The Accident That Made One AI Company Uncatchable

Every AI coding tool in 2024 worked the same way. You copied a piece of code, pasted it into a chat window, and asked for help. The AI only saw what you showed it. It was smart, but blind. Like handing a doctor your symptoms on a napkin and expecting a diagnosis.

Then an engineer at Anthropic named Boris Cherny did something nobody had tried. He gave Claude access to an entire codebase. Not a snippet. Everything. The folder structure, every source file, the dependencies, the tests, the configuration. The model could suddenly see how a project fit together, trace how functions connected, understand what the team was actually building.

The result was immediate and dramatic. Boris shared the prototype with a few colleagues. Within days, half of Anthropic’s engineering team had adopted it. CEO Dario Amodei asked if engineers were being forced to use it, because the adoption curve was vertical. They were not. People just kept telling each other about it.

What Boris had started was Claude Code. The insight behind it was not a breakthrough in model intelligence. It was context. Give a powerful model access to the right information, exactly what it needs to do the job, and the quality of its work transforms. By April 2026, that insight had helped push Anthropic past OpenAI to $30 billion in annual run rate and a valuation in the hundreds of billions.

The Most Powerful Flywheel in Tech

Claude Code did something no one anticipated. It made Anthropic’s own engineers dramatically more productive at improving Claude itself. Better coding tools produced a better model. The better model produced better coding tools. And the cycle kept accelerating.

That self-improving loop produced Mythos.

Anthropic revealed Mythos in early 2026 as a model so capable it discovered a seventeen-year-old security vulnerability that automated tools had scanned five million times without catching. It scored 73% on expert-level hacking tasks that no previous AI model could complete at all. Anthropic is restricting access to Mythos for security reasons, and the concern is legitimate.

But there is no realistic scenario in which Anthropic’s internal teams are not already using Mythos themselves to build whatever comes next. No other AI team has access to a model this powerful. At Davos in January, Dario Amodei described “an emerging feedback loop in which models help build their successors.” He was describing what was already happening inside his company.

Mythos is building its own replacement. Every turn of the cycle widens the distance between Anthropic and everyone chasing them.

$60 Billion Says the Competition Believes It

Four people run the companies that will define artificial intelligence: Dario Amodei, Sam Altman, Elon Musk, and Demis Hassabis. All four understand that building AI that builds better AI is now the race that matters. Their moves in the last 90 days make that clear.

Sam Altman dropped everything. OpenAI ditched what insiders called its “side quests” to focus entirely on coding tools and enterprise, a direct response to Anthropic pulling ahead. Elon Musk is buying Cursor for $60 billion because xAI cannot build a competitive coding AI fast enough. He said publicly that his own engineering team needed to be rebuilt from scratch. Demis Hassabis knows exactly what is at stake. He has told interviewers that Google needs to move at “startup pace,” and his actions confirm the urgency: Google DeepMind is pouring resources into coding models while Google’s own engineers use Claude Code despite a company ban on rival tools.

All four leaders see the flywheel. All four know what happens if one company pulls far enough ahead that its models are building better successors than anything the competition can produce. That is the point where the race stops being competitive. Anthropic may already be there, and the other three may already be too late to catch up.

The Second Race That Could Decide Everything

Context access is what created Claude Code. It was the insight that a model with the right information outperforms a smarter model without it, and that insight launched the flywheel that put Anthropic in this position. Then Anthropic doubled down.

They open-sourced a protocol called MCP that lets any data source connect to any AI model. Then they built features on top of it: persistent workspaces, folder access, third-party connectors that pull context from apps across your digital life. By the time competitors noticed, thousands of integrations existed and they all worked with Claude first. Give away the platform, let the ecosystem build itself, harvest the network effects.

All four CEOs see this race too. But context infrastructure does not work like model training. You cannot build it faster by buying more GPUs. You need developers to integrate their services one by one. You need users to adopt those integrations and trust them with personal data. Every new connection requires a company to discover your protocol, evaluate it, commit engineering resources, and ship. That process runs on relationships and reputation, not compute. It takes years, not training runs. The three companies chasing Anthropic on models will throw everything they have at closing that gap. They might succeed. But they are also years behind on context, and closing that kind of ecosystem gap is something money alone cannot buy. When Musk needed a coding AI he could not build, he wrote a $60 billion check for Cursor. The question is whether someone will do the same for context before it is too late.

An Accident, Then a Strategy

Boris Cherny did not set out to give Anthropic a decisive lead. He gave a model access to a codebase because he was curious what would happen. But Anthropic recognized the significance. The model transformed the moment it had the right context. That was the lesson, and they applied it far beyond coding.

The coding flywheel gave them the best model. The context strategy gave that model something to work with. Together, those two moves have put Anthropic in the lead in what may be the biggest technology race the world has ever seen, with a valuation heading into the trillions.

Context was at the core of that success. It will be at the core of whatever comes next.