vidtheque.

read-only demo

Videos xIt_mTQp6mY

Loop Engineering from First Principles — Kyle Mistele, HumanLayer

index_state ready data_status ok

AI Engineer· published 2026-07-25· 0:17:57· en-US· indexed 2026-08-10 19:39

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 0:18, 1 of 1 keyframes kept
  5. Shot 4, 0:18 to 0:21, 1 of 1 keyframes kept
  6. Shot 5, 0:21 to 0:33, 1 of 1 keyframes kept
  7. Shot 6, 0:33 to 0:47, 1 of 1 keyframes kept
  8. Shot 7, 0:47 to 1:32, 1 of 1 keyframes kept
  9. Shot 8, 1:32 to 1:54, 1 of 1 keyframes kept
  10. Shot 9, 1:54 to 2:02, 0 of 1 keyframes kept
  11. Shot 10, 2:02 to 2:13, 1 of 1 keyframes kept
  12. Shot 11, 2:13 to 2:31, 1 of 1 keyframes kept
  13. Shot 12, 2:31 to 2:39, 1 of 1 keyframes kept
  14. Shot 13, 2:39 to 2:55, 1 of 1 keyframes kept
  15. Shot 14, 2:55 to 3:13, 1 of 1 keyframes kept
  16. Shot 15, 3:13 to 3:25, 1 of 1 keyframes kept
  17. Shot 16, 3:25 to 3:32, 1 of 1 keyframes kept
  18. Shot 17, 3:32 to 3:35, 1 of 1 keyframes kept
  19. Shot 18, 3:35 to 3:44, 1 of 1 keyframes kept
  20. Shot 19, 3:44 to 3:49, 1 of 1 keyframes kept
  21. Shot 20, 3:49 to 4:03, 1 of 1 keyframes kept
  22. Shot 21, 4:03 to 4:08, 1 of 1 keyframes kept
  23. Shot 22, 4:08 to 4:32, 1 of 1 keyframes kept
  24. Shot 23, 4:32 to 5:00, 1 of 1 keyframes kept
  25. Shot 24, 5:00 to 5:27, 0 of 1 keyframes kept
  26. Shot 25, 5:27 to 5:45, 1 of 1 keyframes kept
  27. Shot 26, 5:45 to 6:02, 1 of 1 keyframes kept
  28. Shot 27, 6:02 to 6:28, 1 of 1 keyframes kept
  29. Shot 28, 6:28 to 6:38, 1 of 1 keyframes kept
  30. Shot 29, 6:38 to 7:00, 1 of 1 keyframes kept
  31. Shot 30, 7:00 to 7:03, 1 of 1 keyframes kept
  32. Shot 31, 7:03 to 7:36, 1 of 1 keyframes kept
  33. Shot 32, 7:36 to 8:05, 1 of 1 keyframes kept
  34. Shot 33, 8:05 to 8:13, 1 of 1 keyframes kept
  35. Shot 34, 8:13 to 8:28, 1 of 1 keyframes kept
  36. Shot 35, 8:28 to 9:07, 0 of 1 keyframes kept
  37. Shot 36, 9:07 to 9:22, 1 of 1 keyframes kept
  38. Shot 37, 9:22 to 9:40, 0 of 1 keyframes kept
  39. Shot 38, 9:40 to 9:49, 1 of 1 keyframes kept
  40. Shot 39, 9:49 to 10:10, 1 of 1 keyframes kept
  41. Shot 40, 10:10 to 10:22, 0 of 1 keyframes kept
  42. Shot 41, 10:22 to 10:29, 0 of 1 keyframes kept
  43. Shot 42, 10:29 to 10:35, 1 of 1 keyframes kept
  44. Shot 43, 10:35 to 10:41, 0 of 1 keyframes kept
  45. Shot 44, 10:41 to 10:47, 0 of 1 keyframes kept
  46. Shot 45, 10:47 to 10:59, 1 of 1 keyframes kept
  47. Shot 46, 10:59 to 11:10, 1 of 1 keyframes kept
  48. Shot 47, 11:10 to 11:17, 0 of 1 keyframes kept
  49. Shot 48, 11:17 to 11:20, 1 of 1 keyframes kept
  50. Shot 49, 11:20 to 11:37, 1 of 1 keyframes kept
  51. Shot 50, 11:37 to 11:44, 1 of 1 keyframes kept
  52. Shot 51, 11:44 to 11:51, 0 of 1 keyframes kept
  53. Shot 52, 11:51 to 12:32, 1 of 1 keyframes kept
  54. Shot 53, 12:32 to 12:59, 1 of 1 keyframes kept
  55. Shot 54, 12:59 to 13:18, 1 of 1 keyframes kept
  56. Shot 55, 13:18 to 13:37, 1 of 1 keyframes kept
  57. Shot 56, 13:37 to 13:45, 1 of 1 keyframes kept
  58. Shot 57, 13:45 to 13:49, 1 of 1 keyframes kept
  59. Shot 58, 13:49 to 14:11, 1 of 1 keyframes kept
  60. Shot 59, 14:11 to 14:23, 1 of 1 keyframes kept
  61. Shot 60, 14:23 to 14:58, 1 of 1 keyframes kept
  62. Shot 61, 14:58 to 15:09, 1 of 1 keyframes kept
  63. Shot 62, 15:09 to 15:36, 0 of 1 keyframes kept
  64. Shot 63, 15:36 to 16:12, 0 of 1 keyframes kept
  65. Shot 64, 16:12 to 16:19, 1 of 1 keyframes kept
  66. Shot 65, 16:19 to 16:55, 1 of 1 keyframes kept
  67. Shot 66, 16:55 to 16:57, 1 of 1 keyframes kept
  68. Shot 67, 16:57 to 17:21, 0 of 1 keyframes kept
  69. Shot 68, 17:21 to 17:40, 1 of 1 keyframes kept
  70. Shot 69, 17:40 to 17:56, 0 of 1 keyframes kept

