Videos 1P1hJ36rxM0

"Software engineering is not about writing code" — Benoit Schillings, Google DeepMind VP of Research

index_state ready data_status ok

AI Engineer· published 2026-07-17· 0:20:26· en· indexed 2026-08-10 22:27

Open on YouTube

51 shot(s).

keyframes kept every frame deduplicated

What was stored

cues: 188; whisperx 188
chunks: 36; from 188 cues
keyframes: 27; kept of 51 captured
frames with text: 27; 793 lines read
chapters: 11; from the source metadata
keyframe bytes: 7.8 MB; word timings on 188 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage	state	model	started	took
`fetch`	done	—	2026-08-10 22:23	1m 15s
`stt`	done	—	2026-08-10 22:24	24s
`chunk`	done	—	2026-08-10 22:25	0s
`text_embed`	done	—	2026-08-10 22:25	1s
`keyframe`	done	—	2026-08-10 22:25	2m 16s
`ocr`	done	—	2026-08-10 22:27	13s
`frame_embed`	done	—	2026-08-10 22:27	5s

Frames, and what the machine read

0:02 #0 done2 line(s)

shot 0·sharpness 456.7
1. AlEngineer0.96
2. World's Fair0.97
0:03 #1 done2 line(s)

shot 1·sharpness 663.3
1. AlEngineer0.95
2. World's Fair0.99
0:10 #2 done25 line(s)

shot 2·sharpness 2742.9
1. LAB & PLATINUM SPONSORS0.99
2. Amazon AGI Lab0.97
3. ANTHROP\C1.00
4. Google DeepMind1.00
5. MINIMAX0.98
6. OpenAl0.89
7. Akamai1.00
8. arize1.00
9. aws1.00
10. Braintrust1.00
11. bright data0.99
12. B1.00
13. Browserbase1.00
14. docker1.00
15. :neo4j0.90
16. ORACLE1.00
17. PayPal1.00
18. qodo1.00
19. reducto1.00
20. Sonar1.00
21. Makers of1.00
22. togetherai1.00
23. Unblocked1.00
24. WorkOS1.00
25. SonarQube1.00
0:16 #3 done2 line(s)

shot 3·sharpness 238.7
1. Google DeepMind1.00
2. AIE1.00
0:26 #4 done4 line(s)

shot 4·sharpness 726.5
1. BENOITSCHILLINGS1.00
2. VICE PRESIDENT OF RESEARCH0.97
3. Google DeepMind0.98
4. AIF1.00
0:28 #5 done6 line(s)

shot 5·sharpness 1554.2
1. ENOITSCHICLINGS0.94
2. 70.93
3. VICE PRESIDENT OF RESEARCH0.99
4. Google DeepMind1.00
5. AIE1.00
6. 90000000.64
0:34 #6 done98 line(s)

