The End of Hand-Written Code: Boris Cherny on AI Programming's Terminal Stage
Source: Boris Cherny interview, Sequoia AI Ascent 2026, 2026-05-06
---
1. Opening: Who Is Still Writing Code by Hand?
At Sequoia's AI Ascent 2026, Boris Cherny, creator of Claude Code, asked the audience, "Who here suffers from Claude Code psychosis? Raise your hand." Many hands went up.
"Don't be embarrassed," he said. "My team says I have it too. Probably true."
This is not a joke. As of 2026, Boris has not personally written a single line of code. His primary workspace is the Claude mobile app on iOS. He merges dozens of pull requests per day, with a personal record of 150 in one day. Hundreds of agents run in parallel, and thousands continue deep async tasks overnight.
A former TypeScript textbook author and engineer at heart, Cherny is demonstrating a thesis in practice: programming as the craft of hand-writing code has come to an end.
---
2. Programming Is Dead? No, It Is Solved
Cherny asked the room three questions:
- Who still writes 100% of their code by hand? A few hands.
- Who writes 100% of their code with AI? More hands.
- Who is somewhere in between? Almost everyone.
- A PR-watchdog that auto-fixes CI and rebases branches
- A CI-healthkeeper that detects and repairs flaky tests
- A feedback harvester that scrapes Twitter every 30 minutes, clusters complaints, and pushes a digest to him
- Zero manual code or SQL across the company
- Agents from different employees messaging each other on Slack
- Autonomous collaboration on unknown problems
- Boris Cherny interview, Sequoia AI Ascent 2026, 2026-05-06
- Chinese transcription, Tencent Cloud Developer Community, 2026-05-14, https://cloud.tencent.com/developer/article/2669104
- Zhidongxian report, 2026-05-08, https://t.cj.sina.cn/articles/view/5213469505/136bf3b41020015w2m
- Jike App digest, 2026-05-11, https://m.okjike.com/originalPosts/69fdbc5feb714cce8c542067
- 51CTO in-depth report, 2026-05-09, https://www.51cto.com/article/842726.html
"It looks like programming is 50% solved," he joked. The room laughed.
Behind the humor is a serious claim: for him personally, programming was completely solved in October and November of last year. The model can write 100% of his code. For complex or obscure languages, he simply waits for the next model release.
Large legacy systems and languages the model has not yet mastered remain harder. But his position is clear: the technical problem is cracked. What remains is a matter of time.
---
3. Loop: From Q&A to Continuous Automation
Cherny's favorite feature is Loop, which lets agents run recurring tasks on a cron schedule.
He personally runs dozens of loops:
Loop turns agents from conversational tools into continuously running automation systems. Anthropic has also launched Routines, server-side loops that execute even when the laptop is closed.
Notably, the 4.7 model has started spontaneously proposing its own loops. Asked to fetch a data point, it replied: "This data is changing constantly. Should I set up a loop that reports every 30 minutes via Slack?" Then it wired itself into Slack MCP and went to work.
The nature of work has changed. Workers are no longer executors but orchestrators who decide who does what, at what cadence, and how results get reported.
---
4. The Phone as Workspace: Hundreds of Agents in Parallel
Cherny's workflow sounds like science fiction. He opens the Claude app and runs five to ten concurrent conversations, each spawning many agents, totaling hundreds in parallel.
He keeps five terminal tabs open, each with a Claude instance, and round-robins between them: launch a plan mode in tab 1, switch to tab 2, launch another, and repeat. When tabs overflow, work spills to the desktop, then to the phone.
> "If someone had told me six months ago that I'd be writing a third or even half of my code on a phone, I'd have called them crazy. But that's what I do today."
The bottleneck is no longer typing or thinking speed. The bottleneck is context switching: how efficiently you manage multiple concurrent workflows.
---
5. Everyone Codes: The Rise of Cross-Disciplinary Generalists
Anthropic has one internal rule: all titles are equal — Member of Technical Staff.
This is not cosmetic. It is an organizational signal. "If your Slack title says 'software engineer,' I assume you write code and I won't ask you product questions. But when everyone has the same title, you're expected to do everything."
Inside Claude Code, engineering managers, product managers, designers, data scientists, and finance staff all write code with AI. Once the coding barrier is flattened, the scarce resource is no longer syntax memorization but deep understanding of the business and the user.
Cherny's prediction: the most valuable talent of the future is the cross-disciplinary generalist — someone fluent in product, design, and engineering, or in data science, product, and engineering. AI compresses execution cost toward zero, amplifying anyone who can think across boundaries.
---
6. The Gutenberg Moment for Software
Cherny compares today's AI coding to the printing press of the 1400s.
The press did not eliminate authors, but lifted literacy from 10% to 70%. Likewise, AI will not eliminate programmers, but will turn "writing code" from a rare specialist skill into a universal basic literacy.
Software development is being democratized. Writing code is becoming as cheap and simple as sending a text message. The best accounting software of the future will be built directly by the accountants who understand the business best, not by pure programmers. Programming is easy; domain knowledge is hard.
Cherny even predicts: Claude Code's own codebase may shrink to about 100 lines within a year. When the model is smart enough, you do not need product-layer scaffolding to support it.
---
7. A Golden Decade for Startups
AI is eroding two traditional moats of large incumbents: switching costs and process capability.
Large companies face enormous internal inertia — process changes, retraining, organizational resistance. Lean startups can build natively with AI from day one.
Cherny forecasts: the number of disruptive startups will grow 10x over the next decade. Tasks that once required 20-person engineering teams to compete with incumbents now need three to five people.
Network effects, economies of scale, and scarce-resource moats still hold. AI changes how work is executed, not the fundamental nature of business.
---
8. The Organizational Generation Gap
Technology platforms are evenly available, but frontier companies pull ahead on organizational structure and process.
Anthropic has internally achieved:
The gap is not "you have AI" versus "you don't." It is "you have AI woven into the organization's bloodstream" versus "you bought a few AI tools."
For the model, API calls, MCP, and Computer Use are all just tokens. It does not care about the access path, only whether it can get the information and execute the actions it needs. The real differentiator is whether you can weave those capabilities into daily operations.
---
9. Product Overhang: Building for the Next Model
Claude Code was an accident. In late 2024, Boris joined Anthropic Labs, a small incubator. They observed that model capabilities had outgrown the prevailing product form (inline autocomplete), yet no product on the market had unlocked that capability.
Internally they call this product overhang: the model is ready, the product is not.
The first six months had almost no product-market fit. Boris used the tool for about 10% of his own code. The real inflection came with Opus 4 in May 2025. Every subsequent model release — Opus 4, 4.5, 4.6, 4.7 — drove a growth kink, compounding exponentially to today.
The lesson for builders: most people study current needs, but if you can predict that a technology direction will mature in six months and prepare the product form in advance, you capture the first wave.
---
10. Closing: You Are Not AI's Rival, You Are Its Commander
Cherny's interview delivered insights on three levels:
Tool level: Claude Code marks a paradigm jump from code completion to ticket handling to multi-agent repository maintenance.
Organization level: Anthropic's internal agent collaboration network is defining the next generation of workflow standards. Competitive advantage is no longer "having AI" but how your organization coexists with it.
Human level: The printing press did not eliminate authors, and AI will not eliminate programmers. But "writing code" will shift from specialist skill to basic literacy. The best industry software of the future will be created directly by the people who understand the business best.
> "The real divide is not whether AI can write code. It is whether your organization can completely reshape its processes so that everyone — product managers, designers, finance — becomes a Builder."
---