Claude Code, 2025: the bottleneck was never the typing
Boris Cherny had been at Anthropic for less than a month when, in September 2024, he wired Claude to his terminal and asked it to fix a bug. The next morning, three colleagues were using the same tool. He had not told anyone about it.
That is roughly how the history of code editors tends to work: someone builds the thing they personally need, and the engineers around them quietly steal it.
Cherny, a former Meta engineer, had joined Anthropic sensing what he called a “product overhang” — a gap between what the underlying model could do and what existing tools actually let programmers ask of it. GitHub Copilot and Cursor had learned to read a codebase and autocomplete with genuine context. But they still waited for the programmer to lead. The programmer was still the one who identified the relevant file, scoped the change, decided when to run the tests, and interpreted what the failures meant. The AI made individual keystrokes smarter; it had not touched the workflow underneath.
What Cherny built had no graphical interface, no file tree, no menu bar. It ran in a terminal. You described what you wanted — “fix the failing test in the payment module,” “migrate this endpoint to the new schema” — and Claude Code searched the codebase, reasoned about what needed changing, edited the appropriate files, ran the relevant commands, read the output, and iterated. If it was about to do something risky, it stopped and asked. Everything else, it simply did.
The research preview launched February 24, 2025, bundled alongside Claude 3.7 Sonnet. General availability followed May 22, 2025 at Anthropic’s Code with Claude conference, with Cat Wu leading product strategy. Within five days of the November 2024 internal release, half of Anthropic’s engineering team had adopted it. By the public launch, over 80 percent of the company’s engineers were using it every day. Claude Code crossed one billion dollars in annualized revenue within six months.
Cherny, later asked about his own practice, gave an unembellished answer: “I have not written a single line of code since November.”
The code editor had spent sixty years getting more elaborate — windows, menus, syntax highlighting, real-time linting, integrated debuggers, AI autocomplete that read the room. Claude Code stripped all of that away and returned to the terminal. TECO, in 1962, asked you to specify every operation in cryptic shorthand; Claude Code asks only what you want done. Between those two instructions lies six decades of work. By May 2026, more than 80 percent of Anthropic’s own production code was being written by Claude Code — the tool now generating its own successors. The bottleneck, it turned out, was never the typing. The typing was the last thing to go.
Sources
- Claude (language model) — Wikipedia — Claude Code overview, February 2025 research preview, general availability, commercial traction.
- Anthropic’s Claude Code Shows How Far AI Coding Tools Have Come — Built In — Boris Cherny background, launch dates (February 24 and May 22, 2025), adoption metrics, $1B revenue milestone.
- Boris Cherny, Anthropic: “I have not written a single line of code since November” — Station F — Direct quotes from Cherny, product overhang framing, internal adoption story.
- What Is Claude Code? Full History, Timeline & How It Works — Taskade — Detailed timeline including November 2024 internal release, Cat Wu product strategy, May 2026 production code share.