Devin-kun: Cognition opens Tokyo office, bets on Japan

Cognition AI is opening a Tokyo office and betting that Japan's shrinking engineering workforce, legacy code backlog, and government-led digital modernization push will turn into the largest sustained coding agent market outside the United States. The company told Fortune that Japan was already one of its most engaged markets globally by user count, before the formal Asia push started.

Devin, Cognition's autonomous software engineering agent, was already being adopted inside Japanese government and enterprise before the company formalized its local presence. Sapporo's city government needed to modernize more than one million lines of legacy code to comply with a national IT mandate, a job Cognition's president Russell Kaplan estimated would have taken roughly 200 engineering months using traditional staff augmentation. Working with Devin, the Sapporo team reportedly finished the same work in about a quarter of the time. The pattern is the one Japan has been signaling it would adopt since the country's Ministry of Economy, Trade, and Industry (METI) projected a shortage of about 789,000 software engineers by 2030, and the country is now the first concrete proof point that AI coding agents can absorb structural workforce gaps at municipal scale.

Devin-kun and the Japan-first rollout

Devin had already gone quietly viral in Japan before the Tokyo office opened, and the user community settled on a Japanese honorific for the agent that tells you something about how the technology is being framed culturally. "There was a debate of what the correct honorific for Devin was," Kaplan told Fortune, referring to the suffixes attached to names in Japanese to signal social hierarchy. "What the community settled on was Devin-kun," with the familiar "kun" suffix that Japanese speakers use for younger male colleagues, a friend, or a peer rather than a tool. The naming choice matters because it shows Devin is being treated as a coworker inside Japanese engineering teams, not as a piece of software, and that framing is going to be the one the rest of Cognition's Asia expansion has to work against.

Cognition's Tokyo office opened in April, with Singapore set to follow as the Asia-Pacific headquarters later in 2026. The geography is not a marketing decision. Japan's working-age population is projected to decline by more than 30 percent between now and 2060, the country has the world's oldest population with nearly 30 percent of residents over 65, and the programming shortage is the load-bearing constraint on the country's digital modernization plans. OpenAI and Anthropic both opened their first international offices in Tokyo, Microsoft, Alphabet, and the other hyperscalers have committed billions to Japanese data centers, and Japan was the second country to secure access to Anthropic's Mythos model after a U.S. export-control order, with three of Japan's largest banks granted entry through the Project Glasswing program before access was shut off again in mid-June. The country is the rare U.S. AI friendly market that is also large enough to matter, and Cognition is the first autonomous coding agent vendor to make the bet explicit.

The bet extends beyond coding. Devin operates as a full software engineering teammate, given a task it codes, debugs, and deploys inside the tools an existing team already uses, and is one of the earliest examples of an "AI employee" pattern that integrates with workplace tools like Slack, Linear, and Jira in a way that does not require the kind of constant prompting the first wave of AI coding assistants needed. In late May, Cognition raised more than $1 billion in a new funding round that valued the company at $26 billion, more than double its September 2025 valuation, with annualized run rate at $492 million, up from $37 million a year earlier. The cross-link to the open-source toolkit Workato Labs shipped for AI coding agent operations and to the best AI models for coding in 2026 resource page puts Cognition's Japan-first bet in the same conversation as the rest of the coding agent infrastructure stack, where governance, model choice, and run-time environment are all being unbundled from the agent product itself.

What Japan's coding agent bet means for the rest of the world

The Sapporo precedent is the part enterprise buyers will care about most. A municipal government with a million-line legacy code backlog, a national compliance mandate, and a constrained engineering hiring market was able to take a project that would have taken 200 engineering months down to roughly 50, and the work was not a demo or a pilot but a live modernization program with public sector accountability. The pattern is the one enterprise AI buyers have been waiting for since the first wave of coding assistants: measurable productivity gains on real legacy code, not synthetic benchmarks, with the cost of the project falling into a range that a city budget can absorb. The Devin-kun honorific suggests the productivity pattern is also producing a workflow pattern, where the agent is being treated as a peer rather than a tool, and that is the part most enterprise change-management programs have not yet figured out how to design for.

The global stakes are highest in the markets that have sold themselves as software engineering service hubs. Shares in India's four largest IT services firms, Infosys, Wipro, Tata Consultancy Services, and HCLTech, have each fallen between 30 and 40 percent over the past 12 months on fears that AI agents will perform the same back-office coding work at a fraction of the cost. Kaplan does not read it as a wipeout for Indian engineering, but the read is more nuanced than the consensus. "On the ground in India, the job of an engineer can become more fun and impactful," Kaplan told Fortune. "Suddenly you have someone who has been working by themselves on a specific part of a project, and they're getting a promotion where they have a whole team of AI agents working for them." The model is a senior engineer with an agent team, not a junior engineer replaced by an agent, but the ratio of work done per engineer is the part that is going to put pressure on IT services pricing through 2026 and 2027.

The compute-arbitrage story is the second-order effect. Cognition's demand for inference compute is doubling roughly every seven weeks, and a geographically distributed team means compute that is sitting idle on the U.S. West Coast at 3 AM local time is productive when it is 3 PM in Tokyo. "When people are at work in Japan, people in New York are asleep," Kaplan said. "There's a lot of efficiency you get as an AI company working that way." The model will look familiar to anyone who has followed the global offshoring economics of the last 30 years, but the cost curves are different. An engineering hour in Bangalore was cheaper than an engineering hour in San Francisco; an inference minute in Tokyo is not cheaper than an inference minute in California, it is just a different minute on the clock, and the same inference minute that is wasted in California at 3 AM is the inference minute that is on the critical path for a Japanese municipal government. The economic arbitrage is on time, not on cost, and the AI companies that learn to staff across time zones the way the global services industry staffed across labor markets are the ones that are going to win the 2026 enterprise AI buildout.

How Cognition is positioning beyond Japan

Cognition is not putting all of its Asia expansion chips on Japan. The company has launched what it calls an Applied AI Engineering program in Malaysia, identifying engineers who excel at directing agents and training them to teach entire teams how to work effectively with AI. Kaplan described the engineers his team encountered in Kuala Lumpur as among the most skilled in the world at managing AI agents, and Malaysia's combination of a large English-speaking talent pool, lower operating costs, and proximity to the rest of Southeast Asia makes it the natural regional hub for the engineering side of the bet. South Korea and Australia are also under evaluation for Asia-Pacific expansion. The Malaysia and Singapore moves are also the part of the strategy that the rest of the agent ecosystem is going to have to match, because the global services industry has spent 20 years building engineering centers in those markets and the agent vendors are now competing for the same talent with a different pitch: instead of replacing you with someone cheaper, we make you ten times more productive.

The investment math behind the bet is the part that matters for the rest of the agent ecosystem. Cognition's $26 billion valuation and $1 billion raise put the company on a similar footing to the model labs, even though Devin is a single product rather than a foundation model platform. The bet is that the agent layer is going to capture enough of the value created by AI coding that the agents are worth more than the models they are built on, and the Japan-first expansion is the test of whether that bet is correct. If Sapporo-style municipal deployments scale, the enterprise coding agent market is large enough to support multiple Cognition-tier companies, and the geopolitical alignment between Japan and the U.S. on AI policy, with Japan's preference for U.S. AI over sovereign AI alternatives, is the part of the bet that the model labs cannot make on their own. The full Fortune article on the Devin-kun Japan launch was published July 3, 2026, and Cognition's Tokyo office is now open with the Singapore APAC headquarters set to follow later this year.