The PodKnowledge|← All Episodes
Based on Lenny's Podcast data
Lenny's Knowledge Sketch

Every Engineer Is
Now a Manager

Sherwin Wu
Head of Engineering, OpenAI API & Platform
FEB 12 2026
The Shift

The Engineer's Job
Just Changed

PRs SHIPPED WITH CODEX
"Engineers are becoming tech leads. They're managing fleets and fleets of agents. It literally feels like we're wizards casting all these spells."
  • The shift was fast: from writing every line to reviewing what the agent wrote
  • Engineers now run 10 to 20 parallel Codex threads, steering more than typing
  • Trust climbs weekly as Codex proves it can be left to do more on its own
  • "This is the worst the models will ever be" sets the trajectory (Kevin Weil, OpenAI)
Framework

Programming as Sorcery

MACHINECODELANGUAGESPLAININTENT
95%
of OpenAI engineers use Codex
100%
of PRs reviewed by Codex
70%
more PRs from heavy Codex users
  • SICP called engineers "wizards" and code "incantations" back in 1980
  • AI is the next rung: tell Codex what you want and it goes and does it
  • Like the Sorcerer's Apprentice, huge leverage, but don't let the brooms run wild
  • The abstraction keeps rising: tab complete → full function → full feature
The insightA senior engineer fluent with these tools is now extraordinarily high-leverage. The seniority is exactly what keeps the agents from going off the rails.
Inside OpenAI

Running on 100% Codex

  • The experiment: one team maintains a fully Codex-written codebase
  • No escape hatch: they can't roll up their sleeves and hand-code the fix
  • Why it matters: removing the fallback forces the real agent problems into the open
  • The payoff: a blog post of paradigms and best practices is falling out of it
Most failures are context

When the agent goes wrong, it's usually underspecified. The fix is encoding tribal knowledge into the repo: comments, structure, .md files, Skills.

Code review, collapsed

Codex reviews every PR. A 10 to 15 minute review drops to 2 to 3 minutes; human attention goes from 100% to about 30%.

Playbook

Build Where Models Are Going

  • Build for where the models are heading, not where they sit today
  • Winning products work at ~80% today, then click when the next model ships
  • As a manager, spend more than half your time with your top 10% performers
  • For AI adoption, pair top-down buy-in with a bottoms-up evangelist team
The manager's moveLook around corners and clear blockers before they land. When engineers ship PR after PR, the bottleneck turns organizational, not technical.
Contrarian

Hot Takes on Building With AI

Always listen to your customersINSTEAD →The models will eat your scaffolding for breakfast. Chasing every ask traps you in a local maximum.
Engineering is AI's biggest prizeINSTEAD →Business process automation may be bigger. Most work is repeatable operations, not open-ended code.
OpenAI will squash your startupINSTEAD →The space is too big to worry about it. Startups die building things people don't want, not from getting crushed.
The billion-dollar startup is one person, aloneINSTEAD →That one person buys from a hundred tiny bespoke-software startups. A golden age of B2B SaaS.
Based on Sherwin Wu's episode on Lenny's Podcast. All ideas on this page are from the episode.Watch on YouTubeFollow @sherwinwu on X
Every sketch here began as a transcript.

Send us yours and we’ll draw it. $1 a transcript.

Sketch mine →
𝕏︎ X / Twitterin LinkedIn📸 Instagram🔗 Copy link
0:00