shot 6·sharpness 3577.7
1. MERGE1.00
2. World'sFair1.00
3. AlEngineer0.97
4. promptql0.98
5. RedHat1.00
6. Z.AI0.94
7. orld'sFa0.97
8. World's Fal0.93
9. orld's Fair0.97
10. THE VELOCITY ROOM0.97
11. mezmo+0.92
12. bright data0.82
13. ©fiddler0.95
14. igg0.51
15. comet1.00
16. rld's Fair0.97
17. Cate0.61
18. Snorkel0.97
19. INNGEST1.00
20. World's Fair0.97
21. PUSTHAN0.89
22. PRIOR1.00
23. SRLOK0.70
24. )pen0.93
25. Id's Fair0.89
26. Worid's Fair0.94
27. Z.AI0.95
28. © BAND0.91
29. World's Fair0.93
30. Keycard1.00
31. AUTOMATTIC0.99
32. VorkOs0.92
33. World's Fair0.97
34. :neo4j0.97
35. Amazon AGI Lab1.00
36. World's Fair0.95
37. Google DeepMind0.97
38. World's Fai0.94
39. World'sFa0.95
40. qodo0.93
41. World's Fa0.96
42. arize1.00
43. Cleric0.99
44. World's Fai0.93
45. À ATLASSIAN0.92
46. GRAVITEE0.96
47. Gradium0.98
48. Meliculous0.99
49. FACTORY1.00
50. Jdrid0.57
51. Qorae0.58
52. d's Fair0.95
53. Braintrust1.00
54. World's Fai0.98
55. Microsoft0.96
56. Worla0.99
57. reducto1.00
58. ANTHROPIC0.98
59. orld0.95
60. VS0.71
61. World's Fair0.98
62. World'sFair0.95
63. OpenAI0.92
64. World'sFair0.99
65. MINIMAX1.00
66. bright data0.93
67. Microsoft0.94
68. Googe DepMind0.81
69. Crusce0.82
70. togetherai1.00
71. World'sFai0.98
72. Z.AI0.99
73. ANTHROPIC1.00
74. sFair1.00
75. World's Fair0.97
76. DATADOG1.00
77. World's Fair0.96
78. World's Fai0.95
79. Resolve.ai1.00
80. @Arbyte0.85
81. greptle0.88
82. ceDB1.00
83. World'sFair1.00
84. builder.io0.96
85. World'sFair0.99
86. Ravenna1.00
87. Copi1.00
88. World's Fai0.90
89. CODERI0.94
90. Optiver0.99
91. Fair1.00
92. World's Fair0.97
93. cognee1.00
94. World'sFair0.99
95. BeNCORD0.91
96. Worlc0.83
97. Workds Fair0.88
98. rgus0.67
0:54 #7 done15 line(s)

shot 7·sharpness 487.5
1. Lab1.00
2. World's Fa0.99
3. Microsoft1.00
4. AlEngineer0.98
5. OpenAl0.91
6. ir0.82
7. World's Fair1.00
8. AlEngi0.93
9. ai1.00
10. World1.00
11. Akamai1.00
12. AlEngineer0.97
13. DAT1.00
14. ir0.99
15. World's Fair0.97
1:21 #8 done15 line(s)

shot 8·sharpness 533.4
1. Lab1.00
2. World's Fair1.00
3. Microsoft1.00
4. AlEngineer0.99
5. OpenAl0.93
6. r1.00
7. World's Fair1.00
8. AlEnginee0.96
9. ai1.00
10. World's1.00
11. Akamai1.00
12. AlEngineer0.98
13. DATAD1.00
14. World's Fair1.00
15. r0.96
1:45 #9 done20 line(s)

shot 9·sharpness 2461.6
1. AlEngineer0.98
2. World's Fair0.97
3. From Assembly to Vibe Coding0.98
4. A technical roadmap of constraints, ML capabilities, and the future of engineering intent.0.99
5. Benoit Schillings0.97
6. Google DeepMind1.00
7. orld's Fair0.99
8. Mi1.00
9. OpenAl0.91
10. Worlc0.85
11. AlEn0.87
12. - AlEngineer0.93
13. orld's1.00
14. Ak0.75
15. Research to Reality with Google DeepMind0.99
16. AlEn0.87
17. Benoit Schillings/ VicePresident of Research0.99
18. Google DeepMind1.00
19. DATADO0.99
20. Norle0.99
2:30 #10 skipped

shot 10·duplicate of #9
2:41 #11 done28 line(s)

shot 11·sharpness 2935.2
1. AlEngineer1.00
2. |Origin story0.99
3. World'sFair1.00
4. Resisting Technical Shift1.00
5. Google X & Pitchfork Roots0.99
6. It is natural to resist progress. From writing 680000.98
7. Project Pitchfork was initially designed to optimize0.99
8. assembly and eyeing early compilers with strong0.99
9. codebase evolutions by matching review edits.1.00
10. suspicion, to dismissing GC/interpreted languages, the1.00
11. cycle of skepticism is predictable.0.97
12. We never anticipated LLMs would scale this fast. The1.00
13. Today, building with modern high-level abstraction1.00
14. once impossible concept of vibe coding has become0.99
15. loops turns even the most cynical purist into an avid ml0.98
16. an daily workflow reality.1.00
17. coder.1.00
18. Vorld's Fai0.97
19. N0.97
20. OpenAI0.90
21. Wor1.00
22. -AlEngine0.95
23. Vorld's1.00
24. Research to Reality with Google DeepMind0.99
25. Benoit Schillings/ VicePresident of Research0.99
26. Google DeepMind1.00
27. DAT0.99
28. Wlor0.68
3:21 #12 skipped

