vidtheque.

read-only demo

Videos WkBPX-oDMnA

Understanding is the new bottleneck — Geoffrey Litt, Notion

index_state ready data_status ok

AI Engineer· published 2026-07-10· 0:19:33· en-US· indexed 2026-08-10 19:46

Open on YouTube

Scene timeline

  1. Shot 0, 0:00 to 0:03, 1 of 1 keyframes kept
  2. Shot 1, 0:03 to 0:05, 1 of 1 keyframes kept
  3. Shot 2, 0:05 to 0:12, 1 of 1 keyframes kept
  4. Shot 3, 0:12 to 0:16, 1 of 1 keyframes kept
  5. Shot 4, 0:16 to 0:25, 1 of 1 keyframes kept
  6. Shot 5, 0:25 to 0:32, 1 of 1 keyframes kept
  7. Shot 6, 0:32 to 1:12, 1 of 1 keyframes kept
  8. Shot 7, 1:12 to 1:31, 0 of 1 keyframes kept
  9. Shot 8, 1:31 to 1:59, 1 of 1 keyframes kept
  10. Shot 9, 1:59 to 2:27, 0 of 1 keyframes kept
  11. Shot 10, 2:27 to 2:55, 1 of 1 keyframes kept
  12. Shot 11, 2:55 to 3:04, 1 of 1 keyframes kept
  13. Shot 12, 3:04 to 3:31, 1 of 1 keyframes kept
  14. Shot 13, 3:31 to 3:58, 0 of 1 keyframes kept
  15. Shot 14, 3:58 to 4:25, 0 of 1 keyframes kept
  16. Shot 15, 4:25 to 4:54, 0 of 1 keyframes kept
  17. Shot 16, 4:54 to 5:22, 0 of 1 keyframes kept
  18. Shot 17, 5:22 to 6:02, 1 of 1 keyframes kept
  19. Shot 18, 6:02 to 6:28, 0 of 1 keyframes kept
  20. Shot 19, 6:28 to 6:44, 1 of 1 keyframes kept
  21. Shot 20, 6:44 to 6:52, 0 of 1 keyframes kept
  22. Shot 21, 6:52 to 7:29, 0 of 1 keyframes kept
  23. Shot 22, 7:29 to 7:38, 1 of 1 keyframes kept
  24. Shot 23, 7:38 to 7:57, 0 of 1 keyframes kept
  25. Shot 24, 7:57 to 8:21, 0 of 1 keyframes kept
  26. Shot 25, 8:21 to 8:50, 0 of 1 keyframes kept
  27. Shot 26, 8:50 to 9:29, 1 of 1 keyframes kept
  28. Shot 27, 9:29 to 9:49, 0 of 1 keyframes kept
  29. Shot 28, 9:49 to 9:50, 0 of 1 keyframes kept
  30. Shot 29, 9:50 to 10:07, 1 of 1 keyframes kept
  31. Shot 30, 10:07 to 10:43, 0 of 1 keyframes kept
  32. Shot 31, 10:43 to 10:58, 1 of 1 keyframes kept
  33. Shot 32, 10:58 to 11:00, 1 of 1 keyframes kept
  34. Shot 33, 11:00 to 11:02, 0 of 1 keyframes kept
  35. Shot 34, 11:02 to 11:25, 0 of 1 keyframes kept
  36. Shot 35, 11:25 to 11:41, 1 of 1 keyframes kept
  37. Shot 36, 11:41 to 11:56, 1 of 1 keyframes kept
  38. Shot 37, 11:56 to 12:03, 0 of 1 keyframes kept
  39. Shot 38, 12:03 to 12:37, 1 of 1 keyframes kept
  40. Shot 39, 12:37 to 13:03, 0 of 1 keyframes kept
  41. Shot 40, 13:03 to 13:49, 1 of 1 keyframes kept
  42. Shot 41, 13:49 to 14:10, 1 of 1 keyframes kept
  43. Shot 42, 14:10 to 14:21, 1 of 1 keyframes kept
  44. Shot 43, 14:21 to 14:30, 0 of 1 keyframes kept
  45. Shot 44, 14:30 to 14:40, 0 of 1 keyframes kept
  46. Shot 45, 14:40 to 14:57, 0 of 1 keyframes kept
  47. Shot 46, 14:57 to 15:17, 0 of 1 keyframes kept
  48. Shot 47, 15:17 to 15:36, 1 of 1 keyframes kept
  49. Shot 48, 15:36 to 16:05, 0 of 1 keyframes kept
  50. Shot 49, 16:05 to 16:19, 0 of 1 keyframes kept
  51. Shot 50, 16:19 to 16:59, 0 of 1 keyframes kept
  52. Shot 51, 16:59 to 17:05, 0 of 1 keyframes kept
  53. Shot 52, 17:05 to 17:34, 0 of 1 keyframes kept
  54. Shot 53, 17:34 to 18:19, 1 of 1 keyframes kept
  55. Shot 54, 18:19 to 18:44, 1 of 1 keyframes kept
  56. Shot 55, 18:44 to 18:54, 1 of 1 keyframes kept
  57. Shot 56, 18:54 to 19:15, 1 of 1 keyframes kept
  58. Shot 57, 19:15 to 19:32, 0 of 1 keyframes kept

