Y Combinator removes Delve after audit allegations; new directories spotlight AI-agent startups and infrastructure
This edition contrasts Y Combinator’s removal of Delve amid disputed audit allegations with its expanding pipeline of AI-agent and infrastructure startups. New products and financings show continued commercial momentum, but directory descriptions and company-reported metrics remain early signals rather than proof of adoption, accuracy or revenue.
Wednesday, Oct 7, 2026
The latest from Y Combinator
The Latest from Y Combinator
YC cuts ties with Delve amid audit allegations
Y Combinator removed compliance startup Delve from its public portfolio after allegations that the company reused near-identical SOC 2 reports across clients. An anonymous writer said 493 of 494 reviewed reports contained the same typo, with client names and logos changed.
Delve’s co-founders disputed the account, calling the leak unauthorized access and denying that the company fabricated evidence or issued audit reports itself. YC has not explained its decision, while scrutiny has also reached lead Series A investor Insight Partners.
Why it matters: The dispute highlights the risks of using AI-enabled shortcuts in compliance, where paperwork is supposed to represent real controls.
If the allegations are substantiated, affected customers could face renewed manual work and reduced confidence in their certifications. The episode also shows that accelerator and investor backing does not eliminate the need for independent verification.
YC’s pipeline moves AI agents into specialized workflows
YC’s publicly listed startups include Rote, which prepares disputed vehicle-repair claims, and LinkLane, whose agents quote, book and manage freight.
YC also promoted Collar, an investor-relations product that turns company and investor data into briefs, answers and targeting decisions.
The listings and promotion indicate a focus on software that performs work inside specialized processes rather than simply generating chat responses.
Why it matters: AI agents are moving toward workflows involving claims, freight and investor outreach, where errors can affect external commitments.
These products remain early-stage signals—not evidence of customer adoption, accuracy or revenue—so approval points, source records and human fallback will be important tests.
YC’s infrastructure directory underscores the systems behind AI agents
YC’s October 7, 2026 infrastructure directory update lists 285 funded companies and highlights several F2026 startups building systems for AI agents.
The companies include Terse, Agent 37, Tilion, Hopper, Maritime and Boat, with products spanning agent coordination, persistent sandboxes, browser automation, speech-model improvement and cloud computers.
Why it matters: The directory points to a broader AI product stack: agents need persistent state, access to software, browser capabilities and scalable compute before they can perform work continuously.
The entries establish positioning and cohort presence, but not customer traction or funding outcomes. Sustained usage will be the key test.
YC’s fintech directory shows a portfolio at very different stages
YC’s October 2026 directory lists 574 financial technology and services startups, alongside a broader directory covering more than 5,000 companies.
The portfolio ranges from active startups to public companies: Dakar-based Wave is listed with 1,600 employees, Bengaluru-based Groww is public with 1,050, and Razorpay is active with 2,700.
Razorpay’s profile says its valuation reached $7.5 billion after Series F and describes expansion from payments into banking and offline point-of-sale services.
Why it matters: This is a portfolio snapshot, not a new investment announcement. It shows how YC affiliation can span companies at sharply different stages, so current operating status, employee scale, geography and financing history matter more than accelerator membership alone.
Axle raises a $17.5 million Series A for insurance infrastructure
YC-backed Axle announced a $17.5 million Series A led by Base10 Partners to continue building an AI-native insurance clearinghouse.
Y Combinator and Gradient Ventures continued investing, alongside Stage2 Capital, Cover Genius co-founder and CEO Chris Bayley, and other industry angels. The round adds to a backer group that includes the early team at Plaid.
Why it matters: YC’s continued participation shows ongoing support as Axle moves beyond its earlier financing stage. The funding is intended to support product development, but the announcement does not establish customer, revenue or adoption metrics.
Execution and industry uptake remain the important next signals.
A GitHub project creates a maintained index of YC-backed open-source startups
A GitHub repository tracking YC-funded open-source startups said its data refreshes daily as of October 7, 2026. It draws the startup list from YC’s open-source API and separately maps companies to canonical GitHub repositories, with contributor verification requirements.
Automated updates refresh repository statistics and the generated table.
Why it matters: The project offers a repeatable way to track the overlap between YC funding and open-source development.
Repository links and star counts provide limited visibility into activity, but the index depends on YC’s classification and ongoing contributor maintenance; it is not a measure of traction or product-market fit.
Egoist Machines launches AI Passport with a $3 million seed round
YC-backed Egoist Machines launched AI Passport as a permission-based layer for carrying personal context between assistants such as ChatGPT, Claude, Gemini and Muse. Users can review information from connected apps, control where it is shared, keep it in one app or delete it.
The company announced a $3 million seed round involving YC, Goodwater Capital, a European family office, and scientists from DeepMind and Meta.
Egoist says its earlier beta grew 41% week over week, attracted more than 10 B2B2C pilots with at least 1 million users each, and now connects to 800 apps including Gmail, Notion and GitHub.
Why it matters: AI Passport targets the fragmentation of user context across assistants and could give users a central point for permissions and deletion.
Its prospects depend on trust, recurring user engagement and broad app support; the reported pilots and connectors indicate early ecosystem interest, not proven adoption.
Advice for early-stage founders
Advice for early-stage founders
The guidance below is evergreen startup advice, separate from the reported Y Combinator developments above.
Start with the customer’s problem
Talk to people who regularly experience the problem—not just people who say the idea sounds interesting. Ask what they do today, what the workaround costs, how often the problem occurs, and what they have already tried. Look for repeated, specific behavior rather than hypothetical enthusiasm.
Build the smallest useful product
Choose one user, one painful job, and one clear outcome. Remove features that do not help the user reach that outcome. A narrow product that delivers value quickly will teach you more than a broad product that takes months to complete.
Prioritize evidence, not volume
Keep a short list of assumptions that must be true for the product to work. Rank them by risk and test the most important ones first. A useful priority filter is:
Does this address the core customer problem?
Will solving it improve activation, retention, or trust?
Can we test it quickly?
What important work would it displace?
Treat distribution as part of the product
Identify where your target users already gather, search, buy, or ask for help. Test one or two repeatable acquisition paths before expanding widely. Track whether a channel produces qualified users who continue using the product—not only clicks or sign-ups.
Turn feedback into learning
Watch users attempt the task, ask what confused them, and record the exact language they use. Separate requests from underlying problems: several feature requests may point to one unmet need. Review feedback regularly, identify patterns, and close the loop by telling users what changed—or why it did not.
The goal is not to follow every request. It is to learn quickly enough to make better product and distribution decisions with each iteration.