· Fred Turner

$5BN in Revenue, 7 to 7,000 Employees in 9 Months, 206,000 Tests in a Single Day: The Craziest Story in Startups — Curative with Fred Turner

A business built on a demand shock can generate $5BN of revenue and almost no durable enterprise value; the transferable asset is not the business but the operator's proven ability to compress cycle time inside regulated industries.

hypergrowthcrisis-operationshealthcareregulated-industriesai-agentsdemand-shocksupply-chainpivot0% confidence

Why this is in the corpus

The sharpest available counterweight to the corpus's compounding/durability doctrines. Curative went 7 to 7,000 employees in nine months, peaked at 206,000 tests/day, did $5BN of revenue over three years, and ended up converting roughly $500M of that into an entirely different company. It is a rare, fully-documented case of building for a peak you know is temporary — including the deliberate decision to tell every hire the job lasts three months. It also contains an unusually concrete account of agentic labour replacement inside a regulated payer (credentialing 3 months/$50 to 12 hours/$0.20; contracting 100/week to 100/day).

Summary for skimmers

Fred Turner shut down his sepsis diagnostics company in Dec 2019 (selling its lab licence for $150K), then five months later bought an equivalent licence for $27M to enter COVID testing. Curative scaled to 206,000 tests/day, 7,000 employees and $5BN revenue by building an "orthogonal supply chain" — deliberately sourcing inputs nobody else was competing for. Margins were excellent during surges and negative during lulls because peak capacity had to be maintained through the dips. He assumed from the start it would end, hired on explicit three-month terms, and began searching for the next business in mid-2020. He landed on health insurance because the payer controls behaviour in US healthcare, funded it with ~$500M of COVID profits, and is now rebuilding it around Claude agents — credentialing to zero headcount, an agent named Gwen signing provider contracts end-to-end with his DocuSign signature, and an 80% cut to SaaS spend.

Briefing

What survives the editorial filter

This page should feel like a smart colleague already listened for you and left only the operating logic worth keeping. Not everything said in the episode makes it through.

Trust signal

Direct episode extraction

Best used for

Decision-grade retrieval metadata not yet added for this episode.

Hold lightly

No explicit downgrade reason stored yet for this episode.

Principles

Durable claims that survive beyond the speaker's biography — each with explicit limits, transferability judgment, and evidence.

Principle

Peak-capacity businesses lose money in the trough by design

If your capacity must be sized for the peak but demand is episodic, negative-margin troughs are a permanent structural feature, not an execution failure.

Model the trough, not the peak. If you cannot survive 20% utilisation for an indefinite period, the peak revenue is a mirage.

Principle

The binding constraint on AI adoption is deployment discipline, not model capability

When capability stops being the constraint, deployment throughput becomes the durable competitive variable.

Stop benchmarking models against your hardest task. Start measuring how many workflows per quarter you can actually configure, permission and ship.

Principle

Run the TAM arithmetic before you build the technology

A market ceiling computed in two minutes can invalidate years of technology work; compute it first, not at Series A.

Do the units-times-price calculation before you commit engineering years. If the ceiling cannot support your intended round, either reprice or re-aim.

Principle

Consolidation on both sides of a market destroys the market

Efficient pricing requires that both sides have alternatives; bilateral consolidation replaces price discovery with leverage extraction.

If you are entering a bilaterally consolidated market, your wedge is being the credible alternative that restores optionality — not matching scale.

Principle

Regulatory complexity is a moat against your own suppliers

Regulated complexity converts your model provider from a potential competitor into a pure supplier.

When choosing where to apply AI leverage, prefer domains where the regulatory surface makes vertical entry unattractive to your model vendor.

Principle

A buyer who needs surge capacity must overpay in the calm

When a buyer needs guaranteed surge capacity, price must include an availability premium — the peak margin is payment for the idle period.

If you are selling standby capacity, negotiate the availability premium explicitly rather than relying on peak-period volume to cover the troughs.

