Join us

Pencil AI is a small company building one product: an agent harness with a fast decision model at its centre. We are early, and what we do next depends partly on who joins.

The work sits between machine learning and systems engineering. Some of it is training and calibrating small models and measuring when they can be trusted. Some of it is plain careful software: edit tools that cannot silently corrupt a file, rails that hold when a prompt tries to move them, logs that can be replayed. Both halves need people who check their own work.

We write things down, we say what we have not measured, and we change our minds when the evidence says to. If that sounds like a place you would like to work, we would like to hear from you, even if there is no listed role that fits yet.

What we look for

1. Evidence over opinion. You want the experiment before the argument, and you report the result when it goes against you.
2. Care with small details. You notice when an edit matches twice, or a threshold was set on the wrong data.
3. Plain writing. You can explain what you built to someone outside the team in short sentences.
4. Range. You are strong in one area, such as models, systems or evaluation, and curious about the others.

Write to hello@pencilton.com with a few lines about something you have built and what you would like to work on.