May 20, 2026 · 57m · big-technology
Claude Code Head Boris Cherny: Insane Growth, Tokenmaxxing, AI Agents' Next Frontier
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of the Big Technology Podcast, host Alex Kantrowitz interviews Boris Cherny, Head of Claude Code at Anthropic, to explore the rapid rise of autonomous AI agents and self-writing software. Cherny explains how agentic tool execution, parallel multi-agent workflows, and enterprise restructuring are reshaping software engineering and the broader knowledge economy.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Alex holds 40.8% of the talking time here. How this is scored →
Alex as informed peer 5.4 Guest teaching 4.1 Guest disagreement 1.9 Alex pushing back 3.4
speaking balance: gold is Alex, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 26:24 Direct rejection of LLM architectural limits
Boris firmly disagrees with a critic's claim that probabilistic LLMs cannot achieve agentic reliability, pointing out that Claude Code already writes 100 percent of its own codebase.
Hardest push from Alex ▶ 18:35 Host confronts guest with FT report on fake AI tasksAlex refuses to accept that token consumption is purely organic, confronting Boris with Financial Times reporting and firsthand employee accounts of automated dummy loops.
Biggest teaching moment ▶ 20:10 Reframing AI adoption via 1990s PC productivity studyBoris educates the host by drawing a direct historical analogy to a 1990s Harvard Business Review article about the structural business changes required for personal computers to yield measurable productivity.
Alex holds their own ▶ 42:27 Host uses Ethan Mollick's ASI critique against lab claimsAlex cites Ethan Mollick's observation about AI labs hiring Salesforce admins and forward-deployed engineers to challenge the narrative of autonomous artificial superintelligence.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Alex as informed peer | Guest teaching | Guest disagreement | Alex pushing back | Why |
|---|---|---|---|---|---|---|
| Anthropic's Exponential Growth and Dual Product Strategy | 5 | 3 | 1 | 3 | Alex demonstrates strong industry context by quoting Dario Amodei's ARR milestones and questioning the revenue split between API and first-party products. Boris provides inside color on product demand and Anthropic's mission while deflecting exact revenue proportions. | |
| Defining Claude Code and Agentic Tool Use | 4 | 2 | 1 | 1 | Alex offers a concise definition of Claude Code and clarifies the concept of tool integration. Boris validates the framing and explains the historical pivot from autocomplete text editors to agentic tool use. | |
| Agentic Execution and Changing Mental Models | 5 | 3 | 1 | 1 | Alex articulates the paradigm shift from next-token autocomplete to multi-step computer control. Boris illustrates this capability through a concrete anecdote about booking multi-city itineraries autonomously. | |
| Analyzing Tokenmaxxing and Engineering Productivity Gains | 6 | 4 | 2 | 5 | Alex pressures the core narrative by introducing the concept of tokenmaxxing and asking whether enterprise adoption is artificially gamified. Boris draws on his Meta background to contrast marginal engineering gains with 250 percent AI productivity spikes. | |
| Corporate AI Incentives and the 1990s PC Productivity Paradox | 7 | 6 | 2 | 6 | Alex cites reporting from the Financial Times and verified Amazon employee accounts regarding artificial token consumption. Boris addresses the skepticism by reframing AI adoption through the lens of the 1990s PC productivity paradox. | |
| Model Efficiency, Token Looping, and Effort Controls | 6 | 4 | 1 | 4 | Alex brings up model inefficiencies with a personal example of Claude looping on PDF exports and references a listener comment claiming LLM probabilistic flaws are unfixable. Boris explains the trade-offs between intelligence, speed, and user-selected effort controls. | |
| Self-Writing Codebases and Overcoming LLM Limitations | 4 | 6 | 4 | 2 | Boris explicitly rejects the critic's thesis that LLM architectural limits prevent agentic reliability, citing that Claude Code is entirely self-written and referencing Y Combinator founder adoption. | |
| Autonomous Delegation and Trusting Claude Cowork | 4 | 2 | 1 | 1 | Alex compares trusting autonomous agents to his initial white-knuckle rides in autonomous Waymo vehicles. Boris agrees, detailing how non-engineers adopt agentic workflows to diagnose OS settings. | |
| Rate Limit Infrastructure and Parallel Agent Swarms | 5 | 5 | 2 | 4 | Alex pushes on user churn driven by strict rate limits. Boris responds with operational metrics showing low overall hit rates while explaining how advanced users deploy swarms of hundreds of parallel instances. | |
| Datacenter Buildouts, Compute Scaling, and Market Competition | 6 | 4 | 2 | 5 | Alex questions whether Anthropic's compute discipline puts them at a disadvantage against OpenAI's heavy datacenter spending and Codex. Boris defends Anthropic's capacity expansion via Colossus and frames rival products as flattering copycats. | |
| Mid-Show Reset and Enterprise Agent Expansion | 4 | 4 | 1 | 2 | Alex recaps enterprise integrations into accounting tools like QuickBooks. Boris breaks down technical developments in Auto Mode, explaining multi-Claude safety routing that replaces repetitive user confirmation prompts. | |
| Scaling Parallel Agents and Proactive Chatbot Interfaces | 7 | 5 | 3 | 6 | Alex cites Ethan Mollick's critique about AI labs hiring Salesforce admins and consultants as evidence of agent capability limits. Boris counters by explaining that AI creates individual leverage requiring humans to manage recursive prompt chains. | |
| The SaaSpocalypse and Evolving Software Moats | 6 | 5 | 2 | 4 | Alex explores whether universal agentic interfaces destroy traditional SaaS moats. Boris applies the Seven Powers framework, arguing that network effects and manufacturing scale economics remain durable defensibility moats. | |
| Recursive AI Self-Improvement and Long-Term Safety | 7 | 5 | 3 | 4 | Alex probes self-improving AI timelines from Jack Clark and Yann LeCun's world model criticism. Boris offers to live-demo Claude Code with LeCun and highlights research on emergent planning inside next-token prediction. |
Statements from this episode (23)
Cherny: The defining difference of an AI agent is tool use
“The thing that made quad code different from chatbots at the time was quad code can use tools. And this is it. Like, this is just the difference. It's, with a chatbot, you're going back and forth and you're talking, but an agent, and Cloud Code is an agent, it…”
Cherny: Claude Code was first AI tool able to edit local computer files
“Even something as simple as like editing a file on your computer. You know, like a year and a half ago, there was no AI product that could actually do that. But this is the first thing that quad code was able to do. It could edit a file on your desktop.”