Principle

Efficiency-optimised incumbents are structurally incapable of 10x

Extreme operational efficiency and extreme scalability are opposing designs; incumbents optimised for one cannot pivot to the other in a crisis window.

When you need step-change output, do not assume the largest player will win. Look for whether their operating model rewards elasticity or efficiency.

Principle

Build an orthogonal supply chain — win by not competing for the same inputs

In a supply-constrained surge, the winning move is input substitution, not procurement aggression.

Before you scale procurement, ask which of your inputs is manufactured on an elastic base and which is not. Redesign the process around the elastic ones.

Principle

The second company is easier because the failure modes are already priced in

The transferable asset from a failed company is procedural competence, and it is worth more than the company was.

When evaluating a repeat founder, price the procedural library, not the prior outcome.

Principle

Cash conversion, not demand, is the binding constraint on physical scaling

In physical hypergrowth, negotiating payment terms is a financing decision that outranks price.

Treat payment terms as your cheapest source of growth capital. Daily invoicing at net-one can outperform a bridge round.

Principle

Own the layer that controls the money, not the layer that delivers the service

In fragmented systems, control of payment is control of behaviour; the payer seat has more leverage than any provider seat.

Map who funds behaviour in your system before choosing where to enter. Adjacency to the money beats adjacency to the customer.

Principle

Deliberately assemble incompatible perspectives to triangulate ground truth

Ground truth emerges from the intersection of incompatible professional lenses, not from the best single lens.

For a genuinely novel operating problem, hire across at least three professional cultures and budget explicitly for the translation overhead.

Principle

Relentlessness is the scarce sales input, and it is now automatable

Persistence was a human-limited scarce resource in sales; agents remove the limit and therefore remove the differentiator.

Audit where your funnel loses deals to giving up rather than to rejection. That gap is the first place an agent pays for itself.

Principle

Capping margin percentage makes cost growth the only path to profit growth

Percentage margin caps convert cost reduction into a profit penalty and make cost inflation the rational strategy.

When designing or evaluating a rule, test whether it caps a rate or a level. Rate caps almost always reward inflating the base.

Frameworks

Reusable systems and operating models — including when they help and when they break.

Framework

Design out the failure mode instead of controlling it — the regulatory-approval shortcut

In heavily regulated markets, an architecture that makes the failure physically impossible beats one that controls it well, because it changes the burden of proof.

In a regulated domain, ask what architecture would make your worst outcome structurally unreachable. That design wins the approval race even if it loses on elegance.

Framework

The two-axis retention filter: technical skills and relationships

Invest headcount at the two poles, technical leverage and relationship trust, and let the paperwork middle compress.

Sort every role by whether it deploys leverage, holds a relationship, or moves paperwork. Fund the first two; plan the third down.

Framework

Displacement vs expansion: the volume-target test for agent deployment

Set the agent's target as a volume multiple of prior output, not as a replacement for prior cost — that framing is what makes the economics robust to price increases.

Write the agent's success metric as a volume multiple. If your business case only survives at current token prices, you framed it as displacement.

Framework

Market-first pivoting: hold the technology core, re-aim at willingness to pay

When a technology core is sound but the market ceiling is low, search across markets by willingness to pay rather than by technical adjacency.

Write down your true technology core in one sentence, then rank candidate markets by price per unit of that core, not by how similar they look to your current one.

Signals

What appears to be shifting, for whom it matters, and what happens if you ignore it.

Signal

Model spend compounding 6x per month is the adoption inflection tell

Month-over-month inference spend growth is the cleanest available leading indicator of whether an organisation has genuinely crossed into agentic operations.

Use inference spend trajectory, not headcount or announcements, to judge whether a company (yours or a competitor's) has actually made the transition.

Signal

A counterparty going quiet through an intermediary is the earliest read on a systemic shock

Out-of-character communication behaviour from an enthusiastic counterparty is a higher-fidelity early shock signal than any public data.

Treat abrupt, out-of-character withdrawal by an engaged counterparty as a signal to investigate, not as a lost deal to write off.

