Anthropic's Boris Cherny: Why Coding Is Solved, and What Comes Next

À retenir

1. Date / source. interview vidéo Sequoia Capital (YouTube), URL https://www.youtube.com/watch?v=SlGRN8jh2RI. Date précise non visible dans le transcript ; références internes (Opus 4.7 comme modèle courant, "we also just launched routines") la situent fin avril / début mai 2026. Date d'ajout dossier veille : 2026-05-05.
2. Speaker. Boris Cherny, créateur Claude Code, Anthropic. Interviewer : Lauren Reader (Sequoia). Intro : Asia (Sequoia).
3. Background Cherny révélé.
4. YC alumni, premier employé d'une boîte YC, plusieurs startups passées.
5. Auteur d'un livre "Programming TypeScript".
6. N'a pas écrit une seule ligne de code en 2026. (déclaration faite à Sequoia avant l'interview).
7. Anecdote middle school : a écrit un guide BASIC pour TI-83 Plus, "still on the internet, extremely embarrassing".
8. Ouverture comique.
9. "Show of hands, who here uses Claude Code?" → majorité.
10. "Show of hands, who here has Claude code psychosis?" → rires.
11. "My team lovingly says I have Claude code psychosis, which may or may not be true." ### Histoire de Claude Code (récit nouveau)
12. Origine. Cherny rejoint Anthropic Labs fin 2024 — incubator interne Anthropic, équipe de quelques personnes.
13. Output Anthropic Labs. Claude Code, MCP, application desktop. Équipe disbanded une fois la mission accomplie.
14. Round 2 actuel. équipe reformée, dirigée par Mike Krieger (Chief Product Officer Anthropic, ex-cofondateur Instagram).
15. Pourquoi Cherny a démarré coding. "product overhang" — "the model can do all the stuff that no product has yet captured."
16. État de l'art coding fin 2024. "type ahead" (open IDE, press tab, complete one line). C'est ce que Sonnet 3.5 avait débloqué.
17. Pari. aller plus loin — "the model was almost ready for the next big step. So we don't have to do type ahead anymore, we can just have the agent write all of the code."
18. 6 premiers mois. "it just really didn't work. It was like not very good. It was barely usable. I used it for maybe 10% of my code."
19. Première release Claude Code. "It was not a hit. There's a lot of people that used it, but it did not have this exponential growth that it has today."
20. Inflection point. Opus 4 en mai 2025. "That's like when the exponential growth started, and then it kind of inflected with every model release. It started with Opus 4, then 4.5, then 4.6, now 4.7. It just kind of keeps inflecting."
21. Pari pré-PMF assumé. "essentially we were trying to build this thing that was pre-PMF, and we knew that it wouldn't have PMF for 6 months because we were building for the next model." ### "Coding is solved"
22. Sondage à main levée :
23. "Who writes 100% of their code by hand?" → quelques-uns
24. "Who writes 100% using an agent like Claude Code?" → quelques-uns
25. "Who's somewhere in between?" → majorité ("like 50% solved")
26. Cherny lui-même : 100%. .
27. Codebase Claude Code.
28. "It leaked, so people know."
29. TypeScript + React. — choisis car "very on distribution for the model" (le modèle de l'époque maîtrisait bien ces technos).
30. "Pretty simple. There's nothing really complicated."
31. Modèle écrit 100% du code depuis octobre/novembre 2025. .
32. "For me today, the model writes 100% of my code. I write somewhere usually a few dozen PRs every day."
33. "There was a day last week I did like 150 PRs in a day. That was a record."
34. Limites assumées. "this is not the case everywhere. There's very big complicated codebases. There's weird languages the model isn't good at yet. Usually the answer is just wait for the next model." ### Setup personnel actuel (mises à jour)
35. iOS first. "most of my work I do from my phone."
36. App Claude → onglet code. "a little code tab on the left-hand side, I just have a bunch of sessions going."
37. 5-10 sessions. ouvertes, "a few hundred agents" actifs en journée, "a few thousand" la nuit pour deeper work.
38. Sub-agents. + /loop slash command = nouvelle primitive principale.
39. *"Loops are the future at this point. If you haven't experimented with it, highly highly recommend it.". * ### Loop : la nouvelle primitive
40. Mécanisme. "You have Claude use cron to schedule a job for some point in the future, and it's a repeat job."
41. Cadences. every minute / every 5 minutes / every day — paramétrable.
42. Exemples Cherny.
43. PR babysitter. "fixing CI, auto-rebasing"
44. CI healthkeeper. "if there's a flaky test or whatever, it'll go and fix it"
45. Twitter feedback clusterer. "grabs feedback from Twitter and clusters it for me every 30 minutes"
46. Routines. = équivalent serveur du loop : "we just launched routines. Even if you close your laptop, it keeps going."
47. 4.7 démarre des loops spontanément. "I'll tell it 'pull this data query.' And it's like 'Hey, I noticed the data is changing over time. I'll start a loop and I'll give you a report every 30 minutes.' And I'm like 'Great. Can you send it to me over Slack?' And then it uses the Slack MCP to do that." ### Vision sur les équipes
48. "There's going to be a lot more generalists than there are today."
