中文
Figma Config

An OpenAI PM directs product like a stage director: how one sentence — the 'spine' — gets a team to work the averaging machine will never produce

Be the director: how stage craft informs product craft ft. Kim Beverett (OpenAI) | Config 2026 · Isaac

22 min
AI Product

21 min total·Actually worth watching closely: ~8 min·3 must-watch clips

Orange = the 8 minutes worth watchingFor the rest, the guide is enough
Segment guide · 7 segments
  1. 0:11 2:40Listen

    Lesson one: there is no neutral default on stage

    The talk opens by putting a proscenium next to an app screen: both frame the audience's attention, both say focus here, forget that world and enter this world. The audience assumes everything they see was intentional, so leaving the curtain open or using a plain chair is a choice too — within five seconds users read meaning off the choices you didn't think you made.

    There is no neutral; the choices you didn't make still speak for you.

    A purely spoken analogy — the section has no visual dependency and the slides are supporting imagery, so following the logic is enough.▶ Jump to 0:11
  2. 2:43 6:02Listen

    The uncanny valley of choices: weird isn't the problem, unintentional is

    A button in a strange spot takes all of the user's attention and gives back only confusion — that's the uncanny valley of choices. Reversed, a strange choice that is clearly deliberate gets the audience interested and engaged from the moment they sit down.

    A weird choice can be a clear choice; what matters is that it reads as done on purpose.

    The argument runs on spoken contrast with nothing to watch — good for listening closely.▶ Jump to 2:43
  3. 6:05 10:33Skim

    A Midsummer Night's Dream and the spine

    Peter Brook's 1970 RSC production introduces the spine: a single overarching sentence to which every choice must adhere. Brook's was evocation over illustration — put a real forest on stage and the imagination turns off.

    It matters less what the spine is than that it exists and resonates with the whole team.

    There's a still of the 1970 white box worth a glance (around 512s), but the section is mostly narrative — a quick look while listening is enough.▶ Jump to 6:05
  4. 10:33 13:07Skim

    The white box wasn't one genius, and no form lasts

    The iconic white box is credited to set and costume designer Sally Jacobs — a strong spine is what let different disciplines make different choices that tell the same story. And Brook's Dream couldn't exist today, just as 2009's skeuomorphism is no longer needed.

    The same spine needs entirely new choices in a new era; the script is not the show.

    Production stills and the skeuomorphism comparison are worth skimming, but this is mostly argument — no need to watch frame by frame.▶ Jump to 10:33
  5. 13:07 14:36Watch

    Theater inside software: the cursor animation

    The computer use cursor goes far past what the feature required: a glow, personality, a designed path, a custom-built simulator, and days of engineering on corner cases where it didn't land just right. Every detail says what the team believes.

    One choice says it doesn't matter; one choice says everything matters.

    Real product motion on screen (799s is the LLM-flagged key visual) — the feel of the cursor has to be seen for the point to land.▶ Jump to 13:07
  6. 14:36 18:46Watch

    App Shots: a year of rehearsing with users

    The shortcut that pulls an app's screenshot and text into Codex took a year to get right: several approaches failed in a row, and the solution came from rehearsals with real users. The audience is a collaborator, a mirror, a witness.

    If good work hinges on your choices, those choices have to be validated with real users early and repeatedly.

    The App Shots interaction demo is on screen (876s is the key visual); the failed attempts next to the final form are worth seeing.▶ Jump to 14:36
  7. 18:50 21:27Listen

    A 1969 moment: don't expect the averaging machine

    The close lands on the AI era: build as if the rules were set in stone, hand AI a generic prompt, and you get the average of all of humanity's collective choices. There's nothing wrong with the tools, but the spine has to come from your own experience and intent.

    We cannot expect the averaging machine to produce anything other than averages — only you can supply the spine.

    A purely spoken closing with no visual dependency; it's where the whole argument condenses, so listening is the strongest way to take it in.▶ Jump to 18:50