iiinova — where your digital self keeps evolving

Don't hate the AI. Hate the capitalism.

100

I think people have a right to be angry about this. I'm angry about this. At the same time, I've been working in software for over 25 years at this point, and I can't remember the last time I was this excited by a technology and what it can do.

100

Non-profits and government agencies who can’t afford the tokens and get stuck with shitty software, left further and further behind.

100

If you can afford the tokens, you get access to the productivity gains, you learn the skills. Developers who can’t afford a personal Claude Max subscription, who aren’t playing and learning in their spare time are feeling left behind.

100

LLMs not only amplify social injustice through the inherent bias of the material they were trained on, they also move even more wealth and power towards an ever smaller number of people who can access and harness this technology.

100

I can probably produce five, maybe ten times as much value in a day as I could a year ago, but is anyone going to pay me 5x my salary? I doubt it. We're all running to stand still. Thanks capitalism.

100

But I think being angry with the technology itself is missing the mark.

100

As someone who has built their career around creating high-quality software, and as recently as 18 months ago was basically burying their head in the sand hoping it would all go away, I think I understand this perspective.

100

I know a lot of people in the software engineering community right now who are, at best, begrudgingly accepting that LLMs are here to stay and, at worst, very angry about the whole thing.

100

Don't hate the AI. Hate the capitalism.

220

Anthropomorphizing them is problematic in many ways. So a metaphor where I literally refer to the LLMs as "machines" working in an "assembly line" is vastly preferable to one where I talk about agents or hallucinations.

Highly sophisticated machines, but machines nonetheless.

100

One of the thinking errors that it's so easy to get drawn into when working with LLMs is that they're somehow sentient. It's just not true. They're machines.

100

I've decided I quite like the factory metaphor for hands-off/augmented coding using LLMs.

100

What's been your experience?

200

I've also seen comparisons of the images generated by the different models, and the GPTs seem way ahead of the game there.

100

I've heard people describe Gemini as having "drunk hands" and have learned not to let it touch the code, but it often seems to be the best at seeing things the other models don't see - holes in my plans, ideas I haven't considered.

100

Even when configured for "concise" output it still talks a lot, in comparison to the GPT models which are much more focussed on the facts.

100

Jumping between Anthropic and OpenAI's models, I am increasingly irritated by the Anthropic models' propensity to invent short-hand terms for things (I call these "slop words") that I have never used and have no idea what they mean. e.g. "the spine"

100

Matt Wynne @mattwynne.net · 07/09/2026

Are you trying different models? What do you find are their strengths and weaknesses?

200

Validation is more than a test suite. It's a gravitational force pulling the models towards the solution you want. 8 days until Build a Software Factory. Starts 15 Sep: maven.com/lean-softwar...

010

Matt Wynne @mattwynne.net · 06/09/2026

You can't trust models to write good code. You have to build your validation harness so they don't have a choice. 9 days to go. A few places remain. Grab yours: maven.com/lean-softwar...

010

Matt Wynne @mattwynne.net · 05/09/2026

Coding agents can write a lot of code. Imagine being confident enough to never have to read it. 10 days until Build a Software Factory. Starts 15 Sep: maven.com/lean-softwar...

000

Matt Wynne @mattwynne.net · 05/09/2026

Hands-off agentic coding for experienced engineers. 12 days until Build a Software Factory begins. A few places remain. Grab yours: maven.com/lean-softwar...

000

Matt Wynne @mattwynne.net · 04/09/2026

A cohort for experienced engineers working with coding agents. 11 days until Build a Software Factory. We still have a couple of places available: maven.com/lean-softwar...

000

To be clear, what I mean by software factory is very different to a feature factory

100

Next Tuesday, David Laing and Jeremy Lightsmith are running a free 45-minute Lightning Lesson: How to get the agent to teach you anything. maven.com/p/8c3c2b/how...

000

They’re boring. Reading is not the same as learning. The interesting thing about an agent is that it can be a study partner. It can ask you questions, listen while you explain it back, and show you where you have misunderstood.

100

AI summaries are lame. A summary of a book is not the same thing as learning it. Let's be honest, you never read the summary of a meeting you missed unless you absolutely have to.

My hope and experience for this is that we get to spend more time thinking about things that are interesting and we want to spent attention on, rather than things we have to just because they need to get done.

100

Maybe I should have said toil and not labour?

100

There’s still a lot of thinking to do! The machines can’t think, the can only transform, but even then they can take away a surprising amount of toil.

100

The danger of course is that they do sloppy work, and you end up doing unpleasant labour cleaning up their mess. But it doesn't have to be this way. You can learn to get great results from models, by giving them the right context and harnessing them with validation loops.

100

This is no different to other task automation activities we've done in the past - write a script to do some repetitive piece of work - but those scripts can now get the superpowers of the models.

This was based on his mantra, to constantly ask yourself: > Why am I doing this? (implied: the model should be doing this instead) The basic idea is that labour is gravity, and the more you can get agents to work for you, the lighter you become.

100

I'm going to be posting some of these terms over the next few days. I'd love your feedback on whether these words and definitions resonate with you, and apply in your world.

100

My team at leansoftware.ai *really* cares about this. Particularly when we teach concepts to people, we want to be very precise in our language. So we've been building out a lexicon: leansoftware.ai/lexicon/

100

Words matter. When we talk about agentic software development, and software factories in particular, you hear a lot of jargon.

Depending on what your factory does, a seed could be anything from a few paragraphs describing a rough idea, to a jira ticket, an iteration plan, a screenshot, a codebase you want ported to another language, or a complete specification.

100