What we hand over to it
Getting to know an existing piece of software, writing the repetitive parts, reviewing a change, reading and sorting documents, producing a first draft. Everything that can be checked.
ARTIFICIAL INTELLIGENCE
AI, we master it.
And we know how far to take it.
We build with AI every day: agents, custom tools, prototypes taken through to product. We also know where it helps, where it exposes you, and what we do not hand over to it.
Mastery is knowing how much to use
Every use of AI has a price and an exposure. We know them, we say them, and we set the dial with you.
What we keep
Deciding, designing, validating, going live. Everything that commits you, and everything you choose to keep.
Our AI know-how
Agents working as a team, tools plugged into yours, prototypes taken over, documents read, content readable by agents.
What it can do, and what it cannot
It all comes down to one word. Intelligence means two things: the intellect, and the gathering of knowledge, as in military intelligence. What got built is the second one. Nobody invented artificial intellect, but artificial knowledge. Not the same animal.
Knowledge on tap
It has read every manual, every piece of public code, every answer already given. When a tool escapes us, it knows it. A formidable accelerator, as long as we know what we are looking for.
Volume, not reuse
A machine trained to produce is rewarded on what it outputs. Reusing what exists does not show in what it produces; duplicating does. Volume looks like work.
No experience
It has never picked up its own work six months later. It does not know what a lost fix costs, a rare case forgotten, a copy that breaks nothing and ages in silence.
We keep a hand on the plug
We use it to explore, write and check. Never to decide.
It gets to know an unfamiliar piece of software faster than we do, writes the repetitive parts we have already written so many times, rereads a change looking for a forgotten case. At that, it is formidable, and we would be fools to do without it.
But every line it produces goes through our hands before it enters a project. A building block copied instead of reused is a dead copy the day it is pasted: it loses every fix its community will bring tomorrow, and nobody notices, since nothing breaks.
Same reflex when we hear “no more need for a CMS, AI does that very well”. A CMS is twenty years of fixes, contributors and edge cases already solved by others. Replacing it with code produced on demand means trading an asset for a deliverable, and taking the contributor’s autonomy away: they no longer publish, they ask.
What we refuse
A building block copied rather than reused. A rule laid down because “it is good practice”, without knowing what it protects. A component rewritten on demand where a proven one exists. A change approved because “it passes”, without having been read.
Three times, only one that counts
Production time has collapsed. Understanding time has not moved. Review and validation time has gone up.
Production: collapsed
Writing the repetitive parts, laying out a form, wiring a third-party service: what used to take time hardly takes any. It is the time everyone sees, and the only one AI brings down.
Understanding: unchanged
Knowing what needs doing, why the original logic was written, which cases it protects. It is the time we have always paid for, and it has not moved an inch.
Review and validation: up
The more the machine produces, the more there is to read. A shortcut that passes every test breaks precisely the case the rule was written for, and nobody sees it before the customer does.
Since knowledge went self-service, everyone believes they got the rest with it. The beginner takes themselves for a veteran, the tool for a peer, and the buyer builds prices on production time. Having access to an artificial intelligence makes nobody a doctor or a carpenter: it gives access to knowledge, not to its mastery.
Fixing is fast. Knowing what to fix is experience and understanding. That cannot be prompted.
Against median-crity
Statistical tools push every production towards the middle of the curve. We work against the current.
Median-crity (n.)
An aesthetic and intellectual regime in which statistical tools (AI, algorithms, recommendations) push every production towards the middle of the curve. Neither sublime nor scandalous. Just acceptable, just forgettable.
Handing someone the best brush in the world does not make them a painter. Handing them the best code generator does not make them a developer. What makes a piece of work memorable is the signature of those who made it: their choices, their refusals, the way they name things.
We hold on to that signature. AI sits in our toolbox, on the same shelf as the rest of the bench. Never in the author’s seat. The hard part is swimming against the current so as not to slip into the easy way, and using it as well as possible rather than as fast as possible.
Let's talk about your project
With AI in the toolbox, and someone who answers for the result.