Petrichor, a San Francisco startup in Y Combinator's Fall 2026 batch, has launched a wet lab service in which trained scientists run physical experiments for AI labs and send back the complete experimental records as data. All three of its founders are 19, according to YC's company profile.
The pitch rests on a gap the founders see in how biology models learn. Petrichor's website argues that published papers leave out protocol changes, failures and the small observations that could work as training signals. Models, it says, should learn from the process of experimentation and not only from the tidy, successful endpoints that make it into print.
What Petrichor does
The company sells execution of physical experiments. An AI lab submits a request, scientists do the benchwork, and the lab receives a record of what happened. Petrichor's "How Petrichor works" page says trained scientists perform the experiments, so this is not a fully robotic, human-free laboratory.
All of that comes from the company itself. The page is undated, and I found no independent benchmarks of Petrichor's capabilities.
The stated long-term goal is to produce the experimental foundation for more capable biological AI and for disease research. The homepage presents this as a thesis. No clinical impact has been demonstrated.
The founders
YC lists Wilhelm Hedenskog as founder and CEO, age 19, and a former YC Fellow, though the listing gives neither the year of the fellowship nor the project. Jesper Johansson and Elias Reinfeldt are co-founders, also listed at 19. Grey Baker is named as the company's primary YC partner.
An Expressen article dated October 11, 2026, as summarized by News Minimalist, describes Hedenskog as being from Gothenburg and the founders as high-school classmates. Petrichor's own pages carry no such detail, and the original Expressen page could not be retrieved, so the account remains secondhand.
Customers and funding
Petrichor's pages describe categories of customers but name no paying clients or research collaborators. Time News reported on October 11, 2026 that the company hoped to secure contracts with major AI companies before an upcoming investor presentation. That is an ambition, not a set of signed deals.
No company-announced funding round appears in the sources reviewed. YC's standard terms, according to its website, provide $500,000 through separate SAFEs (simple agreements for future equity, a common early-stage instrument): $125,000 for 7%, plus $375,000 on an uncapped MFN SAFE. That is context about the program, not a verified Petrichor round or valuation. Recent Funding lists a $500,000 figure for the company, but its own footnote says starred amounts are estimated from an accelerator's standard deal and not from company disclosures.
A different name, and an earlier idea
The footer of Petrichor's website shows Qokedas, Inc., and MentionFox lists Qokedas as an earlier name. A Qokedas site that remains accessible names the same three founders and says the company is in YC F26. It describes turning scientific-instrument signals into model experiences and evaluations, with plans to read instruments, predict measurements and eventually operate them. That page is also undated, so the sequence of the two pitches is hard to pin down.
Competitors and context
Cloud labs, which let customers run experiments remotely, are not new. Ginkgo Bioworks announced a browser-based cloud lab on March 2, 2026, with more than 70 instruments and an AI agent called EstiMate that checks protocol compatibility and pricing. Ginkgo's current page advertises 70-plus instruments and 18 protocols, 15 of them certified and three in beta. Those are company-displayed figures, not audited capacity.
Emerald Cloud Lab's public documentation lists remote functions including liquid transfers, serial dilution, PCR and DNA/RNA synthesis. Its automated lab also belongs to the GEMSTONE project with Stanford University, Purdue University, Morehouse College and U.S. Pharmacopeia. Medra announced a DARPA collaboration on June 24, 2026.
Public money is moving into the category as well. The NSF announced $400 million over four years on July 22, 2026 for 20 programmable cloud-lab test-bed nodes, with the Astera Institute. Stanford reported an August 5, 2026 award of $20 million to a collaboration led by Mark Musen for remotely operated, AI-driven laboratories.
The AI Read, in editorial commentary added October 6, 2026, said Petrichor's distinction is that its customer is an AI agent, since cloud labs themselves are hardly novel. The same commentary flagged capital intensity and unproven purchasing demand as risks.
A correction on earlier coverage
Founderland published a launch article on October 3, 2026 that attributes to Hedenskog a statement about compliant AI for regulated laboratories. That wording appears in Kristina Hoang's announcement for a different company, Ordo Labs. Hedenskog's LinkedIn activity shows he liked the post. It is not a Petrichor quote, and it says nothing about the company's regulatory compliance.
Neither Founderland nor MentionFox, which dates its directory record to August 3, 2026, establishes a company-confirmed commercial launch date.
Pricing, turnaround times and the size of Petrichor's lab footprint do not appear on the company pages reviewed, its homepage and "How Petrichor works" page. The next test is whether a single AI lab signs on.
