Harness Engineering is not Enough: Why Software Factories Fail — Dex Horthy, HumanLayer

Faster code generation can leave a team spending its time reviewing changes that solve the wrong problem or fit the existing system poorly. This talk on software factories argues for resolving those questions before delegating substantial work in an established codebase: agree on the intended behavior, component responsibilities and implementation sequence.

The proposed sequence uses small vertical slices, meaning changes that connect the necessary parts of a feature and can be checked along the way. Reviewers can then compare the implementation with decisions already made. This is the speaker's recommended response to maintainability problems encountered in practice; the speaker explicitly says there is no benchmark proving the broader claim that models cannot maintain code quality over time.

“no amount of harness engineering or loops maxing can solve what is fundamentally a model training issue”

Dex Horthy

“Verifying code quality and maintainability is orders of magnitude harder than the code runs and the test pass. Because the cost function of bad architecture is measured in months and years.”

Dex Horthy

“if the model knew what good code looks like, it would probably write it in the first place”

Dex Horthy

“If you're drowning in PRs, you actually have too many bad PRs.”

Dex Horthy

“30 minutes over here in pre-planning and alignment can save you hours in review”

Dex Horthy