49. Pas seulement "product engineers cross-platform" (iOS+web+server) mais cross-disciplinary generalists : "engineers that are really good at product engineering, but also really great at design. Or really great at product and data science and engineering."
50. Sur la team Claude Code.
51. "Every single person on our team writes code."
52. Liste : engineering manager, PM, designers, data scientist, finance guy, user researcher.
53. "They're specialist in something, but now also everyone's just coding." ### Vision SaaS / disruption
54. Pas d'apocalypse SaaS. , mais réordonnancement via le framework 7 Powers de Hamilton Helmer.
55. Powers qui deviennent moins importants.
56. Switching costs. — "because you can just use the model and you can port from one thing to a different thing."
57. Process power. — "Claude is getting really good at figuring out process. And especially with 4.7, it can just hill climb anything. So if you give it a target and you tell it to iterate until it's done, it will just do it. I think this is the first model like that."
58. Powers qui restent importants.
59. Network effects.
60. Scale economies.
61. Cornered resources.
62. Prédiction startup. "the number of startups in the next 10 years that are just going to disrupt everything is going to increase like 10x."
63. Pourquoi maintenant. "a tiny startup can build a thing that's as valuable as a large company and actually compete head-to-head, because the large company has to evolve their business process, retrain everyone, face internal resistance. But you, starting fresh, can build with AI natively from the ground up."
64. "It's the best time to build. It's the best time to be a startup. There's so much disruption coming." ### Analogie pivot — Presse de Gutenberg
65. 1400s Europe avant la presse. "essentially 10% of the European population was literate. They knew how to read and write. They were often employed by kings and lords that were not literate."
66. 50 ans après l'invention. "more literature published in Europe than in the thousand years before."
67. Coût d'un livre. ↓ 100× sur la même période.
68. Quelques siècles plus tard. "literacy globally went up to like 70%. Now we can all read and write. You don't need a degree in reading and writing to know how to read and write. Although still there are professional writers."
69. Analogie applicquée à 2026+. "software will be a thing that is fully democratized, that anyone can do. And it's going to be much faster than 50 years."
70. Phrase-pivot mémorable. "The best person to write accounting software, I think maybe even today, is not an engineer, it's a really good accountant. Because they know the domain really well, and coding is the easy part. It's knowing the domain that's the hard part." ### Modèle vs harness (Q&A)
71. "If you asked me a year ago, the ratio was maybe 50/50. If you asked me 6 months ago, would be 50/50. In 2 years? I don't know dude, we plan in like 1 week out."
72. Le harness devient moins important. au fur et à mesure que le modèle s'améliore.
73. Évolutions du harness.
74. Loops first-class
75. Easier to run lots of agents
76. Sub-agents (1 idée parmi d'autres)
77. "There's a bunch more stuff that we're cooking."
78. Dans 1 an. "the model will be much better aligned. And so all the safety mechanisms that we have today around prompt injection and static verification of commands and permission modes, human in the loop, all this kind of stuff is just going to be less important because the model will just do the right thing." ### Démocratisation : tous codeurs comme on sait envoyer un SMS
79. "I think it's going to be a skill like — yeah, like I know how to send a text message."
80. Référence forte au modèle de la presse de Gutenberg ci-dessus. ### Pratiques internes Anthropic
81. Pas d'avantage modèle. "internally, we use the same models everyone else does." Mythos pour test, beaucoup d'Opus 4.7 pour dogfood.
82. Pratique radicale.
83. "We use Claude for literally everything."
84. "Our Claudes are talking all day. As I'm coding, as my Claudes are coding in a loop, they will communicate over Slack to talk to other people's Claudes that are also running in a loop, to figure out unknowns."
85. "We have no more manually written code anywhere at the company."
86. "All of the SQL is written by models. Everything is just built by the models."
87. L'avantage Anthropic n'est pas la techno mais le process. "the place that we're ahead is not the technology — same technology available to us is available to everyone here, because fundamentally we are building a platform. There's actually a far bigger lead in organizational structure and organizational process." ### Multi-agent / parallélisation
88. "On the product side, it really just comes down to prompting. As the model gets better, it just naturally does this."
89. "It's not on users to figure out how to hold the tools better. If that's the case, it's actually a product design problem and I'm not doing a good job." ### Local vs cloud, computer use, MCP
90. Local vs cloud : "doesn't matter. By a couple years from now, the model is just going to be doing all the code. It's going to be starting the agents. It's going to be building the environments. If it decides 'actually I'll use local models to do this,' that's what it'll do. These I don't think these will be decisions that we are making as engineers anymore."
91. Knowledge work : "largely, it's there already with Salesforce and Docs. For us, it's always just the simplest answer. It's MCP."
92. Computer use : "I think currently, Anthropic is pretty far ahead on computer use. It's quite good. It's very slow, but it does it quite well now, especially with 4.7."
