The Creator of Claude Code on The Hottest Piece of Software in the World
Episode
66 min
Read time
3 min
Topics
Productivity, Remote Work, Leadership
AI-Generated Summary
Key Takeaways
- ✓Model capability drives adoption, not UX: Claude Code's growth inflected three distinct times — at Opus 4 (May), Opus 4.5 (November), and Opus 4.6/Fable (February) — demonstrating that product adoption is almost entirely driven by underlying model intelligence improvements, not interface changes. Businesses evaluating AI coding tools should prioritize frontier model access over feature sets, since raw capability improvements deliver the largest productivity jumps.
- ✓Prompt injection defense via neural probes: Anthropic ran a one-week external red-team competition offering $20,000 to researchers who could prompt-inject Claude Code. Researchers successfully attacked every competing model but failed on Claude Code. The defense uses three layered mechanisms: alignment training, mechanistic interpretability probes that detect injection in model neurons, and Claude Code's auto-permission mode that eliminates manual approval prompts entirely.
- ✓Enterprise adoption follows a sequential ladder: Companies scale Claude Code usage in stages — starting with one AI session per employee, then expanding to 10, 100, and eventually 1,000 concurrent Claude sessions per engineer. Businesses seeing the largest productivity gains are those that restructure workflows around Claude as the center, mirroring how PC-era companies that digitized core processes outperformed those that simply added a computer in the corner.
- ✓Software engineering roles are fragmenting into five types: As code-writing becomes automated, Cherney identifies five emerging roles replacing traditional engineering titles: prototypers (rapid first-idea iteration), builders (bringing products to market), maintainers (managing scale), scalers/growers (10x-100x expansion), and perfectors (eliminating rough edges). Notably, designers and product managers on the Claude Code team already write their own code directly, eliminating handoff bottlenecks.
- ✓COBOL migration is now economically viable: Bun's engineering team migrated an entire codebase from Zig to Rust in 11 days using one engineer and Claude Code dynamic workflows, at a cost of approximately $150,000 in API credits. Previously, equivalent migrations required multiple engineers working for roughly a year, making them cost-prohibitive. Banks currently use Claude Code for legacy COBOL modernization projects that were previously impossible to justify financially.
What It Covers
Boris Cherney, creator of Claude Code at Anthropic, explains how the AI coding tool evolved from a safety research instrument into software used by NASA, major banks, and Fortune 500 companies. By 2026, Anthropic reports 98% of its internal code is written by Claude Code, with growth inflecting sharply at each new model release from Opus 4 through Fable.
Key Questions Answered
- •Model capability drives adoption, not UX: Claude Code's growth inflected three distinct times — at Opus 4 (May), Opus 4.5 (November), and Opus 4.6/Fable (February) — demonstrating that product adoption is almost entirely driven by underlying model intelligence improvements, not interface changes. Businesses evaluating AI coding tools should prioritize frontier model access over feature sets, since raw capability improvements deliver the largest productivity jumps.
- •Prompt injection defense via neural probes: Anthropic ran a one-week external red-team competition offering $20,000 to researchers who could prompt-inject Claude Code. Researchers successfully attacked every competing model but failed on Claude Code. The defense uses three layered mechanisms: alignment training, mechanistic interpretability probes that detect injection in model neurons, and Claude Code's auto-permission mode that eliminates manual approval prompts entirely.
- •Enterprise adoption follows a sequential ladder: Companies scale Claude Code usage in stages — starting with one AI session per employee, then expanding to 10, 100, and eventually 1,000 concurrent Claude sessions per engineer. Businesses seeing the largest productivity gains are those that restructure workflows around Claude as the center, mirroring how PC-era companies that digitized core processes outperformed those that simply added a computer in the corner.
- •Software engineering roles are fragmenting into five types: As code-writing becomes automated, Cherney identifies five emerging roles replacing traditional engineering titles: prototypers (rapid first-idea iteration), builders (bringing products to market), maintainers (managing scale), scalers/growers (10x-100x expansion), and perfectors (eliminating rough edges). Notably, designers and product managers on the Claude Code team already write their own code directly, eliminating handoff bottlenecks.
- •COBOL migration is now economically viable: Bun's engineering team migrated an entire codebase from Zig to Rust in 11 days using one engineer and Claude Code dynamic workflows, at a cost of approximately $150,000 in API credits. Previously, equivalent migrations required multiple engineers working for roughly a year, making them cost-prohibitive. Banks currently use Claude Code for legacy COBOL modernization projects that were previously impossible to justify financially.
- •Agentic loops represent the next product overhang: Cherney identifies a "product overhang" — where model capability exceeds what current interfaces allow users to experience. The solution is shifting from single-prompt interactions to persistent multi-agent loops (Claude Tag) running for weeks continuously, given goals rather than detailed instructions. Anthropic's internal Slack bot proactively joins relevant conversations unprompted, pulling data from Datadog and BigQuery simultaneously to surface analysis without being asked.
Notable Moment
Cherney revealed that during early desktop app development, a Claude prototype tasked with ordering a pizza completed the order and then, apparently idle, began browsing Hacker News on its own. He attributes current models staying on task to years of alignment research rather than any hard constraint preventing autonomous behavior.
