Agentic AI coding tools are erasing the line between engineers and everyone else. In an interview with CNBC aired on 20 May 2026, Boris Cherny, Head of Claude Code at Anthropic, described a lawyer at the company who built a side-scrolling game — with no coding experience — to teach employees how to work with the legal department. Cherny framed it not as a curiosity, but as a signal of an accelerating shift: AI coding tools are turning product managers, designers, finance staff, and lawyers into active contributors to codebases, at companies from early-stage startups to large enterprises.
What AI Coding Tools Are Doing to the Traditional Engineering Role
Cherny told CNBC’s Arjun Kharpal that the pattern is now widespread across the Claude Code user base. “People that are not traditional engineers are coding,” he said. “On the Claude Code team, our designers code, our engineering managers code, our finance guy writes code, everyone writes code.” He cited fintech company Ramp as a corroborating example, noting that Ramp had reported in a blog post that its product managers were coding.
That claim checks out. Ramp built an internal AI coding agent called Inspect, designed from the ground up to give every employee — not just engineers — access to the same development environment. According to a February 2026 case study published by cloud infrastructure provider Modal, Ramp did not mandate adoption of Inspect. Within a couple of months, roughly half of all merged pull requests across Ramp’s frontend and backend repositories were being initiated by the tool, with product managers empowered to add features directly and designers iterating between intent and implementation without waiting in an engineering queue.
The system requires no local setup. A product manager or designer sends a prompt, and Inspect spins up a full sandboxed environment with all dependencies installed — the same developer setup an engineer would have, without the overhead.
Claude Code Itself Is Now Written by Claude Code
Cherny’s CNBC appearance builds on a broader narrative that has been developing at Anthropic since early 2026. In March of this year, Cherny confirmed via a post on X that Claude Code is 100% written by Claude Code itself. An Anthropic spokesperson told Fortune that company-wide, the figure sits between 70% and 90%.
On CNBC, Cherny extended that data point to Anthropic’s customers. “It’s everyone from the smallest startups and indie devs to the largest companies,” he said. “100% of their code is starting to be written by Claude Code.” He described the core productivity shift as a 10x acceleration: companies using Claude Code in sophisticated ways can build the same product, but do it roughly 10 times faster than before — unlocking a backlog of ideas that had previously been too expensive to pursue.
The scale of adoption supports this. On Lenny Rachitsky’s podcast in February 2026, Cherny noted that Claude Code now accounts for approximately 4% of all public GitHub commits globally, with daily active users having doubled in the preceding month. SemiAnalysis has projected that figure could reach 20% of all GitHub commits by the end of 2026.
AI Coding Tools and the Blurring of Job Titles
The implications for organisational structure are significant. Cherny told CNBC he is already observing the emergence of a new kind of worker — one who is neither a traditional engineer nor a non-technical contributor, but something in between. The anecdote about the Anthropic lawyer is illustrative: the individual had no coding background, yet produced a functional, interactive application using AI coding tools alone.
This blurring of roles aligns with a broader pattern Cherny described on Lenny’s Podcast earlier this year, where he predicted the job title “software engineer” would begin to disappear, replaced by a more generalised notion of “builder.” On that same episode, Cherny noted that Anthropic’s own engineering productivity has increased 200% per engineer since adopting agentic coding workflows internally.
Ramp’s experience offers a concrete example of how this plays out operationally. According to a case study published by project management tool Linear, Ramp’s Inspect agent accelerated the company’s culture of acting promptly on product feedback, with designers starting implementations directly in Figma before handing off to engineers — reducing the lag that typically leaves small issues sitting in backlogs.
What This Means for Enterprises Evaluating AI Coding Tools
For enterprise technology decision-makers, the shift Cherny describes is not merely a productivity story. It is a structural one. When non-engineers can ship code directly, the traditional gatekeeping function of engineering teams changes — and so does the talent and tooling calculus.
A few caveats are worth noting. Ramp’s engineering team has acknowledged that frontier models still make mistakes and require human oversight for critical decisions. Building a deeply integrated internal coding agent at Ramp’s level also required substantial AI infrastructure expertise — a bar that not every organisation can clear. Cherny’s framing on CNBC was optimistic, but the operational complexity of deploying agentic coding at scale remains real.
What is harder to dispute is the direction of travel. Whether through off-the-shelf tools like Claude Code or purpose-built internal agents like Ramp’s Inspect, the assumption that coding is a specialised, gated function appears to be eroding — faster than most enterprises have planned for.
“We had a lawyer who made a little app… he has no coding experience. We’re starting to see this across the industry.” — Boris Cherny, Head of Claude Code, Anthropic, speaking to CNBC’s Arjun Kharpal, 20 May 2026
Quotes from Boris Cherny are sourced to CNBC’s “Europe Early Edition,” 20 May 2026. The excerpts were generated using speech recognition software and may contain minor errors.
