MarkosWebWeb intelligence for the agentic economy

startup ideas · ai

AI startup ideas, from what AI companies actually build

MarkosWeb classifies 3,014 accelerator companies into five AI clusters, 827 of them from 2025-26 cohorts. Lists of "AI startup ideas" usually name categories; this page shows where the money and the founders are going, which subniches are already full, and which sectors are turning AI-native fastest - the place where the unclaimed ideas are.

3,014

AI companies tracked

827

from 2025-26 cohorts

26%

of YC S26 are vertical AI agents

AI clusters: size and direction

Share is the cluster's slice of classified YC S26 companies; change compares the last two batches with F25+W26. Each cluster links to its own page with companies, trend by year, funding and grants.

ClusterCompanies2025-26Share, S26Change
Vertical AI agents1,44444725.7%-3.2 pp
Agent infrastructure46020616.5%-0.1 pp
AI infra and compute341687.2%+1.6 pp
Data for AI487574.2%-0.8 pp
Horizontal AI assistants282493.0%+0.9 pp

AI startup ideas by sector: where AI-native share is jumping

Outside the AI clusters, the share of new companies whose product is AI at the core. A sector going from a third to two thirds AI-native in two years is where AI ideas are being applied right now - and where the non-AI incumbents are being replaced.

SectorAI-native, 2025-26AI-native, 2022-23New companies 2025-26
Developer tools74%44%95
Security and compliance69%41%61
Healthcare and bio54%29%127
Consumer53%13%88
Robotics and physical world48%25%136
B2B SaaS39%17%98
Fintech34%13%111
Climate and energy28%25%39
Commerce and marketplaces28%12%47
Defense and space23%13%47

Crowded AI subniches: differentiate or avoid

The most repeated subniche labels among 2024-26 AI companies. Three companies with the same label in two years is not a red flag; it is a queue forming.

Concepts written from the data

Idea Lab invents fictional companies from a cluster's brief - graveyard, crowded subniches, fresh rounds, grants, YC requests. Three from each AI cluster that has a set; the full sets with wedge, competitors and risk are on the Idea Lab page.

Vertical AI agents · 2026-08-17

  • Pennant - An AI paperwork department for school district special education teams.

    The radar tracks 473 companies in this cluster from the last two years and none of the 36 most recent funded entries touch K-12, while federal money is clearly flowing into applied education AI: NSF awards of ~$750k for an AI-researcher-teacher network, ~$442k and ~$200k for LLM learning support, plus the NIH NExT AI Hubs opportunity closing 2026-10-02. The vertical-agent playbook is being applied to radiology networks and freight brokerages but not to the most paperwork-heavy public buyer in the country.

  • Rexa Desk - AI reimbursement desk that clears rejected claims for independent pharmacies.

    Remedy died in 2025 pitching broad 'AI Agents for Pharmacies', and Egress Health died in 2025 doing automated revenue cycle for dentists; meanwhile 'revenue cycle' is only used by 4 companies from the last two years, so the space is thin rather than crowded. The failures suggest the wrong wedge, not the wrong market: a rejection-and-remittance agent has a measurable dollar output on day one instead of asking a five-person pharmacy to change how it operates.

  • Indemnis - Underwriting-grade audit layer that makes AI-run professional firms insurable.

    The cluster has 1441 tracked companies and 473 in the last two years, with 2026 already at 203 new entrants, and the newest cohort is explicitly replacing licensed firms: Billow AI Labs as an AI-native accounting firm to replace the Big-4, Last Accounting Company as an agent-native accounting firm, Veltha as an AI claims adjuster for regulated insurance, TovenAI for compliance at institutional trading firms. Hundreds of firms performing regulated work with agents creates a liability-pricing gap nobody in the tracked set is filling.

Full set in Idea Lab →

Agent infrastructure · 2026-08-17

  • Ratchet - Drift firewall that keeps a vendor's API stable for the agents calling it.

    'Self-Maintaining APIs' is the one YC Fall 2026 RFS theme mapped to this cluster, and the 2026 cohort already shows 106 new agent-infra companies, nearly all of them consuming third-party APIs; supply-side tooling is missing while demand-side integration layers (Executor, Rindler) pile up.

  • Quorumline - Consensus gate that requires independent model agreement before an agent commits an irreversible action.

    NSF is actively funding the primitive: three 'Resource-Efficient Byzantine Protocols' awards at roughly $403k, $300k and $286k, plus 'Self-Organizing Decentralized Agents' awards near $150k-$200k, while the tracked cluster's average AI-visibility score of 48/100 suggests most agent runtimes still ship with a single-model decision path.

  • Purser - Spend mandates and machine-readable receipts for agents that buy things.

    Agent payments sits inside the cluster definition but of the 36 most recent tracked companies only Financial Datasets touches money at all, and it sells market data rather than settlement; 220 companies launched in the last two years are increasingly running purchase and procurement tasks with no per-agent spend primitive.

Full set in Idea Lab →

Concept names are invented; numbers inside them come from the cluster brief at generation time.

How to pick an AI startup idea from this data

  1. Start from a sector in the second table, not from the AI clusters: a vertical where AI-native share doubled still has incumbents to replace, while "AI agent orchestration" already has a queue.
  2. Type the idea into Idea Check: it returns the tracked companies building it and how many are still alive.
  3. Compare against demand: Gaps maps YC's current requests to supply, and Public Money shows which of these spaces US federal grants are funding.
  4. Read the graveyard for the cluster: what the last attempt got wrong is usually one specific dependency that has since moved.

More startup ideas by angle

Validate any of them: Idea Check shows who already builds it; Idea Lab writes concepts from a cluster's data; Chat matches ideas to what you can build.