Polywork raised $40M and closed. Read.cv was beautiful and closed too.

If you're going to build a professional portfolio product, the responsible thing is to look at what happened to the last people who tried.

Polywork raised more than $40M, won a great deal of early attention, and shut down in January 2025 - its final pivot pointing toward AI-generated personal websites, which most observers read as a search for defensibility that arrived too late. Read.cv, the closest reference point to what we're doing - a genuinely lovely product, essentially "Behance for people who aren't designers" - wound down in 2025 following its acquisition by Perplexity.

Here's what makes this worth writing about rather than ignoring: neither of them failed on design, and neither failed because the insight was wrong. Read.cv was better-looking than most products in any category. Polywork's core observation - that a job title is a terrible summary of what a person actually does - was and remains true.

They failed on business model and on network effects, with far more capital than we will ever have. That's the sentence we keep on a sticky note.

Lesson one: a beautiful profile is a feature, not a product

You can build the best-looking professional profile on the internet and still watch it sit there. Profiles are inherently passive: made once, updated when someone changes jobs, visited when someone sends a link. That's a beautiful object with no metabolism.

Products in this space usually respond by adding surface area - a feed, a jobs board, messaging, a directory - until they're competing with LinkedIn on LinkedIn's terms, which is not a fight anyone wins with a small team.

So the question we made ourselves answer before writing any code wasn't "what should the profile contain?" It was: what's the object that moves on its own?

Lesson two: the thing that spreads and the thing that returns should be the same object

Most products have to choose. The thing that acquires users (a viral artefact, a free tool) is usually not the thing that brings them back (a dashboard, a workflow, saved state). Maintaining both is expensive and they tend to pull the design in opposite directions.

The Impact Card is our attempt to make them one object:

• It's the acquisition channel - it gets posted publicly, and other marketers see it and want their own.
• It's the activation moment - it's what you make in your first session, and the session has failed if you don't finish one.
• It's the retention hook - quarters end, new results arrive, the card wants updating.

If that collapse holds, the product has a metabolism. If it doesn't, we've built a nicer version of a thing that has now failed twice, and we should find that out cheaply.

Lesson three: the durability objection is real, and it's answerable

One documented reason people under-invested in Polywork: a rational suspicion that it might not be around long enough to be worth the effort. Every small platform asking for your career history faces this, and the suspicion has now been vindicated twice, publicly, in two years. Anyone who raises it with us is right to.

The counter isn't a promise about our runway. It's making the exit door obvious:

• Data export in one click, from day one, not buried in settings.
• Cards downloadable as images you keep regardless of what happens to us.
• Deletion that actually deletes.

"Your data, always exportable" costs almost nothing to build and is one of the few honest answers available to a small platform. Anyone who tells you their startup will definitely still exist in five years is guessing.

What we're doing differently, concretely

Starting with the artefact, not the network. No feed, no likes, no followers, no directory to browse. There is no cold-start problem if the product's value doesn't depend on anyone else being here. Your card works on day one whether we have fifty users or fifty thousand.

Measuring the loop before building the platform. The number that matters is not signups. Signups measure whether our outreach worked. The number that measures whether theproduct works is: of the people who make a card, how many post it publicly without being asked? If almost nobody does, the growth loop is broken and no amount of feature work fixes it. We'd rather learn that from thirty marketers in a fortnight than from a burn chart in a year.

Choosing the flex over the credential. There are two products hiding inside this idea. One is a credential for recruiters - which self-reported numbers on an unknown site serve poorly, because recruiters won't change their workflow for unverified data. The other is a personal-brand artefact for a profession that treats its own brand as a craft. The second one is stronger, more fun, and much more likely to spread. We're committing to it, which means the card matters more than the profile.

The part we're not confident about

Three things could still be fatal, and we'd rather write them down than discover them later.

Confidentiality could cap the market. If most marketers can't publish their numbers, the addressable audience shrinks to freelancers, agency people and active job-seekers. That might be a real market. It might be too small. We're asking people bluntly rather than assuming.

LinkedIn has antibodies against humblebragging. Post the wrong framing and the card reads as cringe, which is a distribution problem disguised as a copy problem. "What I shipped this quarter" survives scrutiny; "my impact" does not. We'll be watching the reception of the first wave carefully.

Stickiness might not arrive. A quarterly update rhythm is a hypothesis, not a mechanic we've proven. Two well-funded companies have already discovered that "come back and update your profile" isn't a reason to come back.

None of that means don't build it. It means build the smallest thing that tests the loop, tell people the truth while you do it, and stay genuinely willing to conclude that the loop doesn't work. That's the plan. We'll keep posting what we find, including the parts that go badly.