coding assistants for real is just how "meh" they are compared to the claims.
There's lots of "Ah, it wasn't supposed to do that" that folks seem to forget when recounting their "agentic" exploits.
And there's a LOT of not noticing problems.
There's lots of "Ah, it wasn't supposed to do that" that folks seem to forget when recounting their "agentic" exploits.
And there's a LOT of not noticing problems.
November 30, 2025 at 8:03 AM
The interesting thing about videos of people using "AI" coding assistants for real is just how "meh" they are compared to the claims.
There's lots of "Ah, it wasn't supposed to do that" that folks seem to forget when recounting their "agentic" exploits.
And there's a LOT of not noticing problems.
There's lots of "Ah, it wasn't supposed to do that" that folks seem to forget when recounting their "agentic" exploits.
And there's a LOT of not noticing problems.
We get the best results from "AI" coding assistants when we:
* Take small steps
* Test, review and refactor continuously
* Use version control to keep our code shippable
* Integrate continuously
* Cleanly separate concerns
If only there was somewhere we could learn these skills!
* Take small steps
* Test, review and refactor continuously
* Use version control to keep our code shippable
* Integrate continuously
* Cleanly separate concerns
If only there was somewhere we could learn these skills!
November 26, 2025 at 8:18 AM
We get the best results from "AI" coding assistants when we:
* Take small steps
* Test, review and refactor continuously
* Use version control to keep our code shippable
* Integrate continuously
* Cleanly separate concerns
If only there was somewhere we could learn these skills!
* Take small steps
* Test, review and refactor continuously
* Use version control to keep our code shippable
* Integrate continuously
* Cleanly separate concerns
If only there was somewhere we could learn these skills!
But that policy will help kids who don't need an unfair advantage
November 19, 2025 at 8:24 PM
But that policy will help kids who don't need an unfair advantage
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?
codemanship.wordpress.com/2025/11/12/t...
Simple, deterministic games can reveal their true nature.
Can you see the tiger?
codemanship.wordpress.com/2025/11/12/t...
November 12, 2025 at 9:29 AM
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?
codemanship.wordpress.com/2025/11/12/t...
Simple, deterministic games can reveal their true nature.
Can you see the tiger?
codemanship.wordpress.com/2025/11/12/t...
Current status: an evening with @nicolabryant.bsky.social
November 11, 2025 at 7:38 PM
Current status: an evening with @nicolabryant.bsky.social
A fun thing to do is play GPT-5 at rock, paper, scissors and let it go first in every round. Then ask it to figure out why it keeps losing.
But a *really* revealing experiment is to go first every round yourself.
"Reasoning" my arse!
But a *really* revealing experiment is to go first every round yourself.
"Reasoning" my arse!
November 11, 2025 at 9:29 AM
A fun thing to do is play GPT-5 at rock, paper, scissors and let it go first in every round. Then ask it to figure out why it keeps losing.
But a *really* revealing experiment is to go first every round yourself.
"Reasoning" my arse!
But a *really* revealing experiment is to go first every round yourself.
"Reasoning" my arse!
Okay then. That's my capacity for delayed gratification exhausted.
November 10, 2025 at 7:10 PM
Okay then. That's my capacity for delayed gratification exhausted.
The data's in: teams who haven't addressed the bottlenecks in their dev process will see bigger delays and more production issues using "AI" coding assistants.
There are aways to avoid it.
But they don't come with any Claude Code plan, I'm afraid.
codemanship.co.uk
There are aways to avoid it.
But they don't come with any Claude Code plan, I'm afraid.
codemanship.co.uk
November 10, 2025 at 7:40 AM
The data's in: teams who haven't addressed the bottlenecks in their dev process will see bigger delays and more production issues using "AI" coding assistants.
There are aways to avoid it.
But they don't come with any Claude Code plan, I'm afraid.
codemanship.co.uk
There are aways to avoid it.
