Ian Clarke spent the better part of two decades building systems designed to spread power around. First came Freenet, the peer-to-peer network meant to withstand government censors. Now he's back with something equally ambitious, if less obviously rebellious: a platform that wants to make negotiations fairer by letting math do the talking.
Mediator.ai, Clarke's latest venture, sits at the intersection of game theory and generative AI. The pitch is deceptively simple. Instead of haggling over equity splits or partnership terms until someone blinks, why not feed your preferences into an algorithm and let it find a deal both sides can live with? Traditional mediation costs money and drags on. DIY bargaining stalls when egos get involved. Clarke's answer: outsource the hard part to a machine.
Whether founders will actually trust a black-box optimizer to referee their most sensitive conversations—well, that's the bet.
The Mechanics of Mathematical Fairness
Here's how it works in practice. Each party chats privately with an AI assistant, answering what Clarke describes as hundreds of pairwise questions. Would you rather have this clause or that one? More equity or more control? The system doesn't ask you to quantify your priorities explicitly—instead, it infers an approximate utility function from the pattern of your choices.
Once Mediator.ai understands what you value, it runs a genetic algorithm to evolve draft agreements. Clauses get swapped, combined, tested. The goal is what game theorists call the Nash bargaining solution: the deal that maximizes the combined gains for both parties over walking away empty-handed. The platform uses Lua scripts as "mutators" to tinker with contract language until the optimizer settles on a proposal scored against both sets of preferences.
A single negotiation session, according to an early blog post from the company, runs about $2 in LLM infrastructure costs—though what users ultimately pay remains unpublished. The platform is live now, with public pricing yet to be announced.
From Bakery Splits to Buyout Clauses
The examples Clarke has posted suggest he's targeting a wide swath of messy, low-stakes disputes—the kind most people muddle through without hiring counsel.
Take the bakery partnership case. Two founders needed to reset their equity after one had stepped back. Mediator.ai proposed a 60/40 split with an earn-back provision: the partner holding 40 percent could reclaim an additional 10 points by working full-time for six months or by forgoing $24,000 in distributions. It threw in a management salary and a shotgun clause for future exits.
Another scenario involved a house purchase between two buyers with unequal stakes—70/30. The platform suggested a $10,000 reserve transfer to balance post-closing cash, an income-based mortgage split, protections for caregiving duties, and buyout mechanics with defined timelines.
Then there's the freelance scope creep problem, which anyone who's worked on contract knows intimately. Mediator.ai drafted a $6,500 addendum payable at project launch, carved out Phase Two work from the original deliverables, and formalized a change-order process the client's CFO had been asking for anyway.
These aren't earth-shattering deals. Roommate agreements, parenting plans, chore schedules. But Clarke's long game includes prenups, business partnerships, even multi-party negotiations—terrain where stakes climb and emotions run hotter.
The Man Behind the Algorithm
Clarke, who operates out of Austin, made his name in the late 1990s with Freenet, a censorship-resistant network that became a cause célèbre among digital rights advocates. His new venture, structured as Mediator.ai LLC, appears to have launched in 2025, with SignalHire listing Clarke as founder starting in May of that year.
The shift from decentralized information networks to negotiation platforms might seem abrupt, but Clarke sees a through line: both projects are about distributing power more evenly. On Hacker News, where the launch announcement drew 145 upvotes and 74 comments recently, Clarke clarified that the platform proposes drafts and facilitates rather than arbitrates. "Both sides can reject the proposal," he wrote. The system assumes participants actually want a fair outcome—not maximum leverage.
That assumption, of course, is doing a lot of work.
What Could Go Wrong
Clarke doesn't sugarcoat the system's limitations, which is perhaps the most refreshing part of his pitch. Utilities are inferred from conversations with an LLM, which means your stated preferences might wobble depending on how you phrase things. The genetic search only improves among the candidates it generates—it can't promise a globally optimal deal. And there's always the risk of specification gaming: an agreement that scores beautifully on paper but feels wrong the moment you read it.
More troubling, the platform can't detect bluffing. Misstate your walk-away point and you can nudge outcomes in your favor. Mediator.ai also assumes both sides want a deal and neither wants to feel outmaneuvered. A counterparty looking to extract maximum advantage could exploit the data each participant shares.
Then there's the legal fine print. Drafting an agreement isn't the same as making it enforceable. The site includes the standard disclaimer: Mediator.ai does not provide legal advice; consult an attorney for anything binding. Fair enough. But it raises the question of whether a mathematically optimal contract is worth much if no one's checked whether it would hold up in court.
What Happens Next
Mediator.ai is accessible now through the company's website, with a prominent call-to-action inviting users to try the platform. Early coverage has been modest but intrigued—Gigazine ran a piece shortly after launch, walking through how the system translates natural-language inputs into utility functions and maximizes the Nash product.
No funding announcements have surfaced. No investor names. The company lists a contact email but hasn't published details on pricing tiers or which LLM providers power the backend. Clarke has hinted at extensions for multi-party negotiations but hasn't released technical documentation on how that might scale.
For now, the value proposition is clean: negotiate with math, draft with AI, skip the part where someone walks away feeling fleeced. Whether startup founders—who tend to trust their own instincts above all—will hand over sensitive equity talks to an algorithm remains very much an open question.
Clarke, though, seems to think they're exhausted enough by adversarial bargaining to at least give it a shot. Perhaps he's right. Or perhaps there's a reason most negotiations still happen the old-fashioned way, with all the inefficiency and hard feelings that entails.
