Software factory publication
A free publication about software factories: reported company cases and scale figures, every article linked to its sources.
How to combine three ways of organizing agent work: a production line for one repeatable job, Geoffrey Huntley's Ralph loop, and Steve Yegge's multi-agent work graph.
Read the field analysis
A personal manifesto about building more software by improving the system around coding agents.
The details that make each approach work: its tools, feedback, division of labor and review.
Jane Street's coding agents test terminal apps by reading the screen as text. In June 2026 the firm said agentic coding had led it to start building a formal-methods team.
Read the caseAn employee delegates once. Minions prepares a development computer, changes code, runs checks, and returns the result for two stages of human review.
Read the caseEmployees work with River in shared Slack threads. Shopify's Aquifer design keeps the work history separate from temporary computers so sessions can survive interruptions.
Read the caseCompare the work, the scale and the evidence. Software development is the starting point; operations and research widen the picture.
Find a relevant workflow and compare its implementation, review and public evidence.
Inspect reported worker counts, runs, output and adoption, with the units and limits beside each number.
Follow the uses emerging in finance, automotive software, industrial planning and scientific research.
Clear explanations for understanding a proposal, planning an experiment or judging its results.
The people, agents, checks, and learning loop that make up an AI software factory.
How requests reach agents, how results are checked and released, and how the team learns from the work.
Choose a recurring job, compare it with today's process, prepare the agent's tools and tests, and measure what it takes to deliver an accepted result.
Five numbers for judging speed, rework, quality, human attention, and the real cost of AI-assisted delivery.
Production lineA repeatable task moves through implementation, checks and release.
Embedded loopAn agent tries a change, checks the result and tries again.
Agent organizationA shared task system coordinates workers and their dependencies.