Signal

Headcount guidance that goes down before it goes up

A credible growth company guiding headcount down while guiding revenue up is signalling a genuine decoupling of output from labour.

Watch for the specific shape — near-term headcount down, long-term up, with named human-proportional exceptions. It distinguishes structural change from a cost-cutting cycle.

Opportunities

Only included where there is a buyer, a real wedge, and a plausible revenue path — not vague idea theater.

Opportunity

Agent supervisor: the exception-management role that does not yet exist

Scaling agents scales exceptions proportionally, creating demand for a new role whose whole job is high-throughput judgement on agent escalations.

Build or staff the exception-adjudication layer before you scale agent volume. It is the bottleneck that appears immediately after the automation works.

Opportunity

Admin-cost compression is the only earnings growth available in a capped-margin industry

In capped-margin industries the only non-destructive earnings growth is admin-cost compression, which makes them the highest-value place to deploy agents.

Look for industries where margin is capped and volume is saturated. Their admin line is the largest available profit pool and the least contested.

Opportunity

Vertical software incumbents in regulated industries are newly displaceable

Software whose moat was encoded workflow is now displaceable in months; software that became infrastructure is not.

Classify each SaaS line as workflow-encoding or infrastructure. The first category is a rebuild candidate this year; the second is not.

Lessons still worth keeping

Useful takeaways that did not fully clear the bar for durable principle status.

Lesson

Asset value is regime-dependent — the $150K licence that cost $27M to replace

Slow-to-acquire regulatory assets are priced at their normal-regime value during a wind-down, which is exactly when that price is most wrong.

In a wind-down, separate assets that can be re-acquired from those that take years of regulatory process. Sell the first, park the second even at a carrying cost.

Lesson

Assume the boom ends and design the organisation for it from day one

Search for the next business at the top of the curve, when the boom is funding you, not at the bottom when it is not.

If your growth is driven by an exogenous shock, put a dated next-business search on the calendar while revenue is still climbing.

Lesson

A partner who contributes only the licence becomes the bottleneck at scale

A 50/50 JV where one side contributes a static asset converts that partner into a veto holder over the work they cannot do.

When structuring around a licence or permit, prefer to buy or lease the asset outright rather than take a governance-equal partner into an operation you will have to reshape.

The Plays

Try these this week

Verb-first executable actions — each one tied to a stated outcome in the episode.

Run a SaaS renewal-calendar kill list as a standing agenda item

Outcome: SaaS spend survives on default renewal; a dated calendar with a named owner per contract converts it into an active decision.

“we have a slide of like when are SaaS contracts due and whose job is it to tell them that we're not renewing this year?”
— Fred Turner
annual cycle, reviewed at recurring leadership cadence per
  1. 1

  2. 2

  3. 3

  4. 4

  5. 5

Stop or pivot when

  • →classify each tool as workflow-encoding (replaceable) or infrastructure (keep)
  • →integration depth, as with Slack, can make replacement uneconomic

Scripts

Before you start

  • · in-house engineering capacity to rebuild replaced tools
  • · leadership willingness to absorb migration risk

Broadcast spare capacity publicly to find the buyer you cannot identify

Outcome: Under scarcity, invert the search — broadcast capacity publicly and let unidentifiable buyers self-select, rather than running outbound you cannot target.

“she tweeted, Hey we've got COVID testing capacity, does anybody wants some? And the deputy mayor of LA slid into her dms and was like, yes please, we would like to talk about that.”
— Fred Turner
hours to days from post to first contact per
  1. 1

  2. 2

  3. 3

  4. 4

  5. 5

Stop or pivot when

  • →supply must be genuinely scarce
  • →the offer must be understandable without context

Scripts

Before you start

  • · real capacity to deliver
  • · credibility story ready for a buyer who has never heard of you
  • · ability to pass a pilot against incumbent vendors

Daily invoicing at net-one with a government customer

Outcome: Government payment terms are convention, not law; when the buyer is desperate, daily net-one invoicing is negotiable and functions as free growth financing.