70 shot(s).

keyframes kept every frame deduplicated

What was stored

cues
201
whisperx 201
chunks
32
from 201 cues
keyframes
56
kept of 70 captured
frames with text
56
1,476 lines read
chapters
11
from the source metadata
keyframe bytes
9.1 MB
word timings on 201 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage state model started took
fetch done — 2026-08-09 22:35 0s
stt done — 2026-08-09 06:41 20s
chunk done — 2026-08-09 06:42 0s
text_embed done — 2026-08-10 19:39 1s
keyframe done — 2026-08-09 06:42 2m 38s
ocr done — 2026-08-09 06:44 32s
frame_embed done — 2026-08-10 19:39 10s

Frames, and what the machine read

  • 0:02 #0 done2 line(s)

    shot 0·sharpness 453.9

    1. AlEngineer0.96
    2. World's Fair0.97

  • 0:03 #1 done2 line(s)

    shot 1·sharpness 660.6

    1. AIEngineer0.95
    2. World's Fair0.99

  • 0:10 #2 done24 line(s)

    shot 2·sharpness 2731.3

    1. LAB & PLATINUM SPONSORS0.99
    2. Amazon AGI Lab0.98
    3. ANTHROP\C1.00
    4. Google DeepMind1.00
    5. MINIMAX0.93
    6. OpenAl0.92
    7. Akamai1.00
    8. arize1.00
    9. aws1.00
    10. Braintrust bright data0.99
    11. B1.00
    12. Browserbase1.00
    13. docker1.00
    14. :neo4j0.92
    15. ORACLE1.00
    16. PayPal1.00
    17. qodo1.00
    18. reducto1.00
    19. Sonar1.00
    20. Makers of0.99
    21. togetherai1.00
    22. Unblocked1.00
    23. WorkOS1.00
    24. SonarQube1.00

