Y Combinator leads Q3 startup deals; Vitalize raises $31 million; Yale study finds bias in AI evaluations
Y Combinator’s reach is evident in both the quarter’s deal activity and its expanding robotics directory, while portfolio companies are turning AI into operational tools for healthcare and enterprise customers. The growth comes with practical tests: deployment and compliance must keep pace, and Yale’s findings raise questions about whether AI-based startup evaluation can avoid reproducing bias.
Friday, Oct 9, 2026
The latest from Y Combinator
The latest from Y Combinator
YC led Q3 startup deal activity
Y Combinator was the busiest investor in Crunchbase’s Q3 rankings, participating in 45 post-seed rounds and at least 221 known seed rounds. Its activity remained high even as overall venture funding declined and no new record-setting AI megarounds emerged.
The figures point to a broad financing network spanning initial seed investments and follow-on support. They do not, however, show how much capital YC invested, whether it led the rounds, or whether the startups involved had achieved product-market fit.
Vitalize raises $31 million for healthcare workforce operations
Vitalize announced a $31 million Series A led by Oak HC/FT, with participation from Norwest, Y Combinator, . 406 Ventures, Rock Health Capital, Constellation, and MemorialCare Innovation Fund.
The company says its platform coordinates staffing, scheduling, patient capacity, and workforce planning for nearly 20 enterprise health-system clients.
The financing follows a discovery-led founding process: founder Veeraj Shah says he spent a year embedded in about 15 health systems before launching the company, with nurses and clinical leaders helping shape the product.
Vitalize is now hiring aggressively, making its ability to scale adoption while retaining close customer contact the next key test.
Yale study highlights hidden bias in AI startup evaluations
A Yale study tested 2,000 startup pitches based on ideas submitted to Y Combinator between 2020 and 2023. Researchers found that an LLM’s evaluations changed when identical pitches were assigned names associated with different racial and gender groups.
Some groups received slightly higher scores and were less likely to rank last, but those differences did not translate into a greater likelihood of winning.
The results suggest that AI systems can alter visible evaluation scores without changing their underlying selection patterns, underscoring the need to audit outcomes—not just the language used by screening tools.
YC’s robotics directory reaches 142 startups
Y Combinator’s robotics directory lists 142 active startups, including a new group from the F2026 cohort. The companies span components, deployment, maintenance, care-at-home systems, and general-purpose robotics.
Examples include Silica’s force-sensing actuators, Open Actuation’s motors and actuators, and Rhem Labs’ connected hardware and data tools for care at home.
The roster shows founders targeting specific operational problems—not only building robots themselves. Companies such as Mantle, Deploy, and Shepherd Robotics are focused on field service, task execution, data-center construction, and manufacturing.
The directory indicates where startups are concentrating, but provides no customer or revenue evidence.
Clicks Health turns deployment into a product strategy
A Y Combinator job posting shows Clicks Health hiring a Forward Deployed Strategist to install AI agents inside healthcare organizations. The role involves learning revenue-cycle workflows, translating them into agent instructions and decision rules, testing performance, and taking the systems live.
Clicks says its agents work across tasks such as prior authorization, denial management, and payment posting, and claims more than 90% automation, deployment in under five weeks, and guaranteed savings of at least 50%.
The approach makes implementation a form of customer discovery: repeated deployments reveal where agents work, where they fail, and what the product should build next.
Advice for early-stage founders
Advice for early-stage founders
The guidance below is general startup advice, separate from the reported Y Combinator developments above.
Start with the customer’s workflow
Talk to people who regularly experience the problem—not only decision-makers or enthusiastic acquaintances. Ask them to describe the last time the problem occurred:
What were they trying to accomplish?
What did they do instead?
Which steps were slow, costly, or error-prone?
What would make them change their current process?
Look for repeated behavior and existing workarounds. A problem that produces action is usually more useful than a problem that produces agreement.
Build the smallest useful product
Begin with one user, one painful job, and one measurable outcome. The first version may require manual work behind the scenes; that is acceptable if it helps you learn whether the result matters to customers.
Avoid building a broad platform before you know which workflow customers will repeatedly use. A small product that delivers value is more informative than a large product that merely demonstrates technical capability.
Prioritize evidence over requests
User feedback is input, not an automatic roadmap. When someone requests a feature, understand:
What job are they trying to complete?
How often does the need occur?
What happens if it remains unsolved?
Are other users experiencing the same problem?
Can the request be tested without fully building it?
Prioritize problems by customer impact, frequency, urgency, and how directly they support your product’s core promise. Be cautious about building for the loudest individual user.
Treat distribution as part of the product
A useful product still needs a repeatable path to its users. Test distribution early rather than waiting for the product to feel complete. Speak with potential customers through the channels they already use, and note which message earns a response, meeting, trial, or referral.
Your initial audience should be specific enough that you can explain why the product is relevant to them. If every customer requires a completely different pitch, workflow, or implementation, you may still be searching for a focused market.
Turn usage into a learning loop
After each release or customer interaction, record:
The behavior you expected
What actually happened
The evidence supporting that observation
The change you will make
The result you expect to see
Review this regularly and distinguish a one-off complaint from a recurring pattern. The goal is not to eliminate all negative feedback; it is to shorten the time between learning something important and acting on it.