“they would pay after delivery but they would pay net one on the invoice. And so we would deliver the tests for a day and then we would send somebody to City Hall the next morning to pick up a check for those tests.”
— Fred Turner
24-hour cash cycle per
  1. 1

  2. 2

  3. 3

  4. 4

  5. 5

Stop or pivot when

  • →buyer has urgent, politically visible need
  • →delivery is measurable daily and uncontested

Scripts

Before you start

  • · a named contact inside the buyer's finance department
  • · clean daily volume reporting

Hire on an explicit expiry date

Outcome: When you know the demand is temporary, sell the temporariness explicitly at hire — it prices the wind-down in advance and self-selects for mission motivation.

“When we started hiring people at the beginning we told them This Is three months you have a job for three months. Don't bank on anything beyond three months. This Is a three month gig and we're gonna shut it all down in three months.”
— Fred Turner
three months stated at hire; six months for senior leadership per
  1. 1

  2. 2

  3. 3

  4. 4

  5. 5

Stop or pivot when

  • →demand driver is exogenous and expected to end
  • →local labour market can absorb the return of workers

Scripts

Before you start

  • · founder conviction that the boom terminates
  • · willingness to lose candidates who want permanence

Disposable single-use Python for arbitrary file ingestion

Outcome: Route around weak direct parsing by having the model write, test and iterate a throwaway parser per file, then discard it — no maintenance tail.

“You can tell it to write a Python script to convert any random file into this known format and then test it and loop and iterate on your script until it's working and then you throw away that script. And so it's single use code that never gets used again.”
— Fred Turner
about 15 minutes per file per
  1. 1

  2. 2

  3. 3

  4. 4

  5. 5

  6. 6

Stop or pivot when

  • →output must be schema-validatable so the loop has a stopping condition

Scripts

Before you start

  • · a strict validation target
  • · sandboxed code execution
  • · tolerance for per-file inference cost

Give the agent signature authority inside a guardrail envelope

Outcome: Delegate closing authority inside a pre-defined concession envelope; per-transaction review destroys the throughput that justified the agent.

“she now signs the agreements with my signature, she'll open up the DocuSign link and then click the button and it's my signature on that agreement.”
— Fred Turner
end-to-end contract close in a few hours per
  1. 1

  2. 2

  3. 3

  4. 4

  5. 5

  6. 6

  7. 7

  8. 8

Stop or pivot when

  • →margin/rate band beyond which the agent must escalate
  • →counterparty size above which a human takes the relationship

Scripts

Before you start

  • · lawyer-drafted template
  • · documented concession envelope
  • · founder willing to accept binding agent signature
  • · internal agreement — took multiple rounds of convincing

Migrate off a SaaS vendor with an agentic asset-by-asset loop

Outcome: Vendor lock-in lives in the artefact tail; an enumerate-convert-verify-decommission agent loop dissolves it and reprices every renewal negotiation.

“We did most of it with an agentic workflow that would spin up, find the next dashboard, figure out how to convert it into what we needed and then close it down on the looker side and boot it up on the other side.”
— Fred Turner
a couple of months with 1-2 people, versus an estimated year with a full team per
  1. 1

  2. 2

  3. 3

  4. 4

  5. 5

Stop or pivot when

  • →artefact definitions must be machine-readable or extractable

Before you start

  • · target platform selected and provisioned
  • · output-level verification method
  • · willingness to run old and new in parallel during cutover

The parking-lot five-minute interview line

Outcome: Size your hiring process to the cost of a bad hire, not to convention — for fast-signal roles under time pressure, a five-minute filter beats a five-stage loop.

“We'd have people like lineup in the parking lot socially distanced like down the street and then give them five minute interview slots and just have somebody sit there with a clipboard and it's like five minutes and next just to get the volume of people in the door.”
— Fred Turner
same-day decision; continuous while demand ramps per
  1. 1

  2. 2

  3. 3

  4. 4

  5. 5

  6. 6

Stop or pivot when

  • →role must not require a licence
  • →performance must be observable within days
  • →cost of a bad hire measured in days, not quarters

Scripts

Before you start

  • · large motivated applicant pool
  • · short engagement horizon
  • · supervisors able to observe output immediately

Decision Moments

Actual decisions, real outcomes

Specific decisions narrated in the episode with their outcomes and transferable lessons.

