AI Agent Gwen Closes 3,500 Medical Contracts in 8 Weeks — at $70 Each

Source: 20VC with Harry Stebbings | Published: 2026-07-18T14:00:05Z

Curative's AI agent Gwen independently signed 3,500 healthcare contracts in 8 weeks, slashing per-contract cost from $1,500 to $70. Its 45-person predecessor team managed only 2,300 contracts in a full year.

The tweet was from Laura Deming: "We have COVID testing capacity. Does anyone need it?" The Deputy Mayor slid straight into her DMs.

Before that moment, Curative was a hospital sepsis management company that hadn't really gotten off the ground — seven employees, three weeks of cash in the bank. Nine months later, they had 7,000 employees, peaked at 26,000 tests in a single day, and ultimately generated $5 billion in revenue over three years.

From Cows to COVID, With a Detour Through STDs

Fred Turner is British. He moved to Silicon Valley at 19. His first company, TL Biolabs, did genetic sequencing for dairy cows — predicting which cows would produce more milk, which ones had muscle mass that might complicate calving.

The business worked. But when he went to raise a Series A, investors ran the TAM: roughly 100 million cows in the US, $15–20 per test per animal, and even if you could test every single one, the market ceiling was about $1.5 billion. Not enough for an A round.

He pivoted the core DNA testing technology to human diagnostics — first STD testing, then sepsis, a bloodstream infection that raises the mortality rate by 12% for every hour left untreated and kills hundreds of thousands of people each year. They spent two years getting a lab license, found a strategic investor who was interested, signed a term sheet, spent three weeks on legal paperwork — then the investor's CEO killed the deal, citing "competition with core business."

Three weeks of cash left.

In late 2019, he sold the lab license for $150,000 to pay off debts. Five months later, he bought another one just like it for $27 million.

The License He Sold for $150K

This isn't a metaphor about timing — it's literally about timing.

In December 2019, he was winding the company down. That license was worth about $150K without a pandemic, so he sold it at market value. In May 2020, he needed an equivalent license to run COVID tests, and bought one from a Southern California company — a lab that mostly did drug testing for sporting events — for $27 million.

Where did the money come from? Prepayment from clients. By then they already had their first testing contracts, with police and fire departments. The City of LA contract accelerated the cash cycle: after completing each day's tests, they'd send someone to City Hall the next morning to pick up that day's check — because Curative needed that money to buy the next day's supplies.

They Did What Everyone Else Said Was Impossible

Florida put out a statewide bid to test every employee at every nursing home in the state — weekly, for three months. Every other lab turned it down, saying it was impossible.

Curative bid. Won. Delivered.

Fred's read: traditional labs are optimized for efficiency. Their annual test volume grows a few percent at most, and everything is designed to maximize utilization. Ask Quest or LabCorp to 10x their capacity and their entire system breaks — because that's not what they were built to do.

Curative rebuilt the supply chain from scratch, steering clear of anything everyone else was fighting over. Magnetic bead extraction was the industry standard, but only two Chinese factories in the world produced the beads, with no capacity to scale quickly. They switched to filter plates — glass and plastic — which could be manufactured faster. Swabs came from electronics testing suppliers and were self-sterilized. Fred calls this an "orthogonal supply chain" — he admits the phrase was chosen to sound more sophisticated in sales pitches.

When the Pandemic Ended, He Had $500 Million

$5 billion in revenue, minus operating losses (every test outside the peak was money-losing), plus 2.5 million vaccine doses administered — "that was a terrible business, we lost money on every shot, but our partners required it" — and what remained was roughly $500 million, all of which went into a new company: Curative health insurance.

He always knew the testing business would end. When he hired his first employees, he told them: "This is three months of work. Don't expect more." His CFO came out of retirement to help for "six months." Six years later, he's still there.

Why Health Insurance

He looked at the lab testing industry — market cap ceiling around $30 billion. "After what we went through with COVID, I wanted to build something much bigger." They negotiated acquisitions of hospitals in Florida and Texas, then walked away. The problem was too many payers: 50% government, the rest a fragmented mess of small insurers. "They change their minds every five minutes, and you're constantly trying to satisfy everyone at once."