But they don't come with any Claude Code plan, I'm afraid.
codemanship.co.uk
"Do you train teams to use AI coding assistants like Claude or Codex?"
I train teams to be *ready* to use AI coding assistants if they choose - by addressing the bottlenecks, blockers and leaks in their dev process before they attach a code-generating firehose that makes the problems worse.
I train teams to be *ready* to use AI coding assistants if they choose - by addressing the bottlenecks, blockers and leaks in their dev process before they attach a code-generating firehose that makes the problems worse.
November 3, 2025 at 1:50 PM
"Do you train teams to use AI coding assistants like Claude or Codex?"
I train teams to be *ready* to use AI coding assistants if they choose - by addressing the bottlenecks, blockers and leaks in their dev process before they attach a code-generating firehose that makes the problems worse.
I train teams to be *ready* to use AI coding assistants if they choose - by addressing the bottlenecks, blockers and leaks in their dev process before they attach a code-generating firehose that makes the problems worse.
I've been thinking about how we collaborate with LLMs. In an ideal partnership, each plays to their strengths. So I made a handy visual guide.
November 3, 2025 at 5:58 AM
I've been thinking about how we collaborate with LLMs. In an ideal partnership, each plays to their strengths. So I made a handy visual guide.
I think some folks - when they hear "T-shaped developer" - think "front-end" and "back-end".
What does it *really* mean?
Find a copy of this book (there are various editions online - go for a newer one) and look at the chapter headings.
*That* is a T-shaped developer.
What does it *really* mean?
Find a copy of this book (there are various editions online - go for a newer one) and look at the chapter headings.
*That* is a T-shaped developer.
November 2, 2025 at 8:12 AM
I think some folks - when they hear "T-shaped developer" - think "front-end" and "back-end".
What does it *really* mean?
Find a copy of this book (there are various editions online - go for a newer one) and look at the chapter headings.
*That* is a T-shaped developer.
What does it *really* mean?
Find a copy of this book (there are various editions online - go for a newer one) and look at the chapter headings.
*That* is a T-shaped developer.
This totally happened.
October 28, 2025 at 11:53 PM
This totally happened.
I've watched Claude Code and Cursor go round & round in circles trying to get something working and just plain failing so many times.
LLMs will frequently hit problems that are outside of their training data distribution - the code equivalent of "a wine glass full to the brim".
LLMs will frequently hit problems that are outside of their training data distribution - the code equivalent of "a wine glass full to the brim".
October 23, 2025 at 2:30 PM
I've watched Claude Code and Cursor go round & round in circles trying to get something working and just plain failing so many times.
LLMs will frequently hit problems that are outside of their training data distribution - the code equivalent of "a wine glass full to the brim".
LLMs will frequently hit problems that are outside of their training data distribution - the code equivalent of "a wine glass full to the brim".
Job adverts for "Developer" have fall by ~80% in recent years. But advertised "Junior Developer" vacancies are near the floor.
Weapons-grade short-termism.
Weapons-grade short-termism.
October 22, 2025 at 2:53 PM
Job adverts for "Developer" have fall by ~80% in recent years. But advertised "Junior Developer" vacancies are near the floor.
Weapons-grade short-termism.
Weapons-grade short-termism.
For the avoidance of doubt, this is how the inventor of "vibe coding" actually writes code.
Can we all stop pretending now?
Can we all stop pretending now?
October 16, 2025 at 7:11 AM
For the avoidance of doubt, this is how the inventor of "vibe coding" actually writes code.
Can we all stop pretending now?
Can we all stop pretending now?
I'm trying some prompt injection in my LinkedIn profile.
October 6, 2025 at 10:57 AM
I'm trying some prompt injection in my LinkedIn profile.
Me sir! Me sir! I know this one. Is it "reality"?
October 4, 2025 at 10:08 PM
Me sir! Me sir! I know this one. Is it "reality"?
Another fun one to try
October 3, 2025 at 10:48 AM
Another fun one to try