Saturday Jun 6
STRATEGYOther

Honeycomb CTO Reframes The AI Debate

2JUN
WINSCOSTS

The AI coding debate has a structural bug, not an opinion problem. Honeycomb CTO Charity Majors argues wins get loud in all-hands while costs hide in SRE retros, so the loop never closes. Treat it as engineering, not culture war.

Charity Majors reframes the AI coding fight as plumbing, not culture. Both sides see real evidence, neither sees the same room.

Fin's CTO got 3x more PRs in 9 months, with defects halved. But code quality declined for 18 months before recovering. It required a decade of engineering discipline to pull off.

Majors's question for skeptics: 'what would it take to feel comfortable shipping?' Answer that honestly and the team unsticks itself.

full brief & sources

⚡ Why this matters

  • The AI coding fight isn't a culture war — it's a structural gap between teams that ship and teams that operate.
  • Most teams have evidence for BOTH 'AI accelerates us' AND 'AI breaks production.' Without a feedback loop, both grow louder, not closer.
  • Treating this as engineering (build the loop) instead of opinion (win the debate) is the only path that scales past 50 engineers.

🔍 What happened

  • Charity Majors, Honeycomb CTO, published an essay arguing the AI coding debate has a structural feedback bug.
  • Her framing: wins get announced in all-hands; costs surface in SRE retros and on-call pages. Same company, different rooms.
  • Cited example: Fin (formerly Intercom). CTO Darragh Curran asked R&D to 2x output, got 3x merged PRs in 9 months.
  • Side effects at Fin: defect backlog halved, downtime down 35%, code quality declined for 18 months before recovering.
  • Recovery required a decade of engineering discipline already in place. Most teams don't have that base.
  • Prescription: tell the whole story, close the feedback loop, treat as an engineering problem.

💬 Smart takes

  • Charity Majors, Honeycomb CTO: 'Ask skeptics what would it take for you to feel comfortable shipping without reading it?'
  • Darragh Curran, Fin CTO: 3x merged PRs in 9 months, but 18 months of code-quality decline before recovery.
  • Skeptic: Without Fin's decade of engineering rigor, the 18-month dip becomes a permanent productivity tax.

🧭 Where this goes

  1. LikelyAI tooling teams add 'cost surface' metrics (defect rate, MTTR, on-call hours) to their main dashboards by Q4.
  2. Likelya wave of CTO essays over the next 6 months echoing Majors' loop framing as the topic crystallizes.
  3. Possiblea vendor builds a 'shipped-with-AI / paged-after-AI' linking dashboard as a new product category.
  4. Wild CardHoneycomb launches an AI-coding observability SKU within 12 months, productizing Majors' essay.

🥄 The Spoon Take

This is a maturity test for AI tooling teams. The leaders will track wins AND costs in the same forum and close the loop fast. The rest will keep yelling past each other while production debt compounds quietly. Fin is the case study — without engineering discipline, the 18-month quality dip is a death zone.

🤔 Pushback

Most teams don't have Fin's decade of engineering rigor, so the 18-month quality dip turns into permanent debt instead of a comeback.

·Charity Majors - Enthusiasts vs Skeptics·Simon Willison - Link post·Fin/Intercom - 2x results after 9 months