LAB.00 · Lab Note · Method · 2026
How We Use AI
AI gets things wrong. We use it every day anyway.
Both are true, and most of the argument about AI lives in the space between them. Some people hear “AI” and picture a machine doing work a person should do. Others picture a shortcut everyone should already be taking. If you’re thinking about working with us, you should know where we stand. It’s hard to trust a process you can’t see.
AI is good at text and good at code. It’s quick with a first draft, a rough prototype, a gut check. It has helped every one of us learn a new programming language faster than we could have alone. We use it to speed up how we work. We don’t use it to replace the work.
But it makes mistakes, and it makes them with confidence. We don’t see that as a bug someone will patch next year. People make mistakes just by existing. AI learned from what people wrote, and it carries that error forward, faster and at greater volume. So we assume every draft it hands us is wrong somewhere until we’ve shown it isn’t.
That assumption is the method. A person reads the code before it runs, tests it, and rewrites what’s off. A person checks every source before we lean on it. A person makes every design call. Nothing AI touched reaches you until someone on our team has read it, tested it, and will put their name to it. Mistakes will happen. Catching them is our job, and we don’t let them stay in what we deliver.
We’re not the only ones who see the gap. In Figma’s 2025 AI report, about 2,500 designers and developers were asked about it. 78% said AI makes their work more efficient. 58% said it makes the work better. Faster is easy. Better takes a person.
There’s also a line we don’t cross. Anything personal or sensitive stays out of AI tools. Not every model’s security has been vetted, and your data isn’t ours to test with.
If you’re wary of AI, we understand. Most of that fear is about AI replacing the person. Here it doesn’t, because a person stays responsible for everything. What’s worth fearing is AI used carelessly, with no one checking. And it’s your call. We use AI only where you’re comfortable with it. If you want it out of a step, it’s out.
AI will get something wrong on your project. We’ll be the ones who catch it.
( 01 )
Credit
All design borrows. We study work we admire and take the principles, not the signature. We don’t copy a living designer’s work, and we don’t ask AI to imitate one. When a Lab piece builds on someone’s idea, we name them.
( 02 )
Giving Back
We build on open-source code, published research and the people who did this work before us. So we give some back. Our methods are free here in the Lab. Small, useful tools go out as open source.
( 03 )
Questions
- Does heurology use AI?
- Yes, every day: drafting code, exploring designs, checking work and research. A person reviews and tests all of it before you see it.
- Will AI build my website?
- No. People design and build it. AI helps with drafts and checks. A person decides, reviews and tests every part.
- Is my data used to train AI?
- No. Your data stays out of AI tools unless you agree in writing, and then it only goes into tools that don't train on it.
- Who owns code that AI helped write?
- You do. At handover you get the code, the accounts, the docs and training, same as all our work.
- Do you tell clients when you use AI?
- Yes. This page is our standing disclosure. Ask about any project and we'll tell you.
- Which AI tools do you use?
- Claude Code, most of all. The rest is the usual stack for full-stack and machine learning work, and most of those tools aren't AI. We review our tools often and update this page when they change.