- Author
- Charity Majors
- Published
- Jun 15, 2026
- Kind
- Personal Substack essay, a reply to technical responses to an earlier piece
- Original
- charitydotwtf.substack.com/p/ai-demands-more-engineering-discipline
- Subject
- Coding with agents
What it is
Charity Majors wrote this follow-up to an earlier post on AI enthusiasts and skeptics. It answers readers who thought the earlier post urged shipping unreviewed code. It explains what excites the author about regenerable code.
What it says
- Majors says AI code now matches a median engineer's for common patterns.
- The economics changed: lines of code went from treasured to disposable and regenerable.
- Quoting Chad Fowler, the post treats code as a cache of understanding, disposable when stale.
- Failing to regenerate code the way servers are regenerated signals a lack of understanding.
- Humans are poor at validation, so rigor should move to tests, evals and production.
- Nondeterministic systems need more engineering discipline, not less.
- The post denies telling readers to skip code review and ship unread code.
What evidence it gives
The post offers argument, quotations from Fowler and the author's sysadmin history. It mentions one practice, killing the oldest Kafka node every Tuesday. It gives no measurements. Its estimate of teams with fast feedback loops is a guess.
Interests and limits
The post mentions Honeycomb, a devtools company, and its AI mandate. The author says to treat the AI views as those of a non-expert. Nothing was tested. Only two reader comments were shown on the page. The post says the tools for regenerating code from specs do not exist yet.