  • 0:16 #3 done49 line(s)

    shot 3·sharpness 3222.8

    1. ORACLE0.97
    2. arize1.00
    3. ;neo4j0.64
    4. Z.AI0.99
    5. extend1.00
    6. vast.ai0.99
    7. Ref.1.00
    8. RedHat0.99
    9. mezmo{*0.77
    10. stigg1.00
    11. RELAI0.95
    12. Modal1.00
    13. promptql0.94
    14. ZERO1.00
    15. AIEngineer0.96
    16. authe0.86
    17. OrOOOO0.59
    18. PRIOR1.00
    19. World's Fair1.00
    20. POSTHAN0.91
    21. Modular1.00
    22. MERGE1.00
    23. Zed1.00
    24. Keycard1.00
    25. Meticulous1.00
    26. Daytona0.97
    27. GRAVITEE1.00
    28. INNGEST0.93
    29. © BAND0.89
    30. Cleric0.99
    31. A ATLASSIAN0.97
    32. FACTORY1.00
    33. baseten1.00
    34. Z.AI0.99
    35. PayPal0.99
    36. Gradium1.00
    37. ANTHROPIC1.00
    38. ;neo4]0.83
    39. arize1.00
    40. reducto1.00
    41. WorkOS0.90
    42. OpenAl0.96
    43. Z.AI1.00
    44. ANTHROPIC1.00
    45. DATADOG0.96
    46. LanceDB1.00
    47. descpe0.94
    48. TOPK0.99
    49. eNCORD0.91

  • 0:20 #4 done63 line(s)

    shot 4·sharpness 3399.0

    1. World'sFair1.00
    2. Daytona1.00
    3. World'sFair1.00
    4. twilio1.00
    5. 80.90
    6. Temporal1.00
    7. World's Fair0.95
    8. Llamalndex1.00
    9. World's Fair0.97
    10. INNGEST1.00
    11. World'sFair1.00
    12. BAND1.00
    13. World'sFair1.00
    14. baseten1.00
    15. World's Fair0.96
    16. Snorkel1.00
    17. World's Fair0.98
    18. Z.AI0.99
    19. World's Fair0.97
    20. Amazon AGI Lab1.00
    21. World's Fair0.97
    22. Browserbase1.00
    23. World'sFair1.00
    24. :neo4j0.98
    25. World'sFair1.00
    26. Google DeepMind1.00
    27. World'sFair1.00
    28. World's Fair0.96
    29. WorkOS0.97
    30. World's Fair0.97
    31. Amazon AGI Lab1.00
    32. World's Fair0.97
    33. Microsoft1.00
    34. docker1.00
    35. World's Fair0.96
    36. Braintrust1.00
    37. World'sFair1.00
    38. OpenAl0.95
    39. World'sFair1.00
    40. World'sFair1.00
    41. aws1.00
    42. World'sFair1.00
    43. togetherai1.00
    44. World's Fair0.96
    45. Lightrun1.00
    46. ain1.00
    47. World'sFair1.00
    48. World's Fair0.98
    49. DATADOG1.00
    50. CLOUDFLARE1.00
    51. World'sFair1.00
    52. ir1.00
    53. LanceDB1.00
    54. World's Fair0.96
    55. builder.io0.96
    56. World'sFair1.00
    57. World's Fair0.99
    58. descope0.99
    59. PRESENTED BY0.98
    60. World'sFair1.00
    61. TOPK1.00
    62. cognee1.00
    63. Microsoft1.00

  • 0:22 #5 done47 line(s)

    shot 5·sharpness 2215.2

    1. AlEngineer1.00
    2. AlEngineer0.99
    3. Amazon AGI Lab1.00
    4. World's Fair0.99
    5. B1.00
    6. Browserbase1.00
    7. World's Fair0.99
    8. :ne0.90
    9. AlEngineer1.00
    10. AlEngineer0.99
    11. AlEngi0.99
    12. World's Fair0.98
    13. Op0.94
    14. World's Fair0.99
    15. WorkOS0.99
    16. World1.00
    17. Al'0.74
    18. AlEngineer1.00
    19. Microsoft1.00
    20. docker1.00
    21. World's Fair0.97
    22. Brai1.00
    23. World's Fair0.97
    24. AlEngineer0.99
    25. World's Fair0.97
    26. AlEngineer1.00
    27. aws1.00
    28. World1.00
    29. AlEngi0.97
    30. AlEngineer0.99
    31. Lightrun1.00
    32. ain1.00
    33. World'sFair1.00
    34. CLOUDF0.96
    35. AlEngineer0.97
    36. AlEngi0.95
    37. World'sFair1.00
    38. ir1.00
    39. LanceDB1.00
    40. World1.00
    41. Al Engineer0.95
    42. World's Fair0.99
    43. descope0.90
    44. PRESENTED BY1.00
    45. World's Fair0.96
    46. TO0.98
    47. Microsoft1.00