58 shot(s).

keyframes kept every frame deduplicated

What was stored

cues
256
whisperx 256
chunks
35
from 256 cues
keyframes
29
kept of 58 captured
frames with text
29
471 lines read
chapters
0
from the source metadata
keyframe bytes
6.6 MB
word timings on 256 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage state model started took
fetch done — 2026-08-10 01:45 1m 31s
stt done — 2026-08-10 01:46 21s
chunk done — 2026-08-10 01:47 0s
text_embed done — 2026-08-10 19:46 0s
keyframe done — 2026-08-10 01:47 2m 00s
ocr done — 2026-08-10 01:49 12s
frame_embed done — 2026-08-10 19:46 5s

Frames, and what the machine read

  • 0:02 #0 done2 line(s)

    shot 0·sharpness 452.3

    1. AlEngineer0.96
    2. World's Fair1.00

  • 0:03 #1 done2 line(s)

    shot 1·sharpness 658.6

    1. AlEngineer0.95
    2. World's Fair0.99

  • 0:10 #2 done24 line(s)

    shot 2·sharpness 2733.3

    1. LAB & PLATINUM SPONSORS0.99
    2. Amazon AGI Lab0.98
    3. ANTHROP\C1.00
    4. Google DeepMind1.00
    5. MINIMAX0.94
    6. OpenAI0.93
    7. Akamai1.00
    8. arize1.00
    9. aws1.00
    10. Braintrust bright data0.98
    11. B1.00
    12. Browserbase1.00
    13. docker1.00
    14. :neo4j0.91
    15. ORACLE1.00
    16. PayPal1.00
    17. qodo1.00
    18. reducto1.00
    19. Sonar1.00
    20. Makers of1.00
    21. togetherai1.00
    22. Unblocked1.00
    23. WorkOS1.00
    24. SonarQube1.00

  • 0:14 #3 done2 line(s)

    shot 3·sharpness 291.3

    1. World's Fair0.95
    2. AlEngineer1.00

  • 0:23 #4 done2 line(s)

    shot 4·sharpness 182.6

    1. AlEngineer0.99
    2. World's Fair0.99

  • 0:30 #5 done11 line(s)

    shot 5·sharpness 1900.2

    1. AlEngineer0.98
    2. World'sFair0.96
    3. Understanding is the0.99
    4. new bottleneck1.00
    5. PRESENTED BY0.99
    6. Microsoft1.00
    7. Geoffrey Litt1.00
    8. → Design Engineer at Notion | @geoffreylitt0.97
    9. N0.91
    10. Engineering the future of Al1.00
    11. World'sFair1.00

  • 1:07 #6 done10 line(s)

    shot 6·sharpness 2181.4

    1. AlEngineer0.98
    2. World'sFair1.00
    3. It's important for humans to1.00
    4. understand how the code works.1.00
    5. PRESENTED BY1.00
    6. Microsoft1.00
    7. N0.98
    8. World's Fair0.96
    9. TRACK 6· JULY 1, 20260.96
    10. Al Designers/Design Engineers0.99

