Pocock says GLOSSARY.md makes him feel more productive
A one file shared vocabulary for your agent, backed so far by a feeling rather than a measurement.
Matt Pocock is asking people how GLOSSARY.md is working out for them. The file is generated by his /grill-with-docs skill, and he says the more work he puts into the glossary, the more productive he feels, because it is a warm, comforting feeling having the agent speak his language.
the more work I put into the glossary, the more productive I feel. It's a warm, comforting feeling having the agent speak my language.
That is the whole claim, and Pocock frames it as a personal impression rather than a result. He is not reporting a benchmark, a token count or a before and after on a codebase. He is asking other users what they are seeing, which is the honest version of a workflow tip that has only one data point behind it so far.
What the file is for
A glossary is a single markdown file naming the terms your domain actually uses, so the agent reaches for your words instead of inventing its own. If your codebase calls something a run, a job and a task in three different places, the agent will cheerfully pick a fourth. Writing the vocabulary down once gives the model a fixed reference it can consult instead of guessing from surrounding code, and gives you something to point at when it guesses wrong anyway.
/grill-with-docs is the skill Pocock has been building out. He has previously described how he uses it in his agent setup. GLOSSARY.md is a byproduct of that process rather than something you are expected to write by hand up front, which is part of why it is cheap to try.
For working engineers
Treat this as a suggestion with a low cost of being wrong, not as evidence. One file, generated once, and you find out within a few sessions whether the agent stops renaming your concepts. It also sits in the same family as the other context files engineers keep adding, which means the usual caveats apply. Anything you write down has to stay true, and a glossary that drifts from the code is a confident source of wrong answers.
The open question is whether the gain is the file or the act of writing it. Pocock says the productivity scales with how much work he puts in, which is as consistent with him clarifying his own thinking as with the agent reading better. Nobody has measured it yet, and Pocock's thread is him looking for people who can.
Have the good opinion, not the loud one
One email a day. Under three minutes. What working engineers found out about coding with AI.