  • 0:44 #6 done12 line(s)

    shot 6·sharpness 1227.5

    1. AlEngineer0.98
    2. World'sFair0.99
    3. We're all building loops wrong0.99
    4. Open/1.00
    5. Wor1.00
    6. AlEngineer0.94
    7. World'sFair0.96
    8. AlEnginee-0.90
    9. Loop Engineering from First Principles | Kyle Mistele | @0xblacklight0.98
    10. Worl0.92
    11. SI0.71
    12. Wor1.00

  • 1:18 #7 done22 line(s)

    shot 7·sharpness 2680.4

    1. AlEngineer0.98
    2. World'sFair1.00
    3. BY GEOFFREY HUNTLEY IN AI-14 JUL 20250.94
    4. Ralph Wiggum as a "software engineer"0.98
    5. X0.82
    6. done1.00
    7. Prompt1.00
    8. Agent1.00
    9. -while (true)0.99
    10. OpenAI0.93
    11. Worl1.00
    12. All1.00
    13. How Ralph Wiggum went from 'The Simpsons' to the biggest name in Al right now -0.99
    14. Venture Beat1.00
    15. AlEngineer1.00
    16. AlEngineer0.99
    17. Loop Engineering from First Principles | Kyle Mistele | @0xblacklight0.98
    18. World'sFair1.00
    19. World's1.00
    20. All1.00
    21. sn1.00
    22. Norl1.00

  • 1:47 #8 done16 line(s)

    shot 8·sharpness 1284.1

    1. AlEngineer0.97
    2. World'sFair1.00
    3. Let's talk about how to build loops0.99
    4. AlEngineer0.98
    5. penAI0.92
    6. Norld'sF1.00
    7. AlEngineer0.99
    8. World'sFair1.00
    9. rld'sFai0.99
    10. AlEngineer0.99
    11. dock1.00
    12. Loop Engineering from First Principles | Kyle Mistele @0xblacklight0.98
    13. snyk0.94
    14. AlEngineer -0.95
    15. 1d'sF0.91
    16. Ch0.99

  • 2:01 #9 skipped

    shot 9·duplicate of #7

  • 2:10 #10 done19 line(s)

    shot 10·sharpness 2230.8

    1. AlEngineer0.96
    2. World'sFair1.00
    3. Peter Steinberger1.00
    4. @steipete1.00
    5. Here's your monthly reminder that you shouldn't be prompting coding0.98
    6. agents anymore.1.00
    7. You should be designing loops that prompt your agents.1.00
    8. 11:58 AM · Jun 7, 2026 ·8.3M Views0.94
    9. AlEngine0.99
    10. OpenAI0.93
    11. Norld's1.00
    12. AlEngineer0.99
    13. World'sFair1.00
    14. lorld'sFair0.98
    15. — AlEngineer0.95
    16. `.doc0.77
    17. Loop Engineering from First Principles | Kyle Mistele | @0xblacklight0.99
    18. snyk1.00
    19. 1d's0.95

  • 2:22 #11 done28 line(s)