  • 1:20 #7 skipped

    shot 7·duplicate of #6

  • 1:50 #8 done27 line(s)

    shot 8·sharpness 2436.5

    1. AlEngineer0.98
    2. How I understand agent code:0.99
    3. World'sFair1.00
    4. Background1.00
    5. Code explainers1.00
    6. The Engine: Phaser 3 and the Flat Canvas0.99
    7. The Zen Garden is built on Phaser 3, a 2D browser game framework. Phaser renders everything0.99
    8. onto an HTML5 ecamvas» element - a flat grid of pèxels. There is no native notion of depth,0.96
    9. height, or three-dimensional space. When you place a rock, Phaser stamps a sprite image at an0.98
    10. Ix. vi coordinate on the screen and moves.on0.89
    11. 1. Why did changing the garden ellipse from a ~1:1 ratio to a 2.4:1 (wide:tall) ratio0.99
    12. make it appear to be a tilted surface?1.00
    13. Quizzes1.00
    14. Because wider shapes are always perceived as three-dimensional.1.00
    15. Because foreshortening tums a tilted horizontal circle into a flattened ellipse, and0.98
    16. the visual system inverts that relationship when it sees the ellipse on a flat screen.0.99
    17. Because Phaser's renderinao engine anolies an automatic isometric transform to0.96
    18. Micro-worlds1.00
    19. <> Program0.98
    20. Stack1.00
    21. father(orville, abe).1.00
    22. R0 Goals0.94
    23. father(abe, homer).1.00
    24. 0:@0.66
    25. World's Fair0.98
    26. TRACK 6· JULY 1,20260.95
    27. Al Designers/Design Engineers0.97

  • 2:03 #9 skipped

    shot 9·duplicate of #8

  • 2:51 #10 done7 line(s)

    shot 10·sharpness 1504.9

    1. AlEngineer0.98
    2. World'sFair0.98
    3. Understand to verify1.00
    4. N0.98
    5. Worid's Fair0.92
    6. TRACK 6· JULY 1, 20260.96
    7. Al Designers/Design Engineers0.99

  • 3:00 #11 done25 line(s)

    shot 11·sharpness 2711.1

    1. Beyang1.00
    2. AlEngineer1.00
    3. @beyang1.00
    4. World's Fair0.98
    5. The bottleneck in coding is now the time it takes a human to review1.00
    6. agent-generated code.1.00
    7. Code review is1.00
    8. Sam Saccone0.98
    9. the bottleneck1.00
    10. @samccone1.00
    11. Everyone is waking up to the fact that code-review is the bottleneck of1.00
    12. 20261.00
    13. Logan Kilpatrick1.00
    14. @OfficialLoganK1.00
    15. The bottleneck has so quickly moved from code generation to code0.99
    16. review that it is actually a bit jarring.0.99
    17. None of the current systems / norms are setup for this world yet.0.99
    18. 9:53 PM·Mar 16,2026·523K Views0.97
    19. Aiden Bai0.98
    20. @aidenybai0.95
    21. N1.00
    22. the bottleneck for coding agents is now testing / code quality0.98
    23. World's Fair0.98
    24. TRACK 6· JULY 1,20260.95
    25. AI Designers/Design Engineers0.97

  • 3:18 #12 done6 line(s)

    shot 12·sharpness 1973.6

    1. AlEngineer0.99
    2. World'sFair1.00
    3. Correct?1.00
    4. World'sFair0.98
    5. TRACK 6· JULY 1, 20260.96
    6. Al Designers/Design Engineers0.99

  • 3:35 #13 skipped

    shot 13·duplicate of #12

  • 4:22 #14 skipped

    shot 14·duplicate of #12

  • 4:50 #15 skipped

    shot 15·duplicate of #10

  • 5:05 #16 skipped

    shot 16·duplicate of #10

  • 5:53 #17 done16 line(s)

    shot 17·sharpness 3294.4

    1. AlEngineer0.97
    2. Cognitive debt1.00
    3. World'sFair1.00
    4. “Even if Al agents produce code that could be easy0.99
    5. to understand, the humans involved may have0.99
    6. PRESENTED BY1.00
    7. simply lost the plot and may not understand what1.00
    8. the program is supposed to do, how their intentions1.00
    9. Microsoft1.00
    10. were implemented, or how to possibly change it.1.00
    11. Margaret Storey1.00
    12. https://margaretstorey.com/blog/2026/02/09/cognitive-debt/1.00
    13. N0.65
    14. World'sFair1.00
    15. TRACK 6· JULY 1, 20260.93
    16. Al Designers/Design Engineers0.98

  • 6:10 #18 skipped

    shot 18·duplicate of #6

  • 6:42 #19 done5 line(s)

    shot 19·sharpness 1073.4

    1. AlEngineer0.99
    2. World'sFair1.00
    3. World's Fair0.92
    4. TRACK 6· JULY 1, 20260.95
    5. Al Designers/Design Engineers0.99

