See how your team
actually ships with AI.
Every engineer tries agents differently. Nobody knows what is sticking.
ax is live per seat today — it turns one engineer’s local coding-agent sessions into receipts. The team aggregation layer that rolls those seats together is in development; we’re building it with design partners. A walkthrough runs on each engineer’s own local ax data, today.
The first AI rollout does not make a team AI-native.
Copilot, Cursor, Claude Code, ChatGPT — the tools arrive before the operating model. Some engineers build real agentic workflows; others stay at autocomplete. Leadership gets demos, not a shared way of shipping.
Tool chaos
Every engineer has a different stack, habit and prompt folder. ax shows the patterns under the sprawl.
No shared playbook
The useful workflows stay private until someone turns them into team practice. ax finds what is ready to teach or package.
Shipping feels the same
If cycle time isn’t moving, AI is still a side experiment. ax ties agent usage to the work that ships.
Real receipts, generated locally — today.
These are the numbers ax already surfaces from a single engineer’s history (run ax studio on your own machine). The team product aggregates these per-seat receipts into one view — same numbers, rolled up.
What ax found on one machine
Every number here is per-seat and local. The team layer aggregates them — never per-person behavior, never transcripts.
The mechanism that spreads a workflow already ships.
You don’t need the aggregation layer to move a good pattern across the team. Two shipped surfaces already do it.
Skills & hooks SDK
When ax finds a fix worth keeping, it becomes a skill or a typed Effect hook (Claude Code + Codex). Commit it once; everyone’s agent runs it. That’s one person’s trick becoming team practice, today.
Public profiles & leaders
ax profile publish turns one seat into a shareable profile. /leaders and /u/<login> are live multi-person surfaces on the aggregates-only model — proof the rollup works without sending anyone’s code.
The improve loop
ax mines repeated mistakes and proposes a small repo-specific fix, reviewed one at a time. Accept it and it’s in the repo — no aggregation server required.
A faster team has a different shape.
Gates open as a workflow goes from one person’s trick to team practice. The aggregation layer that rolls every seat into this curve is what we’re building with design partners — the scroll below sketches the shape it will draw.
Stuck gates are workflows still trapped in one head — ax finds them per seat today.
sketch of the team view — in development; per-seat receipts are live now (see below)
You see the exact JSON before anything leaves.
ax is AGPL-3.0 and runs entirely on each laptop. There is one real export today — ax profile publish — and it shows you the precise JSON in a consent prompt before the first byte moves. The team aggregation layer is built to the same contract: a local consent gate, ax-shaped aggregates only. Below is the design contract we’re building to.
what will leave (after consent)
- Per-seat adoption signal (active days, depth of use)
- Skill / workflow usage rollups (names, not contents)
- Cost & routing aggregates (the spend worth redirecting)
- Team-level aggregates (never per-person behavior)
what never leaves
- Transcript text & prompts (read locally, never sent)
- Your code, diffs and file contents (stay on disk)
- What each person is building (yours to keep)
- Anything not in the consent-prompt JSON you approved
Verifiable today: ax profile publish prints the full payload and waits for your yes. Source is AGPL-3.0 — read the line, don’t trust it.
What gets measured gets improved.
Right now AI adoption isn’t measured at all. The walkthrough runs on each engineer’s own local ax data — live today — and shows what stuck, what stayed shallow, and which workflows are ready to spread. We’re onboarding design partners for the team aggregation layer as we build it.