Two to three months. That's how long clinical trials typically idle before enrolling their first patient, just waiting for data capture systems to get configured. A San Diego startup thinks it has a better answer: five minutes, give or take.
Harbor, which emerged from Y Combinator earlier this year, is making a characteristically bold pitch to an industry not known for moving fast. Upload your clinical protocol—yes, just the PDF—and the company's so-called "Magic Build" feature will spin up your entire study infrastructure before you've finished your second cup of coffee. Complete database schema, schedule of events, case report forms. Done.
It's the kind of promise that makes clinical research veterans raise an eyebrow. And perhaps for good reason. This is a field where Medidata and Veeva have spent years, even decades, embedding themselves into the workflows of Big Pharma. These systems don't just capture data; they've become the operating system for how trials run.
But Harbor's founders, Albert Cai and Nathan Leung, seem unbothered by the incumbents. Their pricing page tells part of the story: $2,000 per month, flat rate. No per-subject fees, no per-site upcharges, none of the byzantine pricing structures that have made enterprise EDC contracts a negotiation unto themselves. Academic trials? Those run free.
The Technical Gambit
The platform's core technology hinges on what the company calls protocol-reading AI. In a blog post from earlier this year detailing their technical approach, Harbor describes how the system parses clinical protocols to automatically scaffold everything downstream. Traditional EDC systems, the company argues, require months of manual configuration precisely because they weren't architected with AI as a first principle.
Whether that explanation holds up under scrutiny is another matter. Legacy vendors would likely counter that the manual configuration exists for a reason—domain expertise, regulatory nuance, the kind of institutional knowledge that doesn't compress neatly into an algorithm.
Harbor's second major feature takes aim at data entry itself. Research sites upload source documents—lab reports, scans, medical records—and the platform's AI extracts the relevant information and populates EDC fields. Human reviewers confirm the entries afterward. The company claims a 90 percent reduction in manual data entry in its own blog post, though that figure is self-reported and hasn't been independently verified.
Then there's Magic Monitor, which layers AI-generated confidence scores over side-by-side views of source documents and captured data. The FDA has been pushing "risk-based monitoring" for years now, a shift away from on-site visits toward centralized oversight. Harbor is betting that AI can make that transition not just feasible but efficient.
Early Traction, Measured Optimism
As of its March launch, Harbor claims five clinical studies running live on its platform, though independent verification of this figure is not publicly available. The most detailed public case study involves MotilityCount ApS, a Danish company conducting an IVF trial across 15 to 25 centers. Cynthia Hudson, CEO of NOVA Genomics, the partner organization on that study, said Harbor delivered a fully validated EDC in nine days at what she described as "a fraction of the cost." The study went from contract signature to launch in a week.
Nine days versus the industry standard of 10 weeks—or longer, depending on study complexity. If those timelines hold across different trial types, it's not an incremental improvement. It's a categorical shift.
Harbor's homepage lists other customers: Biolinq, BioDynamik, Kalevala Therapeutics, University Hospitals. Third-party confirmations of those partnerships weren't immediately available, which is typical for early-stage startups still building out case study libraries. The company exhibited at the WCG MAGI Clinical Research Conference in Chicago this past April, its first conference sponsorship.
The Compliance Question
Regulatory compliance is the third rail of clinical software. Get it wrong and the entire edifice collapses. Harbor claims adherence to 21 CFR Part 11—the FDA's electronic records rule—along with HIPAA and Good Clinical Practice standards. Each trial gets an isolated database, hosted on Google Cloud Platform and AWS infrastructure.
In a September blog post, the company outlined its validation approach: continuous automated testing with audit-ready artifacts generated for each software release. The goal, Harbor says, is to enable rapid feature updates without sacrificing the documentation rigor regulators demand.
That's a tricky balance. Traditional vendors have years of regulatory inspections under their belts, a track record that newer entrants simply can't replicate overnight. Harbor's approach may technically check all the boxes, but clinical operations teams tend to be risk-averse for good reason. A failed inspection doesn't just derail one study—it can cascade across an entire portfolio.
Who's Building This
Cai, Harbor's CEO, comes from Biolinq, a continuous glucose monitoring company, where he led clinical trials and regulatory strategy through De Novo authorization. Leung, his co-founder, spent time at Google and Ramp before joining an earlier YC company as its first employee.
It's a two-person team, at least according to Y Combinator's directory. That's lean even by startup standards, though not unusual for software companies in the early days. The company hasn't disclosed external funding beyond its YC participation, which typically includes a small check and three months of intensive mentorship.
The Broader Bet
Harbor is targeting a market that's both enormous and notoriously sticky. Sponsors across medical devices, pharma, and biotech rely on EDC systems, as do contract research organizations running everything from early feasibility studies to post-market registries. Switching costs are high—not just financially, but organizationally. Training staff on new software, validating new systems, migrating historical data: it's a heavy lift.
But Harbor's pricing and speed may create an opening, particularly among smaller biotech companies and academic institutions that have historically been priced out of enterprise contracts. $2,000 per month is a rounding error compared to what Medidata or Veeva typically charge, especially once per-patient fees start adding up.
The open question is whether AI can genuinely replicate the domain expertise embedded in legacy systems. Clinical trials are messy, full of edge cases and regulatory idiosyncrasies that don't always map neatly to automation. A protocol might call for one thing, but real-world execution often requires judgment calls that seasoned clinical data managers have learned through years in the field.
Still, Harbor is betting that speed and cost will override hesitation. And in a market where trial timelines can mean the difference between beating competitors to market or watching a patent window close, that's not an unreasonable wager.
Whether the bet pays off will depend on more than just the technology. It'll come down to trust—whether clinical operations teams, notoriously cautious and for good reason, will take a chance on the new platform when established vendors offer the safety of familiarity. That's the harder problem to solve, and no amount of AI can automate it away.
