Most consumer apps have a strange problem. They know an enormous amount about their users - every tap, every session, every abandoned checkout - and then they email all of those people the same thing at the same time. The data is sitting there. The message is a guess. Polymorph, a San Francisco company in Y Combinator's Winter 2026 batch, was built to close that gap.
The company's own line for it is blunt: "Most teams already have the data, but messaging is still guesswork across different users and channels." Polymorph's answer is to treat personalization not as a campaign someone hand-builds, but as infrastructure - something you plug in, the way you plug in a payments API or an analytics SDK.
What it actually does
Polymorph takes the behavioral signals a product already produces and turns them into live user profiles that update in real time. From there, AI agents decide the three things that usually get decided by a rushed growth marketer on a Friday: when to reach someone, on which channel - push, in-app, email, or SMS - and with what content. Then, and this is the part incumbents tend to fumble, it reports clear attribution on what actually drove the conversion.
How a signal becomes a message
Underneath sits an experimentation engine. Rather than testing one hypothesis and waiting two weeks for significance, Polymorph's agents run thousands of experiments continuously, letting the product adapt per user instead of settling on a single averaged-out "best" version. The name is the thesis: a polymorphic internet, one that changes shape for each person who touches it.
The early numbers
Polymorph is young - founded in 2026, still measured in employees you can count on one hand - so the evidence is early rather than exhaustive. But the numbers it does share are the kind founders lead with. It reports supporting more than 3.5 million users across its customers, and one early customer saw a 3.6x lift in reactivating dormant users, the single metric the company highlighted when it surfaced through YC.
Who is behind it
Polymorph has three co-founders, and their resumes explain the choice of problem. Andrew Sy, the CEO, built Scale AI's ML inference infrastructure - a system he describes as handling more than 10 million calls a day - and did stints at Meta and Opal Security. David Nie spent roughly six years at Meta in SMB Ads, where he worked on 200-plus experiments. Manas Purohit's path runs through Meta, Palantir as a forward-deployed engineer, and Gusto.
The through-line is people who have already built the expensive, in-house version of what Polymorph now sells. Personalization at the scale of Meta, TikTok, or Google takes years and whole organizations. The bet here is that three people who lived inside those machines can hand the same capability to everyone else through an API.
Where it fits
The lifecycle-messaging aisle is not empty. Braze, Iterable, Customer.io, OneSignal, and MoEngage all move messages across channels; Statsig, Optimizely, and Amplitude all run experiments. Polymorph's wedge is that those tools mostly hand you a console and expect a team to configure it. Polymorph's framing is that the agents do the configuring - the experimentation and the per-user targeting run automatically, and the human just reads the attribution.
Whether that framing holds up under real enterprise load is the open question every young infrastructure company faces. But the target is clear: consumer and self-serve teams who want Duolingo-grade lifecycle messaging without hiring a Duolingo-sized growth org.
The reference pointsOne useful way to read Polymorph is as an attempt to package taste. Great growth teams have judgment - a feel for when a nudge lands and when it annoys. Encoding that judgment into infrastructure, so a two-person startup gets it on day one, is the whole game. It is also why "attribution on what drove conversion" matters more than it sounds: a personalization tool that can't tell you why it won is hard to trust twice.