93. "All this stuff just doesn't matter that much. It could be MCPs, APIs, just some sort of programmatic access — to the model, it's just tokens." ### Produits à venir
94. Claude Design. "It's pretty good today. It's going to get a lot better."
95. Claude Code. "There's also a few things that we're cooking up for Claude Code that are going to be landing over the coming weeks."
96. Loop, batch, massive parallelism d'agents. "that's going to get better."
97. Computer use. . ### Articulation dossier veille
98. Quatrième fiche Cherny. dans le dossier après :
99. cherny-yc-lightcone-claude-code-origin-story-2026-02. (genèse YC Lightcone)
100. cherny-claude-code-10-tips-team-x-2026-02-01. (10 conseils X)
101. nunez-cherny-claude-code-workflow-venturebeat-2026-01-05. (workflow VentureBeat)
102. wu-cherny-use-claude-code-builders-every-2025-10-29. (Wu/Cherny Every)
103. Plus la fiche anthropic-claude-code-quality-postmortem-2026-04-23 dont Cherny est un des auteurs implicites.
104. Apport spécifique de cette fiche. vs les antérieures :
105. Setup iOS exclusif. = nouveau (vs setup CLI/desktop des fiches précédentes).
106. /loop + Routines. = nouvelle primitive d'architecture agentique présentée publiquement.
107. Récit "Anthropic Labs incubator + Mike Krieger round 2". = nouveau détail organisationnel.
108. Inflection Opus 4 mai 2025. = datation précise du décollage.
109. "100% code généré depuis octobre/novembre 2025". = datation personnelle.
110. Analogie Gutenberg. + "the best accountant writes the best accounting software" = phrases-pivots mémorables.
111. "7 Powers" Helmer reordering. = analyse SaaS structurée.
112. Sequoia / Lauren Reader. = nouveau support, public investisseurs.
113. Convergence majeure.
114. /loop agents en cron. = mise en pratique radicale du Ralph Loop (Trivedy 2026-03-10), du Compound Engineering d'Every (Shipper/Klaassen 2025-12-11), et du Compounding Teams (Schillace 2025-09-28).
115. "Claudes talking to Claudes over Slack". = mise en pratique des prédictions Levie Building for Trillions of Agents (2026-03-07) et de l'agent-native infra prônée par Karpathy (2026-04-29).
116. "Coding is solved" + 100% code généré + 150 PRs/jour. rejoint Curran/Intercom 3× R&D productivity (2026-04-16, 93,6% PRs agent-driven), Stripe Minions (2026-02-09 et 2026-02-19), StrongDM Software Factory (2026-02-06).
117. Démocratisation type Gutenberg. = même thèse que Karpathy (2026-04-29) sur les neural computers et que Andreessen sur la redéfinition du programmeur (2026-02). Aussi cohérent avec le projet "tout le monde peut coder" porté par Wescale (2026-05-03) et Habert PROJ-AI (2026-05-05).
118. "7 Powers" et SaaS reordering. = pose un cadre stratégie d'entreprise rare dans le corpus 2026, complémentaire de Sun NYT Permanent Underclass (2026-04-30) côté économique et de Pawlowski Tension Map (2026-03-30) côté market research.
119. "The harness becomes less important as the model gets better". = nuance directe au courant Harness Engineering (Trivedy 2026-03-10, Osmani 2026-04-19, Böckeler 2026-04-02). Cherny anticipe l'absorption progressive du harness dans le modèle aligné.
120. "Best accountant writes the best accounting software". = corollaire de la long-tail de besoins métier que les fiches Wescale (2026-05-03 "besoins historiques trop coûteux peuvent enfin être adressés") et Soto Developer Taste (2026-04) anticipaient.
121. Limites à signaler.
122. Optimisme techno-centric. assumé — Cherny est créateur Claude Code, biais d'évangéliste évident sur "coding is solved".
123. Pas de chiffres externes. — il s'appuie sur sa pratique personnelle et celle d'Anthropic. À comparer avec les chiffres rigoureux de Curran/Intercom (3× R&D, 93,6% PRs agent-driven).
124. Sequoia format. pas d'adversarial questioning, ambiance conviviale, pas de challenge sur la jagged intelligence (Karpathy 2026-04-29) ou sur la permanent underclass (Sun NYT 2026-04-30).
125. "Coding is solved". déclaration provocatrice qui mérite d'être nuancée par Karpathy "Opus 4.7 will simultaneously refactor a 100,000 line codebase and tell me to walk to a car wash 50 meters away" (2026-04-29) — le coding en surface est en effet automatisable, mais le jugement systémique reste un domaine difficile.
126. À mobiliser pour. présentation rétrospective 18 mois Claude Code (récit + datations) ; argumentaire startup "best time to build" en COMEX ; présentation "loops as the future" pour équipes ingénierie ; cours/conférence "démocratisation du software par analogie Gutenberg" ; débat sur la position du harness dans la doctrine 2026.