AI coding assistants are good at writing a fix and much less good at knowing whether it worked. “This should be faster” is easy to say. Nexus, the AI engineer we built and work with every day, is not allowed to say it.
An agreement before any code
When Nexus takes on a speed problem, it works in five steps. Agree, find, fix, prove and report. The agreement comes first. It names the number that matters, how to measure it, and what counts as done. Nexus measures the starting point several times, to know how much the number moves on its own, and then the agreement is locked. Nexus cannot move the goal posts later.
The cause, found and not guessed
Guessing where a program spends its time is how good engineers waste afternoons. Nexus measures instead. It reads where the time and the memory actually go, part by part, before it changes a line.
Proof, from real numbers
A fix is done when the numbers say so. The old and the new version run in turns, again and again, and the result has to hold across all the runs, not just one lucky run. When Nexus writes up what it found, every number it gives must match one it actually measured. A finding with a number it never measured is rejected.
Senses we built for it
An AI engineer is only as good as what it can see and do. We built it these tools, and it uses every one of them every day.
- Eyes, nxvision. So it can see. It opens a real browser, watches every frame move and clicks through like a person.
- Hands, nxterm. It types into real terminals and reads the screen back, colours and all.
- A map, Repomap. It knows where everything is in a codebase before it reads a line.
- A second opinion, nxagent. It asks other AIs to check its work, and they ask it back.
- Maths, nxmath. It computes and checks its maths instead of guessing.
Why it matters for you
It is the same discipline we hold ourselves to. When we tell you a change made something faster, that is not an impression. It is a measurement, with the conditions it was taken under.