Episode Transcript
Now a message from Meta. Meta is launching America's Workforce Academy. The program offers paid training, a job, and a path to America's future because the future is for everyone. Learn more at meta.com/america's workforce academy. The thing about AI for business, it may not automatically fit the way your business works. At IBM, we've seen this firsthand. But by embedding AI across HR, IT, and procurement processes, we've reduced cost by millions, slashed repetitive tasks, and freed thousands of hours for strategic work. Now we're helping companies get smarter by putting AI where it actually pays off, deep in the work that moves the business. Let's create smarter business. IBM. Healthcare doesn't always work great. If you've ever waited on a refill or couldn't schedule an appointment, you get it. That's the kind of stuff Optum is changing. They're using data and technology to integrate patient care, pharmacy, and everything else. So health care is connected, not complicated. What's that look like? Cheaper prescriptions that are easier to get and care that looks at the whole person, how you need it. Optum is helping make health care work as one for everyone. Learn more at business.optum.com. Bloomberg Audio Studios. Podcasts, radio, news. Hello, and welcome to another episode of the Odd Lots podcast. I'm Joe Weisenthal. And I'm Tracy Alloway. Tracy, so I think the most embarrassing moment for me in Go on. It's the most exciting way you've ever started a podcast, Jo. Maybe not the most embarrassing way at the moment I felt like I'm, like, making myself a little stupider or something like that in 2026 was I asked Claude Code Mhmm. To, clean up all the many screenshots that I had on my desktop. Oh. So I was like, just put the I have all these told me about that. All these, like, screenshots on my desktop, various charts and stuff. And I was like, clogged code. Can you do this? And in that moment, I realized that I was essentially outsourcing my computer to another computer. There's big data centers, etcetera, that Anthropic has. And rather than just, like, taking a few seconds, like, drag and drop some screenshots, I was like, no. I'm gonna have another computer use my computer for me. That just seems efficient. But here's here's the big question. Did it do it correct? Absolutely. Yeah. It was it was perfect. Alright. Because you hear the stories about agents going off the rails. Like, there was some software company or, like, car rental software company, and I think they had an agent that deleted their entire database and then admitted that it had violated its core principles in doing so, but didn't have an explanation as to why. There's definitely been times in my cloud code usage, which is not very sophisticated, where it'll just ask me, like, do I do this or this? And I have no idea what it's asking for, and I just, like, hit yes. Hesitantly …
Get the full transcript (14,942 words) + summary by email — free
One-time email with the complete transcript and AI summary of this episode. No account needed.
One email, no spam. We’ll also show you what SignalCast does.
You just read a 3-minute summary of a 63-minute episode.
Get Odd Lots summarized like this every Monday — plus up to 2 more podcasts, free.
Pick Your Podcasts — FreeKeep Reading
Books, tools, and gear mentioned in this episode
SignalCast may earn commission on purchases via these links. As an Amazon Associate, SignalCast earns from qualifying purchases.
Tools
by Google
“Anthropic's internal Slack bot proactively joins relevant conversations unprompted, pulling data from Datadog and BigQuery simultaneously to surface analysis without being asked.”
by Anthropic
“Boris Cherney, creator of Claude Code at Anthropic, explains how the AI coding tool evolved from a safety research instrument into software used by NASA, major banks, and Fortune 500 companies.”
- Claude TagBy guest
by Anthropic
“The solution is shifting from single-prompt interactions to persistent multi-agent loops (Claude Tag) running for weeks continuously, given goals rather than detailed instructions.”
by Slack Technologies
“Anthropic's internal Slack bot proactively joins relevant conversations unprompted, pulling data from Datadog and BigQuery simultaneously to surface analysis without being asked.”
More from Odd Lots
We summarize every new episode. Want them in your inbox?
What's Behind the Big Surge in US Government Bond Yields
Adam Posen Thinks Things Could Get Very 'Messy' for the Fed
Richmond Fed’s Tom Barkin on the Surprisingly Resilient Real Economy
Austan Goolsbee Is Worried the Economy Is Overheating
Kansas City Fed President Jeffrey Schmid on the First Jackson Hole of the Warsh Era
Similar Episodes
Related episodes from other podcasts
The Vergecast
Feb 24
How Claude Code Claude Codes
The AI Breakdown
Jul 5
The Job Positions of the AI Future
Latent Space
Mar 17
Why Anthropic Thinks AI Should Have Its Own Computer — Felix Rieseberg of Claude Cowork & Claude Code Desktop
Lenny's Podcast
Feb 19
Head of Claude Code: What happens after coding is solved | Boris Cherny
Software Engineering Daily
Jul 30
Docker and Sandboxing AI Agents
Explore Related Topics
This podcast is featured in Best Finance Podcasts (2026) — ranked and reviewed with AI summaries.
You're clearly into Odd Lots.
Every Monday, we deliver AI summaries of the latest episodes from Odd Lots and 192+ other podcasts. Free for one show.
Start My Monday DigestNo credit card · Unsubscribe anytime