Y Combinator Fall 2025, One Year On: The Durability Data

The batch, by the numbers

WillItSurvive assessed 10,525 companies in July 2026, and 152 of them belong to the Y Combinator Fall 2025 batch, giving us an unusually complete picture of a single cohort roughly one year after Demo Day. The headline is not flattering. Of the 152 YC Fall 2025 companies assessed in July 2026, 77 landed at "Losing its protection," 28 at "Protection under test," and 47 at "Little protection left," with one record unresolved. Pressure across the YC Fall 2025 batch ran high overall as of July 2026, and 71 of the 152 companies already carried a high-pressure flag before most had any named customers.

A note on labels: "Little protection left" in this batch mostly means the field is unsettled and crowded, not that the company holds an edge. It is a description of the terrain, not a promise about the company. We never declare an outcome; we publish what the evidence showed at a dated moment and tell you what to check next.

Where the pressure comes from

The pattern across the batch is repetition. AI agent tooling, no-code agent builders, observability, browser agents, and health coaching apps each attracted multiple YC entrants chasing the same buyers that Microsoft, Google, Salesforce, Meta, and well-funded specialists like Aidoc, Encord, and Applied Intuition already serve.

The assessment pressure map (July 2026) breaks this into concrete matchups:

• Metorial, Nexus, and Lua AI face Zapier, Microsoft Copilot Studio, Salesforce Agentforce, Decagon, Sierra, and Ada, all of which already sell no-code or low-code AI agent building at scale.
• Dome, Locus, Aside, Thesis, and dari.dev built functions, from agent payments to agentic browsing to agent deployment, that platforms including Polymarket, Coinbase, Stripe, Google, OpenAI, Anthropic, and Modal Labs shipped natively.
• Nori and AirCaps compete with Oura, Whoop, Samsung, and Meta's Ray-Ban Display line, which now bundle AI coaching or captioning directly into devices people already wear.
• Nucleo, Deeptrace, and Crunched watch larger rivals like Aidoc, Datadog, Resolve AI, Microsoft Copilot Agent Mode, and Claude for Excel roll their core feature into products with installed bases.
• Leadbay, Lunabill, and Zarna saw established vendors such as Apollo.io, ZoomInfo, Candid Health, Infinitus Systems, Hebbia, and Rogo ship near-identical features or already sit inside the target customers.

The companies under the most strain

Dome faced the most acute pressure in the batch after Polymarket absorbed it and sunset its APIs by April 28, 2026. Developer attention around prediction-market data had already shifted toward pmxt and PolyRouter by the assessment date.

Locus, competing against agent-payment standards already shipped by Coinbase, Stripe, Visa, Mastercard, Google, and OpenAI, showed little of its own technical differentiation as of the July 2026 assessment. Aside, a five-person team, is up against native agentic browsing already live in Chrome, ChatGPT, Perplexity Comet, and Dia. Thesis, a three-person team, has been outranked on benchmark results by better-funded research-agent rivals and the major AI labs. Selfin had no live product as of July 2026, entering a banking-app market where Monarch Money, Cleo, Copilot, and Era already ship AI features.

In health, Rovi Health's web traffic dropped 41% quarter over quarter per the July 2026 assessment, while Included Health and Transcarent dominated care navigation. Nori's pressure comes from Oura, Whoop, and Samsung bundling AI health coaching into hardware.

The companies with the most room

Four names stood out at the low-pressure end of the batch in July 2026:

• Structured AI stood out for its narrow construction QA/QC niche, which had no recorded direct customer loss.
• Norra's footing is tied to physical NFC hardware in nursing facilities, which is harder for a software-only rival to copy quickly, and it showed the least competitive strain in the batch.
• Telemetron had no confirmed customer losses, focused on hardware and medical device support rather than general support tooling.
• Sava also showed little strain; large trust departments hold the incumbent business, but the assessment found no evidence of Sava losing ground.

The common thread is narrowness. Companies with a hard technical or regulatory foothold, like construction QA checks or NFC hardware tied to specific facilities, held up better than those selling a thin AI wrapper around a workflow that OpenAI, Anthropic, or Microsoft can absorb in a product update.

How much to trust these labels

WillItSurvive checked its method against companies that had already failed, using only facts that were public before their collapse. Across 116 companies that later shut down, the assessment correctly flagged the large majority as being in serious trouble, and it never called a company that failed within 18 months safe to bet on. On a matched set of survivors from the same period, the method held up consistently on review. The supported claim is narrow: given only pre-collapse facts, the assessment separated the doomed from matched survivors and never once said safe-to-join on a company that was dead within 18 months. That is a track record on hindsight-controlled data, not a forecast about any company in this batch.

Watch-items, dated so you can verify them

We publish forward-looking material only as things a reader can check on a date, never as declared outcomes.

• By June 2027: whether Aidoc, which per the July 2026 assessment had over 500 million dollars in funding, more than 20 FDA clearances, and presence in about 2,000 hospitals, ships a cancer segmentation feature for CT scans before Nucleo secures a named hospital customer or FDA clearance.
• By January 2027: whether Meta's Ray-Ban Display line still holds more than 60 percent of the smart glasses market, and whether AirCaps reports any meaningful share gain in live captioning.
• By December 2026: whether Supabase remains the default backend inside Bolt, Lovable, and Replit, and whether Specific appears as an official integration in any of the three.

Broader signals worth tracking for any name in this batch: a platform giant shipping a native feature matching a startup's exact one-liner (the fastest way a company here loses its reason to exist, since few have signed customers creating switching cost); a startup disclosing its first named enterprise contract (one of the few concrete signs of pulling ahead); and a funded rival moving into a currently quiet niche like construction QA or trust automation, which could change the picture for Structured AI or Sava quickly.

Questions this data answers

Which YC Fall 2025 startups showed the strongest durability signals in July 2026? Structured AI, Norra, Telemetron, and Sava showed the least competitive strain in the batch as of July 2026. They share narrow footholds, construction QA, nursing-facility NFC hardware, medical device support, and trust automation, rather than general AI agent tooling.

Did any YC Fall 2025 companies stop operating independently within the first year? Dome effectively did. Polymarket absorbed it and shut down its APIs by April 28, 2026, and developer attention moved to pmxt and PolyRouter. Dome faced the most acute competitive pressure in the batch, per the July 2026 assessment.

How crowded is the AI agent tooling space this batch entered? Very. Metorial, Nexus, and Lua AI compete directly with Zapier, Microsoft Copilot Studio, Salesforce Agentforce, Decagon, Sierra, and Ada, per the July 2026 pressure map. Buyers in this category already have a default option before any of these startups reach them.

Which YC Fall 2025 companies faced the least competition? Norra and Telemetron showed the least competitive pressure in the batch, per the July 2026 assessment. Their focus areas, medical equipment tracking and hardware customer support, are narrower than the crowded general agent space most of the batch entered.

How is Nucleo positioned against Aidoc? Under heavy pressure. Per the July 2026 assessment, Aidoc had over 500 million dollars in funding, more than 20 FDA clearances, and presence in about 2,000 hospitals, and its roadmap points at the cancer segmentation work Nucleo does. Whether Nucleo lands a named hospital customer first is a checkable item through mid-2027.

Assessments reflect evidence available July 2026. Companies change; treat watch-items as things to verify, not conclusions.