Claude Code’s Architect: Why AI Cost Comparison is Misguided

Posted on by

(SeaPRwire) –

By: Christian Pierce
Five years after its founding, Anthropic is gearing up for a potentially massive IPO. Claude Code, its coding agent, already has an annualized revenue run rate over $2.5 billion. At the Brainstorm Tech conference, Boris Cherny, the mastermind behind Claude Code, shared some crucial insights.

Cherny emphasized that the way we view AI costs needs a major rethink. Early on, some Claude Code customers were hesitant due to the price compared to cheaper AI subscriptions. But after six months, they realized it was different. Cherny advised not to compare its ROI to legacy coding tools. Instead, look at what it would cost if an engineer did the work. He gave an example of a developer who rewrote a codebase in six days, a task that would’ve taken a year before. Salesforce, Ramp, and Airbnb have seen similar results. His second tip was to use internal pilots to evaluate adoption. Measure the differences in speed, security, and output quality between a team using Claude Code and one not using it.

Cherny manages a large number of AI agents daily. He pointed out that in AI-augmented workflows, the bottleneck is always changing. At Anthropic, once code generation was automated, code review became the new constraint, and they automated that too. He also learned that intellectual humility is important when working with AI. And at Anthropic, Claude is at the center of everything, even for tasks like filing expense reports.

Cherny’s hiring criteria for AI-native teams are interesting. He prefers generalists over specialists, low-ego collaborators, and empiricists who trust customer data more than internal beliefs. He treats everyone on the team like a CEO, with the job of talking to customers to figure out what to build next. This has been a humbling experience for the engineering process.

Author bio: Christian Pierce, chief financial columnist and markets commentator.