The conclusion: in the American healthcare system, the real power sits with payers. Control the payments, control the behavior.

Though he'll admit — if he'd known AI was coming, he would have designed the whole thing differently.

Credentialing: From $50 and Two Months, to 20 Cents and 12 Hours

Curative needs contracts with roughly 1.2 million healthcare providers across the US. The first step is credentialing — verifying that each physician holds a valid license and has no malpractice suits.

The standard process: someone sits at a computer, pulls up the medical board website, checks the license, verifies education, searches litigation databases, marks it approved. Two to three months per provider. Cost: around $50.

They built an end-to-end agent with Claude. Average completion time is now 12 hours. Cost: about 20 cents.

And this is just the beginning.

Gwen Signed 3,500 Contracts

Their network contracting team is about 45 people. Over the past year, they signed 2,300 contracts.

Earlier this year, they launched an agent named Gwen. Give her a lead — say, a general practice clinic somewhere — and she'll Google the clinic herself, pull their current reimbursement rates from transparency files, find contact emails through ZoomInfo, and start sending emails.

She follows up until someone responds. Gwen sends an average of 15,000 personalized emails per day. She's found that many providers don't respond until the ninth email — a level of persistence that's essentially impossible for a human sales rep, because sending nine follow-ups to the same person requires a kind of shamelessness that people don't have.

After a few rounds of negotiation, Gwen edits the contract terms, modifies the Word document with a Python script, opens DocuSign, and clicks sign — with Fred's own legally binding signature.

About eight weeks in, Gwen has signed 3,500 contracts on her own.

From 100 per week to 100 per day. Cost per contract dropped from $1,500–2,000 to around $70. Even if Anthropic doubled its prices, the math still works.

He's not laying off the 45-person team. They now focus exclusively on large hospital systems and physician groups — clients who want phone calls, in-person meetings, relationships. Small clinics go to Gwen.

This Year, They're Cutting 80% of SaaS Spend

They've already cancelled their Salesforce contract — saving $600K a year — and replaced it with an internal CRM vibe-coded in two months. The team used to have a full-time Salesforce admin.

Looker is gone, replaced by Snowflake. The entire Looker migration — moving all dashboards to the new platform — was originally estimated to take a full year with a team of engineers. They finished it with an agentic workflow, two or three people, in two months.

There's an internal slide listing every SaaS contract's renewal date and the name of the person assigned to call and say "we're not renewing."

What they're keeping is mostly infrastructure — Sentry, things like that, and Slack, though Slack is actually the hardest to cut because too many people have built integrations and workflows on top of it.

He says that if he'd known AI was coming, they simply wouldn't have bought the large claims systems they migrated to just two or three years ago. The APIs are nearly unusable, data is hard to extract, and they've already built their own replacement from scratch — migrating over in July.

Which Jobs Survive in Five Years

He draws a distinction between two categories: technical capability and relationships.

On the technical side, his engineers basically don't write or read code anymore — they deploy everything through Claude Code or Codex, and a senior engineer's output is dramatically higher than a year ago. But they're still hiring these people, because someone has to build the agent systems.

Relationships are also safe. Insured members want to reach a real person when something goes wrong. Large hospital systems want to meet someone in person before signing a big contract. Brokers want final presentations, dinners, rounds of golf. He expects relationship-driven sales to persist for the next five to ten years, because "the relationship itself becomes a scarce resource — the bigger the contract, the more you need eye contact and trust."

He thinks a new role will emerge soon that doesn't really exist yet: agent supervisor. You can't let agents make every decision. There's always a threshold above which a human needs to step in. But when you're operating at ten times the volume, even if only 1% of cases need human sign-off, the absolute number is ten times what it used to be. Who manages those exceptions, and how — that, he thinks, is the real bottleneck ahead.

His advice: start using these models now. Use them constantly. Because most people's mental model of what they can do is already badly behind reality.