Overview of The Creator of Claude Code on The Hottest Piece of Software in the World
Bloomberg’s Odd Lots interviews Boris Cherney, the creator and head of Claude Code at Anthropic, about why AI coding tools have become so central to the software industry, what makes them powerful, and how they change the future of programming, enterprise workflows, and software interfaces. The discussion frames Claude Code not just as a coding assistant, but as a product that helps Anthropic learn about AI safety while also becoming a major business driver.
Why Claude Code Became So Big
- Claude Code began as a way to test and understand model safety in real-world use, not just in lab evaluations.
- Anthropic’s models got especially strong at coding early, making coding a natural first product.
- Cherney says the biggest driver of Claude Code’s explosive growth is model improvement, not just the harness/interface itself.
- Anthropic “dogfoods” the same API and public models its customers use, so product gains and model gains reinforce each other.
Safety, Security, and Guardrails
A major theme is how Anthropic tries to make powerful agents safe enough for real work:
- Prompt injection is one of the core threats discussed.
- Example: a model reads a webpage, finds malicious instructions embedded in the page, and follows them instead of the user.
- Anthropic uses multiple layers of defense:
- Alignment training
- Mechanistic interpretability / neural probes
- Sandboxing to limit file and website access
- Permission prompts and now more autonomous modes with safer defaults
- Cherney emphasizes that Anthropic continuously red teams and penetration tests its systems.
How AI Coding Actually Works
Cherney explains why coding models can look wrong at first, then fix themselves:
- Coding is a creative process with iterative feedback, not a one-shot execution.
- Claude Code gets better when it can:
- write code
- run tests
- inspect results
- iterate
- He compares this to a sculptor working with more or less feedback:
- the more the model can “see” and test its work, the better the result.
- This is why code generation can feel magical: the model doesn’t just write, it debugs itself.
The Future of Software Engineering
The conversation argues that AI is changing what software engineers do:
- Cherney says Anthropic already writes most of its code with Claude Code.
- He predicts the role of engineers will split into new categories:
- Prototypers
- Builders
- Maintainers
- Scalers
- “Sweepers” / “Perfectors” who polish the product and remove rough edges
- In this view, everyone on a team may increasingly write code:
- designers
- PMs
- engineering managers
- The bottleneck shifts from typing code to judgment, direction, and iteration.
Interfaces Are Changing
Claude Code started in the terminal, but Cherney says that’s no longer the main interface:
- Claude Code now works through:
- IDE extensions
- desktop apps
- mobile apps
- Slack
- He describes the future as less “file system” and more conversation
- AI joins discussions proactively
- it can pull data, generate mockups, and collaborate across teams
- This creates a new kind of interface where the agent is embedded in the workflow rather than being a separate tool.
Enterprise Adoption and Business Impact
The interview spends a lot of time on how companies adopt these tools:
- Adoption tends to happen in stages:
- individual experimentation
- limited team use
- broader deployment with controls
- deep workflow integration
- Enterprises care a lot about:
- spend controls
- permissions
- privacy
- sandboxing
- Cherney argues that the biggest productivity gains happen when companies put the AI at the center of the process, not just as a side tool.
- He compares this to the PC era:
- companies that digitized the whole workflow gained the most
- companies that merely added computers on the side did not
Migration, Legacy Systems, and COBOL
One of the more striking examples is code modernization:
- Claude Code is used for large-scale code migrations, including legacy languages like COBOL.
- Cherney says AI can now do work that used to require:
- multiple engineers
- long timelines
- substantial operational risk
- He cites an internal-style example of a major language migration that was done far faster and cheaper than traditional methods.
Are Programming Languages Becoming Less Important?
Cherney’s view:
- Programming languages still matter, but less than before.
- As models improve, the syntax and language itself matter less because the model can translate intent into working code.
- He suggests this could lead to:
- fewer constraints from language choice
- more experimentation
- even a “Cambrian explosion” of new languages and computing approaches
Big Picture Takeaways
- Claude Code is both a product and a safety instrument for Anthropic.
- The biggest improvements are coming from better models, not just better prompting or UI.
- AI coding is moving software work from typing code to supervising, shaping, and validating systems.
- Enterprise value comes from deep integration into real workflows, not surface-level use.
- The future likely includes AI that is increasingly proactive, conversational, and embedded in day-to-day work.
Notable Insights
- “The thing about coding is it’s not black and white — there are many shades of gray.”
- “We’ve gone from writing code directly, to talking to a model that writes code, to models coordinating with other models.”
- “The biggest productivity gains come when you put the AI at the center of the workflow.”
- “Prompting is getting in the way; the next step is longer-running agents, routines, and loops.”
Bottom Line
This episode presents Claude Code as a symbol of where software is headed: more autonomous, more collaborative, and more deeply embedded in the workplace. The main message is that AI coding tools are no longer just autocomplete—they are becoming the operating layer for how software gets built, debugged, migrated, and maintained.