    shot 11·sharpness 2832.7

    1. AlEngineer0.99
    2. World's Fair0.97
    3. Peter Steinberger@steipeteApr280.98
    4. codex now runs on each commit we land, reviews it - and if a booboo is0.98
    5. found, a new codex spins up and (if still relevant) makes a PR for the fix.0.99
    6. Then a review agent spins up. If an issue is found, another agent will fix the0.99
    7. issues. (up to 5 loops)1.00
    8. openclaw/openclaw1.00
    9. #74065 fix: Found one0.97
    10. low-severity block-reply1.00
    11. edge case in the new...1.00
    12. 4 comments 0reviews 5files +108-80.95
    13. openclaw-clownfish· April 29, 2026 -0-1 commit0.95
    14. OpenAl0.93
    15. World'0.95
    16. AlEngin0.93
    17. fix: Found one low-severity block-reply edge case in the new split-tag by open...0.99
    18. From github.com1.00
    19. AlEngineer1.00
    20. Norld'sF0.97
    21. -AlEngineer0.96
    22. do1.00
    23. Loop Engineering from First Principles | Kyle Mistele |0.99
    24. @0xblacklight0.99
    25. World's Fair0.99
    26. AlEngin0.99
    27. snyl1.00
    28. Wor0.99

  • 2:38 #12 done15 line(s)

    shot 12·sharpness 1228.8

    1. AlEngineer0.98
    2. World'sFair1.00
    3. "I don't prompt Claude anymore.1.00
    4. I write loops - and the loops do the work.0.99
    5. OpenAl0.93
    6. World's1.00
    7. AlEnginee0.99
    8. My job is to write loops."0.99
    9. Boris Cherny, Creator of Claude Code1.00
    10. — AlEngineer —0.92
    11. 'orld's'0.97
    12. doc1.00
    13. AlEnginee0.99
    14. sny1.00
    15. Vorld's1.00

  • 2:49 #13 done24 line(s)

    shot 13·sharpness 2185.3

    1. AlEngineer0.99
    2. World'sFair1.00
    3. Peter Steinberger0.99
    4. @steipete·Jun71.00
    5. Replying to @gauthampai0.97
    6. Don't worry it'll take 3 months until it's there.0.99
    7. We'll be talking about fleets that design your loops then.1.00
    8. 301.00
    9. 1.2K1.00
    10. 山252K0.90
    11. 口0.77
    12. ngineer1.00
    13. d'sFair1.00
    14. azon AGl0.94
    15. AlEngineer0.99
    16. World'sFair1.00
    17. aintrust0.99
    18. AlEngineer0.99
    19. d'sFa0.90
    20. Loop Engineering from First Principles | Kyle Mistele | @0xblacklight0.99
    21. gether1.00
    22. Engineering the future of Al0.98
    23. Engineer1.00
    24. ...rd's Fai0.83

  • 3:08 #14 done20 line(s)

    shot 14·sharpness 1525.3

    1. AlEngineer0.96
    2. World'sFair1.00
    3. DevGuild0.97
    4. WRITE-ONLY CODE SUMMIT0.98
    5. WRITE-ONLY CODE SUMMIT1.00
    6. AI code generation, multi-agent workflows, and shifting release practices have ushered in an era of write-only code, one where0.99
    7. software outpaces human attention. But for those with real customers, velocity cuts both ways. Enterprises still expect a high0.99
    8. Id'sFair0.99
    9. Engineer1.00
    10. OpenAI0.92
    11. bar for quality, performance, and trust. As engineering decision-makers, our work has never been more valuable.0.99
    12. DevGuild gathers technical founders and infrastructure leaders redefining the software development lifecycle. If you're building0.99
    13. licroso0.90
    14. 'orld'sFai0.98
    15. the culture, processes, and tools for a write-only future, join us.1.00
    16. Engineer1.00
    17. Id's0.95
    18. snyk1.00
    19. Engineering the future of Al0.98
    20. Lightrt0.92

  • 3:21 #15 done17 line(s)

    shot 15·sharpness 1939.9

    1. AlEngineer0.97
    2. World'sFair1.00
    3. Build loops0.99
    4. Spend as many tokens as possible0.97
    5. Don't read the code1.00
    6. OpenAI0.90
    7. World'sl0.92
    8. AlEngineer0.99
    9. AlEngineer1.00
    10. - AlEngineer0.90
    11. :docl0.86
    12. Loop Engineering from First Principles Kyle Mistele | @0xblacklight0.98
    13. World'sFair1.00
    14. orld'sFr'0.95
    15. snyk0.96
    16. orld'sl0.99
    17. Engineering the future of Al1.00