shot 12·duplicate of #11
3:39 #13 skipped

shot 13·duplicate of #11
3:55 #14 skipped

shot 14·duplicate of #11
4:37 #15 skipped

shot 15·duplicate of #11
5:03 #16 done29 line(s)

shot 16·sharpness 2648.1
1. AlEngineer0.99
2. |The Software Eras1.00
3. World's Fair0.96
4. Phase1.00
5. The Core Bottleneck1.00
6. Human Role0.96
7. The Assembly/C++ Era1.00
8. Silicon & Memory (Hardware constraints)0.99
9. Manual optimization, extreme precision1.00
10. The Modern/Cloud Era0.99
11. Human Cognitive Load (Context stack0.99
12. Designing for modularity and maintainability0.98
13. constraints)1.00
14. The Frontier Al Era1.00
15. Verifiability & Intent (Correctness constraints)0.98
16. Inductive thinking, architecture, security1.00
17. guardrails1.00
18. d'sFair1.00
19. Micr1.00
20. AlEngine0.94
21. enAl0.99
22. World's1.00
23. d'sF1.00
24. Engineer0.97
25. Akal0.90
26. Research to Reality with Google DeepMind1.00
27. Benoit Schillings / Vice President of Research0.98
28. Google DeepMind0.99
29. AlEngine0.95
5:36 #17 skipped

shot 17·duplicate of #16
5:51 #18 skipped

shot 18·duplicate of #16
6:32 #19 skipped

shot 19·duplicate of #16
7:04 #20 skipped

shot 20·duplicate of #16
7:24 #21 done26 line(s)

shot 21·sharpness 2750.4
1. AlEngineer0.96
2. |State of Al Software Engineering0.97
3. World's Fair0.98
4. PRESENTED BY0.99
5. Super-human Syntax Generation1.00
6. 95%1.00
7. Microsoft1.00
8. Local Problem Solving & Tasks1.00
9. 70%1.00
10. Multi-step Codebase Planning1.00
11. 45%1.00
12. Architectural System Decisions1.00
13. 25%1.00
14. Models generate localized code rapidly but struggle to form cohesive systems, design complex architectures, and predict overall0.99
15. security flaws.1.00
16. Id's Fair0.92
17. Amazon1.00
18. AlEng0.99
19. enAl0.99
20. World1.00
21. Id'sF0.99
22. Engineer1.00
23. Mic1.00
24. Research to Reality with Google DeepMind0.99
25. Benoit Schillings / VicePresident of Research0.99
26. Google DeepMind0.98
7:53 #22 skipped

