{
  "id": 6833368,
  "title": "How to Build an AI Software Factory: Agents That Open, Review, and Merge PRs",
  "url": "https://urgent.news/2026/09/11/how-to-build-an-ai-software-factory-agents-that-open-review-and-merge",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-11T23:43:53.000Z",
  "source": {
    "name": "Hacker News",
    "slug": "hacker-news",
    "url": "https://www.firecrawl.dev/blog/ai-software-factory"
  },
  "original_language": "en",
  "account": "An AI software factory is a system that integrates agents to automate software development tasks, with gates at each stage to control the flow of work. These gates ensure that only vetted and verified code progresses to the next stage, preventing a bottleneck where agents generate more work than teams can review. Spotify's Fleetshift, launched in 2023, successfully demonstrated this concept two years before they had an agent capable of handling the work. The key to an AI software factory is separating the agent (the coding tool) from the software factory (the infrastructure and processes that manage the agent). The agent requires live web context, which is why a curated developer index is integrated with search and scrape capabilities. One notable example is Stephen Toub, who, during a flight, opened nine pull requests that seven of them merged. This demonstrates how an AI agent can generate PRs faster than a team can review them, transforming the economics of code production. Companies like Stripe, Ramp, Spotify, and Block are building their own AI software factories, recognizing that the moat of software companies will shift from the code itself to the means of production that generates that code. The software factory moves steps like intake, workspace setup, and compilation verification into infrastructure, allowing humans to focus on accountability decisions at the end. A critical component of the software factory is a session log that records all interactions and results, enabling reproducibility and traceability. Successful implementations filter work before it reaches the agent, preventing unproductive tasks from being processed. This approach transforms autonomous coding agents from a demo into throughput that a team can actually absorb.",
  "summary": null,
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}