Boris Cherny
公开发言整理

Claude Code 诞生故事

"When I first started Claude Code, it was just going to be a little hack. We broadly knew at Anthropic that we wanted to ship some kind of coding product. I spent one month hacking, built a bunch of weird prototypes, most of them didn't ship. Then I spent a month doing post-training to understand the research side. Then I came back and just started prototyping what eventually became Claude Code."

"The very first version — I have a video recording. It was called ClaudeCLI back then. I gave it a bash tool and it just was able to use that to write code to tell me what music I'm listening to when I asked it. And this is the craziest thing — I didn't instruct the model to say use this tool for this. The model was given this tool and figured out how to use it to answer a question I wasn't even sure it could answer."

"The team was called the Anthropic Labs team. We built Claude Code, we built MCP, we built the desktop app. You can see the seeds of this idea: coding, then tool use, then computer use — this has been our belief as a company for a long time."

"The biggest reason we stuck with the terminal was the model is improving so quickly. We felt there wasn't really another form factor that could keep up with it."

"Coding is Solved" — 原话

"Over the next few months I think what we're going to see is across the industry it's going to become increasingly solved — for every kind of codebase, every tech stack that people work on."

"100% of my code is written by Claude Code. I have not edited a single line by hand since November. Every day I ship 10, 20, 30 PRs. At the moment I have five agents running while we're recording this."

"Software Engineer" 头衔将消失

"I do think that there's a future where — I think by the end of the year — what we're going to start to see is these start to get murkier. In some places the title software engineer is going to start to go away. It's just going to be replaced by 'builder', or maybe it's just everyone's going to be a product manager and everyone codes."

"On the Claude Code team, everyone codes. Our product manager codes, our engineering manager codes, our designer codes, our finance guy codes, our data scientist codes. Everyone on the team codes."

活字印刷机比喻

"In the mid-1400s, literacy was actually very low — sub 1% of the population. It was scribes who did all the writing. They were employed by lords and kings. Then Gutenberg and the printing press came along, and in the 50 years after, more printed material was created than in the thousand years before. The cost went down 100x. Literacy went up to 70% globally over the next 200 years."

"There was actually a historical document — an interview with a scribe in the 1400s. They were actually very excited about the printing press. They said: 'The thing I don't like doing is copying between books. The thing I do like is drawing the art and book binding. I'm really glad my time is freed up.' As an engineer I feel a parallel — I don't have to do the tedious work of coding anymore. The fun part is figuring out what to build, talking to users, thinking about big systems."

"I imagine a world a few years in the future where everyone is able to program. Anyone can just build software anytime. And I have no idea what that unlocks. It's just the same way that in the 1400s, no one could have predicted the Renaissance."

Cowork 产品 — 是什么、为谁而建

"Co-work is kind of the first way to get at this — bringing agentic AI to people who haven't really used it before. When I think back to engineering a year ago, no one really knew what an agent was. But nowadays it's just the way we do our work. For non-technical folks today, all they've used is conversational AI — a chatbot. No one has really used an agent before. Co-work might be the first time they're using something that actually acts."

"I had to pay a parking ticket the other day. I just had co-work do it. All of my project management for the team — co-work does all of it. It's syncing stuff between spreadsheets, messaging people on Slack and email."

潜在需求（Latent Demand）— 完整阐释

"Latent demand is this idea that if you build a product in a way that can be hacked or misused by people in a way it wasn't really designed for — to do something that they want to do — then this helps you as the product builder learn where to take the product next."

"I remember walking into the office and our data scientist Brendan had Claude Code on his computer — just a terminal up. I was shocked. He figured out how to open the terminal, downloaded Node.js, downloaded Claude Code, and was doing SQL analysis in a terminal. The next week all the data scientists were doing the same thing."

"The modern framing I've been seeing in the last 6 months: look at what the model is trying to do and make that a little bit easier. For Claude Code, we inverted the approach — the product is the model. We want to expose it. Put the minimal scaffolding around it. Let it decide which tools to run and in what order."

