Agentic Software Factories Make Building Blocks the New Distribution
joelclaw's package seams, skills, and HATEOAS CLI can become agent-ready building blocks instead of features trapped inside one application
Mitchell Hashimoto argues that software adoption now flows through high-quality building blocks that agents can assemble into many small artifacts. His receipts are Ghostty, with one million daily macOS update checks after 18 months, and libghostty, which reached multiple millions of daily users through downstream integrations after two months. He points to Pi, Next.js, and Tailwind CSS as similar building-block systems.
The clever bit is the distribution loop. AI agents are better at joining proven, documented parts than rebuilding every layer from scratch. Each downstream fork or niche application spreads the primitive, tests a new use case, and gives the maintainer working research and development to inspect. The mainline application can stay narrow while the surrounding ecosystem explores faster and with a lower maintenance burden.
That maps cleanly to joelclaw: package interfaces, reusable skills, and a HATEOAS command-line interface are not only internal architecture. They are the surfaces agents can discover and recombine. The warning matters too: the building block must remain stable, documented, and load-bearing even when most downstream artifacts are temporary, narrow, or rough. Hashimoto leaves commercialization unresolved and notes that agents currently favor open-source software over closed commercial alternatives.
Key Ideas
- Mitchell Hashimoto defines a building block broadly enough to include libraries, frameworks, and forkable applications such as Ghostty.
- Agentic factories increase software quantity because they remove much of the knowledge barrier that once limited who could combine existing components.
- Downstream artifacts distribute the underlying component into niches the mainline application cannot serve without losing focus.
- Working forks turn ecosystem activity into outsourced research and development that maintainers can selectively bring back upstream.
- A lower quality bar for downstream artifacts does not excuse weak primitives; reusable building blocks need strong interfaces, documentation, and stability.
- Closed commercial software faces a discovery disadvantage when agents can inspect, import, and adapt open alternatives directly.