From San Francisco — where software learns to listen.
A persistent agent that rewrites its own application logic.
The problem — a user complains. A human fixes it. Repeat forever. Mend closes the loop: it watches the complaint, builds the fix, and ships it behind your safety rails.
The UI change Mend wrote
router.get("/tasks", (req, res) => { const { tag } = req.query const where = tag ? {tags:{has:tag}} : {} res.json(db.task.findMany({ where }))})Why it’s safe to ship
The feedback loop
USER
complains
“Can't filter tasks by tag.”
Captured in-product or via support channel.
MEND
observes workflow
“47 users repeat the same hunt for a tag.”
Telemetry, session traces, drop-off points.
MEND
writes new endpoint + UI
“GET /tasks?tag= and a TagFilter control.”
Branch opened, tests added, diff prepared.
MEND
ships
“Promoted to 100% after Axiom + Lodestar pass.”
UI updates in real-time, no human in loop.
Guardrails
Schema limits, auth boundaries, rate envelopes, allowed domains. Mend cannot deploy a change that violates them.
Behavioural parity, regression coverage, invariant checks. Every diff is proven before it reaches production.
Mend listens, builds, proves, ships — while you sleep.