shot 22·duplicate of #21
8:17 #23 skipped

shot 23·duplicate of #21

Transcript

188 cues· 3,099 words· 16,502 chars

chunk 0 · 0:13–0:56 · 42 words · 237 chars
0:13 Please welcome to the stage the Vice President of Research at Google DeepMind, Benoit Schillings.
chunk 1 · 0:49–1:31 · 122 words · 599 chars
0:49 All right, good morning.
0:50 This is really quite exciting to be here and have a chance to speak with all of you.
0:55 My name is Benoit Schillings.
0:57 I'm actually a bit of a noob when it comes to machine learning.
1:02 Till a year and a half ago, I was working for Google X, which some of you may know.
1:08 We've done things like Waymo, which seems to be at every street corner now.
1:13 We also do things like glass so you know we had a mix of hit and success but in many ways this was for me an interesting formative experience on how to run a research team in a place like DeepMind.
chunk 2 · 1:27–2:10 · 114 words · 602 chars
1:27 I do have an incredible team.
1:30 My team
1:32 goal in DeepMind is basically to develop whatever technology will be needed to make Gemini incredible between one month and one year from now.
1:42 So one month because if you start to work on what is needed in one week, that's a very different type of job.
1:49 And one year because I don't think anybody can really predict anything that far.
1:54 So that's already pretty ambitious, in my opinion, to think about things that would happen one year in the future.
chunk 3 · 2:02–2:36 · 89 words · 465 chars
2:02 We do many things under that role.
2:06 A lot of it is related to code, which will be the main subject of my talk today.
2:11 But we also do a lot of research on what is the evolution of reasoning for models, for instance.
2:16 Or we do topology research, what are new type of network that might bring better performance.
2:24 We do fundamental work in the science of reinforcement learning, which is so
2:29 fundamental to what we're doing today with ML.
chunk 4 · 2:35–3:14 · 87 words · 453 chars
2:35 Let's do a bit of an origin story.
2:41 We started the project at X named Pitchfork in 2018, which was aimed at looking at how ML could really improve the way code is being written.
2:54 And this was very interesting because in 2018 when we presented that at Google, honestly nobody would give us the time of day.
3:03 There was that point like, why would you ever need ML to write code?
chunk 5 · 3:09–3:49 · 116 words · 643 chars
3:09 At the same time, I think that we totally underestimated how fast this could go.
3:14 When we did that project originally, the idea was to look at how we could speed up the evolution of a piece of code.
3:22 How could we make many of those small changes which slows down code speed development?
3:27 You know, the small edit which requires a review that takes three days and how we could compress that cycle.
3:35 Some people were talking about vibe coding, writing code in English, and at the time, honestly, I totally dismissed that.
chunk 6 · 3:42–4:20 · 94 words · 527 chars
3:42 That's why we have programming language.
3:43 English is not a programming language.
3:45 Well, I guess I was pretty wrong on that front.
3:50 But the resistance we felt at the time reminded me of how my own career was pretty resistive to change.
3:58 I've been writing code for
4:02 45 years, I started by writing video game for Apple II and Commodore 64.
4:07 So my formation was to write assembly language.
chunk 7 · 4:12–4:55 · 112 words · 622 chars
4:12 And when you spend a long time writing assembly language, you look at compilers with a lot of suspicion.
4:18 Are those things really working correctly?
4:21 And then when you switch to C++ and use compiler, you look at garbage collected languages as this.
4:28 That's not real programming.
4:29 You need to manage your memory.
4:31 Well, today I use Python and vibe coding, so even old dogs can learn new tricks.
4:38 But I do understand what happened there.
chunk 8 · 4:43–5:23 · 109 words · 594 chars
4:43 I think that we have a number of eras in what happened with software.
4:47 And the first one was, you know, the one where I started writing code, where the fundamental limit was really the machine.
4:55 And there was a lot of work to go and extract the last ounce of power out of those machines.
5:02 And that was the days of assembly language where you really needed to be incredibly accurate in the way you were writing code.
5:11 Computing became much cheaper and we switched to the modern cloud era where getting the best performance is not the most critical aspect.
chunk 9 · 5:20–6:02 · 110 words · 615 chars
5:20 You can actually brute force many problems.
5:24 But really what became the limiting factor was the ability for us to design in a modular way.

Chapters

0:00 Introduction and speaker background
2:35 The origin story of the Pitchfork project
4:43 Historical eras of software development
7:08 The current state of AI code generation
9:36 The role of self-play in training models
11:13 Changing economics of software engineering
12:41 Implementing guardrails and security
13:48 Inductive architecture and model planning
14:36 Evolution of evaluation benchmarks
15:45 Moving beyond simple chain-of-thought tokens
17:51 Future applications in chemistry and biology