AnyFrame uses its own AI agent "Gilfoyle" to ship production code
AnyFrame built an internal developer agent named Gilfoyle on its own platform, and the AI agent has shipped full features — including database migrations and UI — to production under its own commit identity.
Score breakdown
The post demonstrates a concrete case where an AI coding agent autonomously shipped a complete feature — database migration and all — to a production codebase, with the "proof-of-work" screenshot/live-URL mechanism replacing the traditional human review step.
- 01AnyFrame built an internal AI developer agent named "Gilfoyle" running on its own platform — a control plane for sandboxed AI agents.
- 02Gilfoyle is triggered via Discord @mentions, with each thread mapped to one isolated cloud sandbox and one unit of work.
- 03The agent operates under its own commit identity, opening and merging pull requests autonomously.
AnyFrame describes how it built and deployed "Gilfoyle," an internal AI developer agent running on its own platform, and used it to ship production code to the AnyFrame product itself. Gilfoyle is not a chatbot: it lives in an isolated cloud sandbox with the company's full monorepo checked out, a terminal open, and the ability to clone repos, run tests, and open pull requests under its own commit identity. Setup took three steps — creating a template (defining the repo, install command, system prompt, and reusable "skills"), spinning up an agent from that template with a Claude runtime, and wiring a Discord integration so any team member can @mention Gilfoyle in a channel to kick off a task. Each mention spawns a dedicated thread backed by a fresh sandbox; if the sandbox is evicted between messages, the next message resumes from a snapshot transparently.
First, AnyFrame consolidated its backend, dashboard, SDK, and integrations into a single monorepo so the agent has full cross-cutting context for tasks that span multiple services.
Two architectural choices are credited with making Gilfoyle effective. First, AnyFrame consolidated its backend, dashboard, SDK, and integrations into a single monorepo so the agent has full cross-cutting context for tasks that span multiple services. Second, a "proof-of-work" skill ships Chromium and Playwright inside every sandbox, so Gilfoyle boots the dev server, drives a real browser to the changed page, and attaches a screenshot — or a live public tunnel URL — directly to the pull request comment before anything merges. The post highlights PR #128, which added recurring agent session scheduling: a database table, background scheduler, migration, and UI, all opened autonomously by Gilfoyle and merged without any human writing code. A cron-triggered instance of Gilfoyle also runs every weekday at 9am to hunt dead code and open draft PRs with no human in the loop.
Key facts
- 01AnyFrame built an internal AI developer agent named "Gilfoyle" running on its own platform — a control plane for sandboxed AI agents.
- 02Gilfoyle is triggered via Discord @mentions, with each thread mapped to one isolated cloud sandbox and one unit of work.
- 03The agent operates under its own commit identity, opening and merging pull requests autonomously.
- 04AnyFrame consolidated its backend, dashboard, SDK, and integrations into a single monorepo to give Gilfoyle full cross-cutting context.
- 05A "proof-of-work" skill uses Chromium and Playwright to attach browser screenshots or live public tunnel URLs to PRs before code merges.
- 06PR #128 — recurring agent session scheduling, including a database table, background scheduler, migration, and UI — was shipped entirely by Gilfoyle with no human writing code.
- 07A cron-scheduled Gilfoyle instance runs every weekday at 9am to find dead code and open draft PRs automatically.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 14, 2026 · 09:08 UTC. How this works →