The problem isn't that artificial intelligence agents can't do the work. It's that they don't know how the work gets done.
That's the hypothesis driving Trace, a London startup that just raised $3 million to build what its founders call a "context engineering" layer for enterprise AI. The round, announced February 26, 2026, drew backing from Y Combinator, Zeno Ventures, Transpose Platform Management, Goodwater Capital, and Formosa Capital, along with a handful of angels who came in through a WeFunder special-purpose vehicle.
The pitch is straightforward, if ambitious: most AI agents stumble in real companies not because the underlying models are weak, but because they lack the institutional memory that human employees accumulate over months or years. They don't know which Slack channel handles which kind of request, or that Jane in customer success prefers Notion while the engineering team lives in Jira. Trace wants to solve that by building a knowledge graph that maps relationships across people, projects, and tools—then generating workflows that can hand tasks off to either AI or humans, depending on what makes sense.
Tim Cherkasov and Artur Romanov, co-founders who now serve as CEO and CTO respectively, have already seen some early validation. As of February 26, 2026, the company reported more than 550 active workflows running across its customer base, with somewhere between 10 and 14 percent of tasks currently routed to AI agents. The rest still go to humans.
Most of the use cases cluster in back-office territory: account management, customer onboarding, document processing, HR screening, sales ops. Not the sexiest work, perhaps, but high-volume and ripe for automation.
A Crowded Field, and Fast-Moving Incumbents
Trace is hardly alone in chasing this opportunity. Microsoft unveiled Scout and a suite of agent-first tools at its Build conference in early June, emphasizing—you guessed it—agents with organizational context. ServiceNow has positioned itself as an "AI control tower" for governed autonomous work, deepening ties with Anthropic and OpenAI over the spring. Anthropic itself pushed into enterprise agents with departmental plug-ins around the same time Trace announced its seed round.
Then there are the developer-first frameworks: LangGraph, CrewAI, and others targeting orchestration from the ground up. Perplexity launched "Computer," a multi-model agent for complex workflows, in a similar timeframe.
The convergence is telling. It suggests the market sees real demand for tools that can bridge the gap between powerful AI models and messy organizational reality. It also raises a harder question: can a two-person founding team—LinkedIn suggests Trace employs somewhere between two and ten people—carve out defensible territory against platform giants?
The company did generate buzz early on. It topped Product Hunt as the number-one product of the day, week, and month back in August 2025, roughly nine months before the seed funding. A full-stack engineering role posted on LinkedIn signals active hiring, though at this stage it's hard to say how quickly the team will scale.
What Comes Next
Trace's roadmap centers on expanding its agent suite, deepening integrations with workplace tools, and building a third-party SDK so external agents can plug into its workflow engine. The founders are also eyeing multi-team deployments to move upmarket—a logical next step given the traction in back-office functions.
Whether Trace can establish itself as the connective tissue for enterprise AI remains an open question. But the investor roster, the competitive landscape, and the persistence of the underlying problem all point in one direction: this won't be the last seed round aimed at making AI agents less alien to the way companies actually work.
For now, Cherkasov and Romanov are betting that context is the missing piece. If they're right, the platform play isn't just orchestration. It's translation.
