Trace builds the manager for a workforce that is half human, half AI

Ask a founder about their company's AI rollout and you tend to hear the same short story twice. First the demo, the one that made everyone in the room lean forward. Then the pilot, the one that quietly ran out of runway a month later. The model was smart. It wrote decent copy, summarized the thread, drafted the reply. It just had no idea how the company actually worked - who signs off on a refund, where the customer's contract lives, what "done" means for this particular team.

That gap, between a capable model and a useful one, is where Trace has planted its flag. The company, part of Y Combinator's Summer 2025 batch and founded by Tim Cherkasov and Artur Romanov, sells a single, unfashionable idea: the intelligence was never the bottleneck. The context was. And context is a product you can build.

01 / THE PROBLEMSmart interns, no org chart

The metaphor does a lot of work, so it's worth sitting with. An intern who is brilliant but brand new is not immediately useful. They don't know the systems, the people, or the unwritten rules. What makes them productive is a manager who assigns the right task, hands over the right background, and checks the work before it ships. Trace's argument is that the AI industry spent two years building better interns and almost no time building the manager.

The symptom shows up everywhere: agents with no shared state across systems, no memory of how the last five deals closed, no clear path from a clever experiment to something that runs every Tuesday without a human babysitting it. Cherkasov and Romanov kept hitting the same wall, and eventually decided the wall was the market.

02 / HOW IT WORKSPlain English in, running workflow out

The mechanics are less mysterious than the pitch. Trace connects to the tools a team already lives in - Slack, Jira, Notion, email, Airtable - and quietly assembles a knowledge graph of the organization: who does what, where the data sits, how a process actually moves. A user then describes a task in plain English. Trace breaks it into steps, and for each step it makes a decision that sounds simple and is anything but: can an AI agent handle this, or does a human need to?

Tasks that can be automated get handed to an agent, with the relevant context stapled to them. Tasks that need judgment, a signature, or a gut check get routed to a person, as an approval gate. Every action is logged. The result is a visual workflow that looks less like a chatbot and more like an assembly line where robots and people stand at different stations.

How a task moves through Trace

Slack · Jira
Notion · Email

→

Knowledge
Graph

→

Task
Router

AI Agent
repetitive, high-volume

Human
judgment & approval

03 / THE HONEST NUMBEROnly 14% is a feature, not a flaw

Here is the detail most AI startups would bury and Trace leaves in plain sight: across those 550-plus live workflows, only about 10 to 14 percent of tasks currently run on agents. The other 86 percent still needs a human. In a market that likes to promise the machines will do everything, that number reads almost like a confession.

It's actually the strategy. By being explicit about which slice AI can own today, Trace makes its automation measurable and, more importantly, trusted. Customers report ROI from the shift, not because a workflow went fully autonomous, but because the repetitive 14% stopped eating a person's afternoon.

WHERE THE WORK GOES TODAY

~14%

AI Agents

~86%

Humans

04 / DIFFERENTIATIONTriggers versus orchestration

On paper, Trace sits near tools like Zapier, Make, and n8n. In practice it's making a different bet. Classic automation is a chain of triggers: when this happens, do that. It's stateless and literal. It doesn't know why the task exists or whether a human should look at it. Trace starts from the knowledge graph instead, so the system has a picture of the organization before it moves a single task - and it treats humans as first-class participants rather than the thing you fall back to when the automation breaks.

Dimension	Trigger tools (Zapier, n8n)	Trace
Model of work	if-this-then-that	org knowledge graph
Humans	fallback / manual step	routed, first-class
Context	passed field by field	shared, persistent
Build step	wire nodes by hand	describe in plain English

05 / THE FOUNDERSA product manager and a Deliveroo engineer

Cherkasov, the CEO, came up as a product manager with a fintech background spanning hedge funds and data science. Romanov, the CTO, was a senior engineer at Deliveroo and a repeat founder with roots in logistics and finance. Their path to Trace was not a straight line - the founders describe it, cheerfully, as escaping "pivot hell" before the workflow-orchestration idea finally held.

The market noticed quickly. Trace became the most upvoted startup in its YC batch, took Product of the Day, Week, and Month on Product Hunt, and climbed to the top of the YC Launchpad leaderboard for Summer '25. The $3M seed, announced in February 2026 and covered by TechCrunch, came from Y Combinator, Goodwater Capital, Xeno Ventures, Transpose Platform Management, Formosa Capital, WeFunder, and angels Benjamin Bryant and Kevin Moore.

06 / THE MARKETWhere a five-person team fits

The business is straightforward B2B SaaS - seat-based subscriptions, with volume discounts reported above 50 seats - aimed at operations, client management, HR, finance, and sales teams that want to deploy AI without hiring engineers to babysit it. Early customers reach for the obvious wins first: client onboarding, HR screening, document processing, account management.

The long game is bigger and, the founders admit, more speculative. They call it a "CompanyOS" - a cognitive layer that understands how an entire organization works and directs action across every team, human and AI alike. Whether that vision arrives is an open question. What's concrete today is a five-person team, a knowledge graph, and a router quietly deciding who does the next task.