App Growth Agent

App Growth Agent was a compact attempt to turn early Shopify-app growth into an observable loop: find merchants describing a real problem, qualify the signal, draft a useful message, force the draft through policy and human review, attribute an install, detect actual product value, and ask for a review only after that value existed. The experiment treated growth less like a stream of clever copy and more like a product system with state, evidence, and explicit reasons to stop.

The prototype did not equate autonomy with unrestricted sending. App and campaign context packs constrained claims and tone; Reddit messages required approval by default; cadence rules and policy lint could block a send; and a global outbound kill switch overrode the pipeline. Agent runs, steps, LLM calls, costs, errors, and fallbacks were modeled as inspectable records. The useful design idea was a control plane where the operator could see why an agent planned, qualified, rewrote, queued, or refused an action.

The source was built from February 2 through February 6, 2026. It reached a locally runnable dashboard, seeded acquisition and activation data, connectors, authentication, orchestration traces, and split web/worker deployment plans. It did not reach the source docs' next milestone of validating the real end-to-end loop in production. The former public endpoint now returns 404, and Maggie classifies the project as killed. Hyphenomenon keeps it because experiments that stop still contain product judgment: the ambition, the safety architecture, and the distance between an integrated prototype and a proven growth engine.