Day 2 · Recorded 7 May 2026

A software factory in public: what happens when agents run the SDLC

Most agent demos stop at the pull request. Zacharias Malguitou asks the harder question: what if agents ran the whole software delivery loop, from planning and implementation through review, deployment, monitoring, bug triage, metrics, and documentation? His public Memo experiment turns the software factory from metaphor into something you can inspect.

Zacharias Malguitou, Software Factory, software-factory.dev

What's in this session

software-factory.dev is a live experiment in whether background agents can build more than snippets. Zacharias set out to build Memo, a Notion-style note-taking app, with agents handling the code while the human steers the spec and vision.

In this session, he walks through the factory's first 10 days: over 375 merged PRs, just under 68,000 lines of code, over 1,000 tests generated, and 16 background automations, all framed around the larger loop from planning to production feedback. The agenda also grounds the stack in Next.js, Supabase, and Sentry, with the full codebase public.

For platform leaders, the useful lesson is where the factory still needs structure. Agents need issue-shaped work, review loops, deployment signals, visual checks, monitoring, and human taste around the spec. The destination is not "more code faster"; it is a delivery system that keeps learning from the product it ships.

Inside the recording

  1. 00:00 A 10-day software factory experiment
    Memo was built in public with agents covering planning, code, review, deployment, and operations.
  2. 03:10 Factory metrics after the first run
    Hundreds of PRs, thousands of tests, and most agent executions triggered without human action.
  3. 04:34 Closing the SDLC loop
    User feedback, bugs, monitoring, and automation events feed work back into planning.
  4. 08:30 Feature builders, bug fixers, and PR reviewers
    The factory is a network of automations, not one prompt, with issues as the unit of work.
  5. 18:30 Taste, spec quality, and visual review
    Storybook and visual verification help agents improve UI before rough edges reach users.
  6. 23:30 Start small and fail visibly
    The practical path is one automation, visible failure modes, and loops that improve over time.