  • 3:30 #16 done16 line(s)

    shot 16·sharpness 1529.3

    1. AlEngineer0.99
    2. World'sFair1.00
    3. PRESENTED BY0.97
    4. Microsoft0.96
    5. Fair0.94
    6. penAl0.94
    7. AlEngineer0.98
    8. World'sFair1.00
    9. osoft0.99
    10. AlEngineer0.99
    11. 'd'sFair0.98
    12. Loop Engineering from First Principles | Kyle Mistele |0.99
    13. @0xblacklight0.96
    14. Fair1.00
    15. Engineering the future of Al1.00
    16. truⁿ1.00

  • 3:33 #17 done13 line(s)

    shot 17·sharpness 1378.0

    1. AlEngineer0.99
    2. World'sFair1.00
    3. PRESENTED BY0.97
    4. Microsoft0.96
    5. Oper1.00
    6. Wc0.88
    7. AlEngineer1.00
    8. World'sFair1.00
    9. AIlEn0.86
    10. Loop Engineering from First Principles | Kyle Mistele | @0xblacklight0.99
    11. Wc0.81
    12. Wc0.89
    13. Engineering the future of Al0.99

  • 3:38 #18 done40 line(s)

    shot 18·sharpness 1735.2

    1. AlEngineer1.00
    2. World'sFair1.00
    3. openclaw1.00
    4. Source: openclaw/openclaw·Install Confidence 0-100.99
    5. PRESENTED BY0.97
    6. GLOBAL SCORE HISTORY0.98
    7. Microsoft1.00
    8. 10.01.00
    9. 10.01.00
    10. 9.81.00
    11. 101.00
    12. 7.61.00
    13. 8.81.00
    14. 7.81.00
    15. 8.31.00
    16. 6.41.00
    17. 5.61.00
    18. 5.51.00
    19. 4.61.00
    20. 4.41.00
    21. 4.41.00
    22. 4.21.00
    23. 3.41.00
    24. 3.41.00
    25. 2.31.00
    26. 1.51.00
    27. OpenAI0.96
    28. World's1.00
    29. AlEngin1.00
    30. AlEngineer0.98
    31. World's Fair1.00
    32. Vorld's1.00
    33. —AlEngineer —0.94
    34. dor0.84
    35. Loop Engineering from First Principles | Kyle Mistele1.00
    36. @0xblacklight0.97
    37. - AlEngin0.93
    38. sn'0.92
    39. Vorld's1.00
    40. Engineering the future of Al0.99

  • 3:44 #19 done20 line(s)

    shot 19·sharpness 2187.8

    1. AlEngineer0.99
    2. World's Fair0.98
    3. $1,305,088.810.99
    4. PRESENTED BY0.99
    5. Microsoft1.00
    6. Peter Steinberger's Monthly Token Spend1.00
    7. Engineer1.00
    8. d's Fair0.96
    9. OpenAl0.96
    10. AlEngineer1.00
    11. licrosof0.87
    12. rld'sFai1.00
    13. AlEngineer0.98
    14. Loop Engineering from First Principles | Kyle Mistele | @0xblacklight0.98
    15. World's Fair0.98
    16. d's Fair0.96
    17. Engineer1.00
    18. nyk1.00
    19. Engineering the future of Al0.98
    20. Lightruⁿ0.96

  • 4:00 #20 done32 line(s)

    shot 20·sharpness 2106.7

    1. EUROPE1.00
    2. AlEngineer0.97
    3. World'sFair1.00
    4. WorkOS0.96
    5. eo40.91
    6. VEngineer0.96
    7. PRESENTED BY1.00
    8. EUROPE1.00
    9. Bad code is the most expensive0.98
    10. Microsoft1.00
    11. Draintrus1.00
    12. it's ever been0.97
    13. ailscal1.00
    14. AlEngine0.93
    15. OpenAI0.94
    16. Worl1.00
    17. All1.00
    18. AlEngineer0.98
    19. ATTER MORE THAN EVER0.99
    20. IT AIN'T BROKE: W0.99
    21. EUROPE1.00
    22. PRESENTED BY1.00
    23. World'sF0.94
    24. AlEngineer1.00
    25. Google DeepMind1.00
    26. Braintrust1.00
    27. WorkOS1.00
    28. OpenAl0.94
    29. snyl0.99
    30. rl0.82
    31. All1.00
    32. Engineering the future of Al1.00

