Adapting for a world of software factories
Summary
Zach Lloyd argues software engineering is splitting into two distinct disciplines: product engineering to ensure quality and factory engineering to optimize the autonomous agent systems that build the product.
Engineering requires two distinct roles.
Product engineering focuses on usefulness and design, while factory engineering optimizes the agentic systems, infrastructure, and model mixes used to generate code.
Product engineering: ensuring the quality and usefulness of product
Automation replaces the inner loop.
Work moves from private local setups to public cloud environments where agents handle implementation, adversarial reviews, and even produce computer-use videos.
prompting agents that live in the cloud through knowledge work tools
Human taste prevents automation failure.
Agents cannot replace the human ability to understand context, design, and user experience, though they can automate simple bugs and dependency upgrades.
Humans have taste, and humans for the time being have more context
Efficiency is measured by human touches.
Warp tracks progress by reducing 'human touches per PR,' aiming to increase the percentage of work handled autonomously by the factory.
At Warp we measure “human touches per PR,”
Factory engineering is a technical discipline.
Improving the factory requires using DORA metrics and LLM-as-a-judge scorers to benchmark performance rather than relying on intuition or vibes.
you don’t adjust your factory on vibes; you measure, test, [benchmark
Highlights
Our engineering team still is responsible for the quality and usefulness of the product – this cannot be delegated to agents.
Read the original ↗Save to your Sponge
Summarized with Sponge. Quotes are the author’s own words.