February-March 2020. Turner's sepsis diagnostics company Shield had just died; its successor Curative had raised $1M of seed money to build a sepsis service inside community hospitals. The first pilot hospital in Wisconsin abruptly cancelled via the clinician's assistant with a three-month blackout because they were preparing for their first COVID-19 patient. All hospitals went into lockdown, making the sepsis clinical studies impossible. The company's chief scientific officer had, in spare evenings and weekends through January and early February, developed a COVID test as a morale exercise for employees — but Curative had no lab licence, because Turner had sold Shield's CLIA licence for $150,000 two months earlier to pay creditors.

Did: Went all-in on COVID testing while explicitly framing it as temporary. Emailed his entire network looking for a CLIA-licensed facility; everyone in the Bay Area refused to have anything COVID-related on site. A former YC batchmate turned VC connected him to a sports-doping lab in San Dimas, LA County. He structured it as a 50/50 JV — Curative bringing the test and the scale-up expertise, the counterparty bringing the licence — flew down with a car full of PCR machines, and started testing. When the partner proved unable to scale and began actively obstructing, he bought them out for $27M funded from forward revenue rather than equity.Outcome: Curative reached 206,000 tests in a single day by December 2020, eight months from zero, and $5BN of revenue over three years — the largest non-Quest/LabCorp testing provider in the US. The $150,000 licence he had sold cost $27M to replace five months later.

A pivot into a demand shock is available only to operators who already hold the domain capability and are willing to abandon their current plan entirely. The expensive part was not the decision but the asset he had already liquidated: wind-down decisions made at normal-regime prices systematically destroy options whose value is contingent on a regime change.

Part of an emerging decision pattern across multiple episodes

Mid-2020 through 2021. Demand for COVID testing required capacity that could not be bought — the existing lab industry was optimised for flat volume and could not 10x, all standard consumable supply chains had broken down, and every competitor was bidding for the same swabs and magnetic-bead extraction kits produced in two Chinese factories. Meanwhile the State of Florida had put its entire nursing-home testing programme out to bid; every established lab said the turnaround requirements were impossible.

Did: Threw away the industry's playbook and built an orthogonal supply chain — deliberately sourcing inputs nobody else was competing for: swabs from electronics-testing vendors, sterilised in-house; filter-plate extraction instead of magnetic beads, because glass and plastic manufacturing can be scaled with partners. In parallel, scaled headcount from about 7 to about 7,000 in nine months using parking-lot queues with five-minute interview slots for unlicensed roles, telling every hire the job was three months. Funded the ramp from customer cash by invoicing the City of LA daily at net-one terms. Bid on and won the Florida contract everyone else called impossible.Outcome: Peak of 206,000 tests/day; the Dodger Stadium site alone ran seven lanes, 7am-7pm, seven days a week, testing 10,000 people a day. Won the Florida nursing-home contract worth hundreds of millions. Margins were excellent during surges and negative in every trough, because peak-sized fixed costs persisted at 20-30% utilisation.

Scaling 1000x in nine months is a supply-chain and hiring-throughput problem before it is a technology problem, and it requires deliberately not competing for the inputs everyone else wants. But the same architecture that makes the surge possible guarantees losses in the lulls — elasticity and efficiency are opposing designs and the choice is effectively irreversible.

Part of an emerging decision pattern across multiple episodes

Mid-2020, at the very start of the revenue ramp. Turner was convinced the COVID demand would not last, but each surge kept extending the horizon and he was repeatedly wrong about the end date. The company was generating enormous cash from a demand curve that would eventually go to zero, with 7,000 employees hired on explicitly temporary terms and no follow-on business.

