We work in agentic engineering every day. Each update takes what actually changed and hands it to your team in a form they can run.
A written piece explaining what the trend actually means. Exercises that end with the technique working in your own project. Files that stay in place afterwards.
One practice, renamed four times in five months. Each of these arrived with working tools behind it, and each one changed what a competent team does day to day.
Context engineering. Managing what the agent knows. What goes into its window on each turn, and what is kept out.
Harness engineering. Building the system that produces the work, instead of steering each output by hand.
Skills. Instructions committed to the repository, so a team shares one way of working instead of each person’s own prompt.
Loop engineering. Writing down what counts as done, and letting the agent keep working until it gets there.
Not named yet. Which is the point. Following it closely enough to choose is our job rather than yours.
Three parts, and the update works each one through in full.
What the trend actually means. What it is, what it is good for, and where it does not work. Worked through on real code, not described.
Exercises your team runs. Short exercises that end with the technique working in your own project, not just in a demo of ours.
It keeps working afterwards. Your team watches the technique run on their own project, and the hooks and skills that made it run stay in place, ready for the next piece of work.
Releases, research, and our own use of these tools across customer engineering work. Sorting what holds up in real code from what does not is the work.
Every technique is measured in our own repositories first. The first update shipped with more than sixty recorded runs behind it, in throwaway repositories, every result written down.
The update says what changed and what your team should do about it, in ordinary language, with the working files under it.
The page below is the teaser: what the technique is, and the example project it runs on. The update itself adds the four exercises that put each piece into a codebase you own, and the files that stay there afterwards.
Updates run each month, or every other month, whichever suits your team. You take one and decide from there. Everything arrives as files your team installs and runs, so nothing depends on us being in the room.