故意欠配置 + 无限 Token 策略

"There's this interesting thing that happens when you underfund everything a little bit — people are forced to use Claude. For work where sometimes we just put one engineer on a project, the way they're able to ship really quickly is because they want to ship quickly. And if you have Claude, you can just use that to automate a lot of work."

"My advice generally: don't try to cost cut at the beginning. Start by just giving engineers as many tokens as possible. We're starting to see 'unlimited tokens' come up as a perk at some companies. At the beginning you just want to throw a lot of tokens at it and see if the idea works and give engineers the freedom to do that."

"At Anthropic we're starting to see some engineers spending hundreds of thousands a month in tokens."

为什么离开 Anthropic 去 Cursor，又两周后回来

"I joined Cursor because I'm a big fan of the product. Honestly I met the team and was just really impressed. They're an awesome team. They saw where AI coding was going before a lot of people did. The idea of building good product was very exciting for me."

"As soon as I got there, what I started to realize is what I really missed about Anthropic was the mission. If you ask anyone at Anthropic why they're here, the answer is always going to be safety. This mission-drivenness just really resonated with me. And I just know personally it's something I need in order to be happy. Whatever the work might be, no matter how exciting — even building a really cool product — it's just not really a substitute for that."

他分享给每个新成员的原则

在 Lenny's Podcast 中 Boris 提及的团队核心原则（原文摘录）：

注：在播客中 Boris 还提到"best thing you can do is have Claude do it"作为团队口头禅，以及"bitter lesson"（更通用的模型总会打败特化模型）作为产品决策依据。

通才在 AI 时代的优势

"Try to be a generalist more than you have in the past. Some of the most effective engineers I work with every day — they cross over disciplines. Some of the strongest engineers are hybrid product and infrastructure engineers, or product engineers with really great design sense, or an engineer with a really good sense of the business, or an engineer that loves talking to users."

"A lot of the people that will be rewarded the most over the next few years won't just be AI native. They're curious, they're generalists, they cross over multiple disciplines and can think about the broader problem they're solving — rather than just the engineering part of it."

安全与 ASL-4 — 为什么安全是一切的核心

"There are three layers of safety. The lowest level is alignment and mechanistic interpretability — when we train the model, we want to make sure it's safe. We have pretty sophisticated technology to understand what's happening in the neurons, to trace it. If there's a neuron related to deception, we're starting to get to the point where we can monitor it. The second layer is evals — a laboratory setting, the model in a petri dish. The third layer is seeing how the model behaves in the wild."

"Opus 4 was our first ASL-3 class model. We just saw this inflection — everyone started to use Claude Code for the first time and our growth really went exponential."

"We released Claude Code really early because we wanted to study safety. We used it within Anthropic for 4 or 5 months before we released it — we weren't sure if it was safe. This was the first big agent that folks had released at that point."

所有电脑工作的未来

"It's going to be a lot of the roles adjacent to engineering — product managers, design, data science. It is going to expand to pretty much any kind of work you can do on a computer, because the model is just going to get better and better at this. An agent is an LM that's able to use tools. It doesn't just talk, it can actually act. It can use your Google docs, send email, run commands on your computer. Any kind of job where you use computer tools — I think this is going to be next."

"Models are running for very long periods of time now. With Sonnet 3.5 a year ago, it could run maybe 15-30 seconds before going off the rails. Nowadays with Opus 4.6, it'll run maybe 10-30 minutes unattended. Some have run for hours or even days. I think over time this is going to become more and more normal."

给感到不安的人的建议

"Experiment with the tools, get to know them, don't be scared of them. Just dive in, try them, be on the bleeding edge, beyond the frontier. Maybe the second piece of advice is try to be a generalist more than you have in the past."

"I have never enjoyed coding as much as I do today because I don't have to deal with all the minutia."