@FeitengLi: The loop / loop engineering that's been trending on Twitter recently comes from this May 4 interview — Anthropic's Claude Code creator Boris Cherny (author of a TypeScript book, a coder at heart)...
Summary
Anthropic's Claude Code creator Boris Cherny systematically introduced the loop engineering method at Sequoia AI Ascent 2026, stating that programming is now 100% done by AI, and predicted that software will become as democratized as the printing press.
View Cached Full Text
Cached at: 06/16/26, 03:39 PM
Recently, the term “loop / loop engineering” has been trending repeatedly on Twitter. The origin is this interview from May 4th — Boris Cherny, the creator of Anthropic’s Claude Code (author of a TypeScript textbook, an engineer who builds things by writing code), spoke for the first time at Sequoia AI Ascent 2026 about how he uses loops. I’ve translated the full interview video into Chinese.
The first point he made: for him, programming is 100% solved.
“For me, 100%.”
Not just talk — he said the model now writes 100% of his code, he submits dozens of PRs a day, and last week his single-day record was 150 PRs. The codebase of Claude Code is no secret (it’s been leaked) — it’s just TypeScript plus React.
So how does one person sustain 150 PRs a day?
By looping. This is the method being discussed everywhere — he has Claude use cron to schedule repetitive tasks and just lets them run:
“And by now, I’ve got dozens of loops running all sorts of tasks.”
One loop watches PRs, fixes CI, and auto-rebases; another grabs feedback from Twitter every 30 minutes and clusters it for him. His verdict is strong:
“I think loops are the future right now.”
He also mentioned the recently launched “routines” — same concept but running on the server — “even if you shut your laptop, it keeps going.”
But is this just a solo geek’s play?
No. He said that inside Anthropic, “there’s no hand-written code anymore — all SQL is written by models.” Each engineer’s Claude runs in a loop, and they even communicate through Slack to solve problems together.
If writing code is this cheap, does that mean software companies lose their value — the end of SaaS?
He doesn’t think so, but believes the moats will reshuffle. Borrowing from Hamilton Helmer’s “7 Powers” framework: switching costs and process barriers will fail first, while network effects and scale economies still hold. He also bet on a number:
“I think the number of startups that will disrupt everything in the next 10 years will grow by 10x.”
So what does writing software itself become?
He brings up the printing press: 50 years after the printing press was invented, Europe produced more text than the previous thousand years combined, book prices dropped nearly 100x, and literacy eventually rose from about 10% to about 70%. He predicts software will follow the same path of democratization — “much faster than 50 years” — and ultimately “become a skill, like being able to text.”
The punchline is sharp: the best people to write accounting software aren’t engineers — they’re accountants. “Coding is the easy part; the hard part is domain knowledge.”
When asked what things will look like two years from now, the creator gave the most honest answer: “Oh, two years? Dude, I have no idea. We only plan one week ahead.”
· Loops are the future: dozens of cron agents watching PRs, fixing CI, gathering feedback · Programming 100% solved, 150 PRs a day · Company: zero hand-written code, Claudes looping through Slack to talk to each other · Switching costs/process barriers collapse first, disruptors 10x · Printing press analogy: software will democratize like literacy
Similar Articles
@Lonely__MH: Prompts Are Dead, Loop Engineering Has Arrived! Recently, the concept of Loop Engineering in AI programming has sparked widespread discussion in the tech community. In a recent interview, Boris Cherny, head of Claude Code, shared what's happening inside the team…
Claude Code head Boris Cherny proposes that AI programming is shifting from prompt engineering to Loop Engineering. In the future, developers' core task will be designing automated loops rather than writing prompts—a trend that could level the development playing field.
@seclink: Developers are shifting from "one-shot prompts" to "agentic loops"—where AI like Anthropic's Claude sets its own goals, uses tools to execute actions, observes results, and iterates until the task succeeds. Boris Chern, developer of Claude Code...
Developers are moving from one-shot prompts to an 'agentic loop' model, letting AI autonomously set goals and iteratively execute tasks using tools. Claude Code developer Boris Cherny has abandoned traditional IDEs entirely, now running hundreds of agents to monitor issues and merge PRs—even doing so from his phone.
@freeman1266: https://x.com/freeman1266/status/2064702757773496552
This article introduces the concept of Loop Engineering, which involves designing automated systems that allow AI agents to work in autonomous loops, including elements such as automated tasks, work trees, skills, plugins, and sub-agents, thereby replacing manual prompting and improving development efficiency.
@FinanceYF5: Anthropic's Boris Cherny points out that programming is moving towards a higher level of abstraction. A year ago, developers' workflow was to first write code in an IDE, then prompt multiple Claude instances in parallel. Now, the workflow has evolved to writing loops to call Claude, and letting it make autonomous decisions...
Anthropic's Boris Cherny points out that programming is moving towards a higher level of abstraction, with workflows shifting from manually writing code to letting Claude make autonomous decisions, and predicts that the next paradigm shift will arrive this year.
@cellinlab: https://x.com/cellinlab/status/2064144608242679822
This article introduces the concept of Loop Engineering — instead of directly writing prompts for AI agents, it designs a system (loop) that recursively lets the agent iterate on tasks until completion. The article provides a detailed comparison of how Claude Code and Codex implement five building blocks: automations, worktrees, skills, sub-agents, etc. It suggests this could be the future trend of collaborating with coding agents, but also warns about token costs and AI slop issues.