Did: Began the search for the next business immediately rather than at the end of the boom. Ruled out staying in lab testing because displacing all of LabCorp and Quest caps out around $30BN of market cap. Ruled out buying hospitals in Florida and Texas because a hospital's payer mix is too fragmented — half government, half many small plans with conflicting and shifting demands — limiting the ability to change anything. Concluded that in US healthcare the payer drives all behaviour, got serious about health insurance in late 2021, and spent roughly $500M of COVID profits building it, plus paying investors a 10x dividend while they retained their shares.Outcome: Curative is now a health insurance company valued at $1.3BN after a $150M insider-led round. Headcount is about 650 and guided down toward roughly 400 as agentic workflows convert the back office. Of $5BN in COVID revenue, roughly $500M became the durable asset.

When the demand driver is exogenous, run the next-business search at the top of the curve while the boom funds it — and select the new market by where control sits in the value chain, not by adjacency to what you already do. The revenue from a demand shock is not the asset; the capital and the operator capability it leaves behind are.

Part of an emerging decision pattern across multiple episodes

2024-2025. Curative had designed its health insurance business in 2022 with no anticipation that LLMs were coming, and had recently migrated onto a large off-the-shelf claims platform that is hard to use, hard to get data out of, and that the vendor will not let them manage. Its network-building team of about 45 people was executing roughly 100 provider contracts a week against a requirement of 60-70,000 contracts — the barrier that gave Blue Cross a hundred-year head start. Credentialing took two to three months and $50 per doctor.

Did: Rebuilt the operating model around Claude agents rather than waiting for better models, and granted them binding authority. Credentialing went to zero headcount. An agent named Gwen now runs provider contracting end to end — research, outreach at ~15,000 customised emails a day, multi-round rate negotiation, redlining via generated Python, and signing with Turner's own DocuSign signature within a guardrail envelope and with no law firm review. Cancelled a $600K/year Salesforce contract for a vibe-coded internal CRM built in two months, migrated off Looker via an agentic loop, rebuilt the claims system in house, and targeted an 80% cut to total SaaS spend. Framed every deployment as a 10x volume target rather than a cost replacement.Outcome: Credentialing: 2-3 months and $50 to about 12 hours and $0.20. Contracting: ~100/week to ~100/day; the agent did 3,500 contracts in eight weeks versus 2,300 by the whole team the previous year, at ~$70 per contract versus $1,500-2,000. Anthropic spend grew 6x per month for six to seven months into the millions. The 45-person team was redeployed to large hospital systems rather than cut.

Granting agents binding authority inside a defined concession envelope is what converts them from assistants into throughput — and the blocker is organisational trust, not capability. Framing deployments as volume multiples rather than cost replacement is what makes the economics survive a supplier raising prices; Turner states a 2x or even 5x price increase would still work.

Part of an emerging decision pattern across multiple episodes

Tensions surfaced

Contradictions and trade-offs the episode raises — judgment calls a thoughtful operator has to navigate.

Tension

Building for the peak and running for efficiency are irreconcilable

Elasticity and efficiency are opposing operational architectures; you must choose one, and switching is a rebuild not an adjustment.

Decide explicitly whether your demand is flat or episodic before designing capacity. The two architectures cannot be hedged.

Tension

Agent authority is an internal trust problem, not a technical one

The gating factor on agent value is how much binding authority the organisation will grant, and that is a political negotiation not a capability question.

Treat the authority envelope as the design artefact. Define what the agent may concede, then delegate signing inside it rather than reviewing every output.

Tension

Five billion in revenue produced almost no durable enterprise value

Revenue scale and enterprise value are orthogonal when the demand curve is known to be temporary.

If your demand is a shock rather than a trend, plan the conversion of revenue into a durable second business from day one; the revenue itself will not be the asset.

Corpus connection

Where this episode fits for retrieval

What kinds of decisions this briefing is best pulled into.

Primary decisions

  • • strategic-bet
  • • hire
  • • scale-operations
  • • pivot
  • • market-selection