Unisson Wants to Clone the One Person Who Actually Knows Your Product

Every software company keeps a small, unofficial priesthood. It is the two or three people who genuinely understand how the product works - the ones whose names get typed into a Slack channel when a customer's migration stalls at 11 p.m. Sales can close the deal. Marketing can write the case study. But when a Fortune 500 buyer needs the thing configured, integrated and actually running, the whole operation narrows to a handful of experts. Unisson, a two-person company out of Y Combinator's Winter 2026 batch, is built around a single observation: that priesthood does not scale.

The San Francisco startup makes AI agents that behave less like a chatbot and more like a new hire who reads fast. Point one at a client's product and, the company says, it learns to use the software in roughly 15 minutes - not by digesting a manual, but by exploring the interface and the codebase directly. Then it goes to work: onboarding customers, running migrations, wiring up custom integrations, auditing account health, and shepherding the change-management chores that usually pile up on a few overloaded desks.

01 / THE BOTTLENECKThe unglamorous part of enterprise software

In B2B software, the sale is rarely where deals die. Deployment is. A signed contract still has to survive onboarding, configuration, data migration, troubleshooting and the slow grind of teaching a customer's team how to use a new tool. That work demands specialized product expertise, and product expertise is expensive, scarce and hard to hire. When it bottlenecks, time-to-value slips and cost-to-serve climbs - two numbers that quietly decide whether a software business is healthy.

Unisson's pitch is aimed squarely at the people who feel this pain: implementation teams, technical customer success, sales engineers, forward-deployed engineers and professional-services leaders. These are the roles asked to be technical, fast and everywhere at once. The company's framing is that an always-available AI subject-matter expert is a better answer to that demand than another round of hiring.

02 / HOW IT WORKSReading the code, not the manual

The mechanism is the interesting bit. Most tools that promise to help customer success are trained on documentation and support tickets - which means they are only as current as the last person who updated the wiki. Unisson's agents instead explore a product the way a curious engineer would: clicking through the interface, reading the codebase, and pulling context from meetings, call transcripts and Slack. That is how the company arrives at its 15-minute claim, and why it argues its agents can execute tasks rather than just describe them.

The product splits into two named agents. Runner is the one that does things - it executes tasks directly inside a client's product and coordinates with other agents when a job has many steps, like a migration. Explorer is the librarian - it keeps a real-time knowledge base and answers questions grounded in live customer data, so the documentation updates itself instead of rotting. There is also an API, letting a client's own product and engineering teams embed Unisson's expertise into agents of their own.

---

03 / DIFFERENTIATIONDoing, not answering

The crowded lane next door is AI customer support - platforms that resolve tickets and answer questions. Unisson positions itself one step deeper into the workflow: instead of replying to a customer, its agents go into the product and complete the task. That is a meaningful distinction. A support bot that explains how to run a migration is useful; an agent that runs the migration, with a human watching, is a different category of help.

04 / BUSINESS MODELYou pay when the agent works

Unisson charges on usage rather than per seat. Clients are billed only while the agents are actively performing tasks; idle agents cost nothing. For a buyer, that is more than a billing footnote - it ties the vendor's revenue to work actually done, and lowers the risk of paying for licenses that sit unused. The company also leans on enterprise-grade trust signals early: SOC2 compliance, encrypted data and audit logging, the table stakes for selling into large organizations even as a two-person team.

05 / THE FOUNDERSDeployment DNA from AI and robots

Unisson's two founders arrived at the same problem from opposite corners of applied AI. Varun Mathur, the CEO, previously led product, engineering, growth and visual-language-model research at Ambient.ai (itself a YC company). He is credited with building a video-understanding model reported to outperform Gemini 2.5 Pro and sold into Fortune 5 accounts. Tom Achache, the CTO, ran Perception and vision-language agents at Chef Robotics, where he helped put dozens of robots into live production kitchens.

Both backgrounds share a theme: getting AI to work reliably in messy, real-world deployments rather than in a demo. That is the same challenge customer success presents - every client's product, data and workflow is a little different, and the work only counts if it actually runs.

06 / WHERE IT FITSA vertical agent in a broad field

Unisson lands in the fast-growing category of vertical AI agents - software that does a specific job rather than offering a general assistant. It brushes against horizontal AI support platforms and against the old alternative of simply staffing up professional services. Its wager is that the winning approach is neither a generic chatbot nor more headcount, but an agent that knows one company's product cold and works in the tools teams already live in. As an early-stage, two-person company, the customer roster is small and undisclosed; what it has so far is a sharp thesis, a named product, and founders who have shipped hard AI before.