  • 6:50 #20 skipped

    shot 20·duplicate of #11

  • 7:18 #21 skipped

    shot 21·duplicate of #10

  • 7:37 #22 done12 line(s)

    shot 22·sharpness 1456.6

    1. AlEngineer0.99
    2. World's Fair0.96
    3. RAKE1.00
    4. ROCK1.00
    5. SHRUB0.94
    6. TEAHOUSE1.00
    7. CLEAR1.00
    8. SOUND1.00
    9. N1.00
    10. World'sFair0.98
    11. TRACK 6· JULY 1, 20260.96
    12. Al Designers/Design Engineers0.99

  • 7:55 #23 skipped

    shot 23·duplicate of #10

Transcript

256 cues· 3,686 words· 19,513 chars

  1. 0:12 What's up, yeah, thank you for coming to the design engineering track at AIE.
  2. 0:16 Is everyone having fun?
  3. 0:18 Yeah, I think this is gonna be a great track, so get excited.
  4. 0:22 All right, let's get going.
  5. 0:24 My name's Jeffrey Litt.
  6. 0:26 I'm a design engineer at Notion currently, and I'm here to drop a hot take for this room, maybe.
  7. 0:35 I think it is still important for people to understand how code works.
  8. 0:42 Now some of you might agree, some of you might disagree.
  9. 0:44 Let's try a poll, I'm curious for this room.
  10. 0:47 Raise your hand if you agree with that opinion.
  11. 0:50 Okay, maybe some selection bias.
  12. 0:52 Okay, raise your hand if you disagree with this opinion.
  13. 0:55 Wow, okay.
  14. 0:57 Maybe we'll do a debate later, yeah.
  15. 0:59 I was hoping we'd be at the AI Engineer Conference so we'd have more, okay.
  16. 1:03 I might be preaching to the choir here.
  17. 1:06 You know, I think the reality is that we are entering an era where this is a legitimate question that people are debating, right?
  18. 1:13 Agents are writing tons of code for us.
  19. 1:15 They're landing 50,000 line PRs.
  20. 1:18 And it is getting harder to keep up.
  21. 1:20 We all feel this.
  22. 1:22 Now, I think the good news is there are lots of ways to understand.
  23. 1:28 The days of just reading code line by line, that's not the only way anymore.
  24. 1:32 And what the point of this talk is about is I want to share with you a bunch of the practices that I use to understand the code that my agents are writing for me.
  25. 1:42 This includes things like explainer docs, teaching me about how my code works.
  26. 1:47 My agents write quizzes for me to test my understanding.
  27. 1:51 Am I still really in the loop?
  28. 1:52 Am I keeping up?
  29. 1:55 I have agents build me micro worlds that I can inhabit to get this intuitive sense of how my code works that's deeper and richer than just a written document.
  30. 2:05 And I think all of these are really exciting new possibilities that are opening up for AI to help us understand better, not worse.
  31. 2:13 And so that's what the point of this talk is gonna be about.
  32. 2:15 And I hope I can leave you with some techniques that you can take home and use yourself.
  33. 2:21 By the way, my timer isn't running.
  34. 2:23 If you could get that running, that'd be great so I know if I'm blabbering.
  35. 2:28 Okay, but let's start, let's back up for a sec.
  36. 2:30 Before we talk about how, let's talk about why.
  37. 2:33 Why bother understanding?
  38. 2:35 This is, again, it's a question now, right?
  39. 2:37 And I think a lot of people get this subtly wrong.
  40. 2:42 So what a lot of people think of why do humans still have to understand, they think we understand to verify.
  41. 2:48 The agents do dumb stuff, we've all seen it, and your job as the human is to keep them in line, right?
  42. 2:53 Make sure they don't screw up.
  43. 2:56 When people say things like code review is the new bottleneck, I think that's the first thing that pops into people's heads is correctness, checking.
  44. 3:05 There's this mental model that's like, hey, the agent's gonna give you something, and what's your job?
  45. 3:10 It's to ask, is this correct?
  46. 3:13 Now, correctness can have lots of definitions.
  47. 3:15 Does it match the spec doc you gave it?
  48. 3:17 Does it take down production?
  49. 3:19 Is it well architected?
  50. 3:21 But fundamentally, those are all kind of thumbs up, thumbs down decisions, right?

Open at this second