  • 4:07 #21 done17 line(s)

    shot 21·sharpness 1586.7

    1. AlEngineer0.98
    2. World'sFair1.00
    3. PRESENTED BY1.00
    4. Microsoft1.00
    5. What Actually Works?1.00
    6. OpenA'1.00
    7. Worl1.00
    8. Al0.97
    9. AlEngineer1.00
    10. World'sFair1.00
    11. AlEngineer —0.93
    12. Loop Engineering from First Principles | Kyle Mistele | @0xblacklight0.98
    13. World'1.00
    14. All0.97
    15. sn1.00
    16. Worl1.00
    17. Engineering the future of Al1.00

  • 4:27 #22 done22 line(s)

    shot 22·sharpness 2641.6

    1. AlEngineer0.98
    2. World's Fair0.98
    3. Build Loops1.00
    4. Read the Code1.00
    5. PRESENTED BY0.97
    6. Microsoft1.00
    7. Preserve Human Ownership0.97
    8. Maintain + Improve Code Quality0.99
    9. Solve Hard Problems in Complex Codebases1.00
    10. OpenAl0.92
    11. World's1.00
    12. AlEngin0.99
    13. AlEngineer1.00
    14. AlEngineer1.00
    15. Loop Engineering from First Principles Kyle Mistele | @0xblacklight0.98
    16. World's Fair0.98
    17. Vorld's'0.99
    18. do1.00
    19. — AlEngin0.96
    20. sny0.94
    21. orld's0.98
    22. Engineering the future of Al0.97

  • 4:36 #23 done41 line(s)

    shot 23·sharpness 4106.5

    1. AlEngineer0.99
    2. World'sFair1.00
    3. Control Theory1.00
    4. Measured1.00
    5. Controller1.00
    6. System1.00
    7. Disturbance1.00
    8. Error1.00
    9. Output1.00
    10. Input1.00
    11. d(t)1.00
    12. System1.00
    13. PRESENTED BY0.98
    14. Set Point1.00
    15. e(t)1.00
    16. u(t)1.00
    17. us(t)0.96
    18. Output0.99
    19. Microsoft1.00
    20. Controller1.00
    21. Actuator1.00
    22. System1.00
    23. w(t)1.00
    24. y(t)1.00
    25. Measured Output0.97
    26. Sensor1.00
    27. ym(t)1.00
    28. orld's Fair0.98
    29. -AlEngineer0.97
    30. Open/0.90
    31. AlEngineer1.00
    32. World'sFair1.00
    33. Micros1.00
    34. World'sI0.96
    35. AlEngineer0.99
    36. Loop Engineering from First Principles | Kyle Mistele | @0xblacklight0.98
    37. orld's1.00
    38. — AlEngineer0.94
    39. snyh0.94
    40. Engineering the future of Al0.98
    41. Light1.00