Cherny: Rapid AI progress requires continuously retesting previously failed tasks
“Every month there's a step change in what it can do. And as a user of this technology, it's just quite hard because you have to kind of keep retraining. You have to keep retrying. You always need this like beginner mindset to retry the technology and use it fo…”
Cherny: Tokenmaxxing does not make up a large percentage of Claude usage
“I don't think token maxing is a large percent.”
Anthropic Engineers Increased Code Output by Hundreds of Percent With Claude Code
“What happened with Claude is now many companies, including Anthropic, and all of our biggest customers are reporting gains on the order of hundreds of percentage points. And I think the last number that we reported is the amount of code written per engineer at…”
Cherny: Do not optimize AI workflows early because top innovators are unpredictable
“The challenge is you can't identify these engineers and these people ahead of time. You don't know who they are. And it's almost always going to surprise you. And so the thing you want to do is let people experiment, give them safety, and then once there's som…”
Cherny: AI models should optimize for intelligence before efficiency
“I think we should probably optimize for intelligence. That's the most important thing. So even if it's like a little bit less efficient, but it's more intelligent and it lets you do more things, that's really useful because the efficiency optimization comes af…”
Cherny: Claude Code and Claude Cowork Were Entirely Written by Claude Code
“QuadCode is a hundred percent written by QuadCode. Cowork is a hundred percent written by QuadCode.”
Cherny: Half of YC Founders Write Entire Codebases Using Claude Code
“I did a talk at Y Combinator, you know, the startup incubator yesterday, and I asked people to raise their hands, you know, everyone, everyone's using QuadCode, and I asked them, You know, raise your hand if a hundred percent of your code is written using quad…”
Kantrowitz: Users Maximize Claude's Value by Granting Autonomous Browser Control
“You realize that you can only get the benefit of this, or you're, you'll get most benefit by letting Claude take over your browser and do things for you.”
Cherny: New AI Users Are Often More Ambitious Than Veteran Users
“I think for people that have kind of grown up with these products and they've seen previous versions, they might not be as ambitious as they could, but for people that are new to the products, I often see them using quad code and cowork for things that I would…”
Anthropic Exec Runs Hundreds of Parallel Claude Instances Every Night
“Nowadays, I'm running, you know, like, on my computer, I run maybe five at a time, and then every night, I run, like, you know, not every night, but most nights, I run, like, hundreds of quads at a time. Hundreds. Yeah, hundreds, sometimes thousands.”
Anthropic Deployed New Colossus Compute Capacity to Serve Claude Code Users
“And of course, we announced the new Colossus capacity, which, you know, we brought online to serve all these new users.”
Cherny: Multi-Claude Automated Safety Checks Are Now Safer Than Human Approvals
“Essentially we found both in the laboratory setting, and now we're finding in the wild, this is safer than what we had before. So as a user, it's a really nice benefit, because you don't have to sit there and say yes over and over, and actually the result is b…”
Cherny: Most Claude Code Users Run Multiple Instances Simultaneously
“One of the cool things about quad, and this is something that we started to see pretty early with quad code users, is actually very few people nowadays run one quad code at a time. Most people run many, many quad codes, you know, ranging from, you know, a few …”
Cherny Stopped Writing Code Entirely, Using Claude to Prompt Other Claudes
“When you look at the kind of engineering that I do, I don't write code. I prompt quad. And actually nowadays, mostly what I'm doing is I have a Claude that prompts other Claude. So I don't even talk to Claude. I have a Claude that's talking to my Claude.”
Cherny: Half of Anthropic's GTM team uses Claude Code
“Actually at Anthropic, I think like half the go to market team uses Claude code and the other half uses core.”
Cherny: Prompt chains will deepen, but humans will still pilot AI
“And at some point, Claude is going to become really good at asking Claude to do this. And that person is going to be asking Claude that asked Claude to do this. And this chain will just keep getting deeper, but in the end, you still need people that are piloti…”
Cherny: AI Commoditizes Code, Making Network Effects the Ultimate Business Moat
“One that I think will increase in importance is something like network effects, because it doesn't matter who's writing the code. It doesn't matter if it's an agent at the core of your product or something else, or if there's intelligence in your product. If t…”
Cherny: AI Agents Will Destroy Switching Costs as a Software Moat
“Some modes get less important, and this is, for example, switching costs, because if you want to switch from vendor A to vendor B, you can, you know, you can just ask Quad to do that, and Quad is going to get better and better over time at it.”
Cherny: AI coding models will eventually form a self-reinforcing loop
“Quad is starting to generate its own ideas for what to build next for QuadCode, but it's, you know, it's not always good ideas, and I still generate most of the ideas, and, you know, at some point it's gonna change. The model's gonna improve, and it's gonna be…”