Transcript

201 cues· 3,401 words· 18,245 chars

  1. 0:27 Hey everybody, my name's Kyle.
  2. 0:29 I'm co-founder of a company called Human Layer, and I'm here to talk about loops.
  3. 0:34 I think we've all been building loops lately, and I realized recently, I think we're all doing it wrong.
  4. 0:41 Loops are really powerful, don't get me wrong, but so much of the discourse around them is hype-driven and just really not helpful.
  5. 0:48 I think we have this idea, just kind of as an industry somehow, that we can pipe a prompt and a loop to a coding agent and that we can build software this way.
  6. 0:59 Maybe we're investing a lot of time in verifiers.
  7. 1:02 Maybe you have six different code review agents.
  8. 1:05 But at the end of the day, if we're doing this, we're still building 40,000 line PRs that just nobody wants to read, right?
  9. 1:12 And this isn't to throw shade to Jeff Huntley, right?
  10. 1:15 This is...
  11. 1:16 Ralph is an innovative, it's a sharp tool that works very well for certain types of problems.
  12. 1:22 It works very well if you're not building on a team.
  13. 1:25 And it works very well if you're not working on critical systems.
  14. 1:28 But most of us are working on teams and we don't fit in that box.
  15. 1:33 So today I want to talk about how to build loops that work in large complex code bases for systems that have real customers, real users, real regulatory obligations, and service level agreements, and everything else that keeps us from shipping YOLO 40,000 line PRs straight to production.
  16. 1:51 In other words, I want to talk about how to build loops for the real world.
  17. 1:55 If you're not aware, this post actually dates back to July.
  18. 1:58 It went viral this past January, which is when a lot of us, I think, started building loops.
  19. 2:03 And of course, much more recently, I'm sure y'all are gonna see this slide a lot this week, but Peter Steinberger said that we shouldn't be prompting coding agents anymore, right?
  20. 2:11 We should just be designing loops that prompt our agents.
  21. 2:14 Of course, OpenClaw is notoriously built on loops.
  22. 2:18 Loops build the code, loops review the code, they merge and release the code, they find and fix the bugs.
  23. 2:23 There's even loops for finding and fixing bugs in the loops that are merging the things, right?
  24. 2:29 It's loops all the way down.
  25. 2:32 It's Boris Turney, also the creator of Claude Code, recently said that this is his entire job as an engineer now, is just writing loops to prompt Claude.
  26. 2:41 And eventually, we might not even need loops, right?
  27. 2:44 We're just going to have swarms of agents designing loops to prompt agents building swarms for loops.
  28. 2:51 And I don't know, somewhere we're writing production code, I assume.
  29. 2:56 And in fact, all of our loops that we're building are producing so much code that we can't possibly read all of it, right?
  30. 3:03 So we might as well just not read any of it, right?
  31. 3:06 We're investing in verification and in code review, but all this code is read-only.
  32. 3:10 This is the thesis of a conference that was here in town last month.
  33. 3:14 So a lot of smart people at the Frontier Labs think that this is the future of software development, and if you're doing this, you're moving 10x faster, and everybody else is getting left behind.
  34. 3:26 Now, it's not clear how well this works yet.
  35. 3:29 It took six months to fix the Cloud Code terminal flicker.
  36. 3:32 The OpenCode team wrote a renderer in a fraction of that time.
  37. 3:36 And OpenClaw, of course, also notoriously has stability issues.
  38. 3:41 What is abundantly clear, however, is that this shit is really expensive if you don't work at a frontier lab and have an unlimited token budget.
  39. 3:50 And all this code that we're writing is actually really expensive, right?
  40. 3:54 Matt Pocock talked about this recently.
  41. 3:57 Bad code is much more expensive in the age of agents than it has ever been at any point in the past.
  42. 4:03 So today I want to talk about what I think works in the real world and what we've started doing at Human Layer, which, to be clear, is still building loops, right?
  43. 4:12 I think loops are super powerful, but we can design loops and still read the code.
  44. 4:17 In fact, we can design loops that make it easier to read the code because the loops are making the code better.
  45. 4:23 We can solve hard problems in complex code bases with loops, and we can build our software for that at three incrementally.
  46. 4:29 But to do this is gonna take some real engineering, y'all.
  47. 4:33 So let's talk about control theory.
  48. 4:35 Control theory is all about how we drive a dynamic system, which would be your code base, towards some desired stable or optimal end state.
  49. 4:43 You have a sensor that measures the current state of the world.
  50. 4:47 You have your set point, the desired state of the world, and the difference between those two things is your measured error.

Chapters

  1. 0:00 Introduction: the 40,000 line PR problem
  2. 1:55 Why more code is not the goal
  3. 3:37 Is the generated code any good?
  4. 4:43 Control loops from control theory
  5. 5:49 Infrastructure as code and Ralph loops
  6. 7:06 Applying control loops to coding
  7. 8:35 Migrating a codebase one procedure at a time
  8. 10:40 Tracking the loop in version control
  9. 12:35 The actuator agent and golden patterns
  10. 13:51 Wiring the loop into CI
  11. 15:44 Avoiding stacked PRs and scaling the controller

Open at this second