Idea Lab
Pick a vertical. The radar reads everything it tracks there - momentum by year, crowded subniches, dead companies, fresh rounds, grant money, open YC RFS themes - and proposes concrete startups someone should build. The companies below are fictional; the data behind them is not.
Vertical AI agents
473
companies, last 2 years
1441
tracked all-time
94%
of checked sites alive
$13M
across 1 recent rounds
Most crowded right now: operating system (21) · law firms (7) · real estate (7)
concept 1
Pennant
An AI paperwork department for school district special education teams.
Pennant drafts, checks and files the IEP documents, progress reports and Medicaid reimbursement claims that special education coordinators produce by hand today. It plugs into the district's student information system, produces compliant drafts for case managers to sign, and tracks deadlines and consent trails for state audits. Buyer is the district's director of special services, not a teacher.
- why now
- 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.
- wedge
- One mid-size district's special education office, priced per active IEP, starting with the annual-review draft and the deadline calendar only, expanding into Medicaid claiming after the first audit cycle passes clean.
- closest real companies
- Nothing in the tracked set targets K-12; the closest analogues are Stratum Industries (applied AI for governments) and Locke (AI-native government affairs firm), both of which sell to agencies and lobbies rather than school districts, and neither touches student compliance records.
- main risk
- District procurement plus student-data privacy review can stretch a first contract past the runway of a three-person team.
concept 2
Rexa Desk
AI reimbursement desk that clears rejected claims for independent pharmacies.
Rexa Desk sits on the pharmacy management system and works the rejection queue: reprocessing PBM claim denials, chasing prior authorizations, and reconciling DIR and remittance shortfalls so owners see what they were actually paid. It is one narrow money-recovery job, billed as a share of recovered dollars, not a general pharmacy assistant.
- why now
- 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.
- wedge
- Single-store and two-store independents in one state, sold through a pharmacy buying group, starting only with top-20 rejection codes on a contingency fee.
- closest real companies
- Remedy (graveyard, 2025) went horizontal across pharmacy operations; Egress Health (graveyard, 2025) targeted dental RCM; Veltha and Qlo work the payer/insurer side of claims rather than the provider's rejection queue.
- main risk
- PBM and switch-level data access is gatekept, and without a clean integration path the agent degrades into screen-scraping that breaks weekly.
concept 3
Indemnis
Underwriting-grade audit layer that makes AI-run professional firms insurable.
Indemnis records, scores and freezes the work product of vertical AI agents in licensed professions - accounting, legal, insurance adjusting - into an evidence file that an errors-and-omissions underwriter can price. Agent companies embed the SDK to get supervision logs, sampling reviews and a signed-off human attestation trail; carriers and brokers consume the resulting risk report. Sold to both sides, with the carrier relationship as the moat.
- why now
- 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.
- wedge
- Ten agent-native accounting and claims startups that already need E&O cover to close enterprise deals; give them the audit export free and charge the carrier for the risk report.
- closest real companies
- TovenAI does compliance monitoring inside trading firms and Definite models a client's books; neither prices agent liability for insurers. Veltha and Qlo are insurance-domain agents, i.e. potential customers rather than competitors. No tracked company sells an underwriting layer for agent work product.
- main risk
- Carriers may keep underwriting AI firms on gut feel and cheap questionnaires, leaving no one willing to pay for measured evidence.
concept 4
Parthex
AI quote desk for industrial distributors buried in messy RFQ email.
Parthex reads inbound RFQ emails, spreadsheets and scanned drawings, resolves part numbers against the distributor's legacy ERP and supplier catalogs, and returns a priced, margin-checked quote for a salesperson to approve. It is priced per quote turned around, and the hard technical work is cross-catalog part matching on dirty data rather than conversation.
- why now
- Soff died in 2024 doing 'AI Agents for Distributors' as a broad agent play, while the current cohort shows the industrial middle is now fundable with a narrow job: Orvixo on planning for manufacturers, Operon on an agentic data layer for process industries, Trope as an AI-native ERP integrator. Also note 'voice agents' is already claimed by 5 companies in two years - the unmet distributor pain is written quotes, not phone calls.
- wedge
- One electrical or fluid-power distributor with 20-60 inside sales reps and a 1990s ERP, starting with repeat-order RFQs where the part matching is easiest to validate.
- closest real companies
- Soff (graveyard, 2024) attacked distributors horizontally; Orvixo and Operon serve manufacturers upstream; Trope integrates ERPs rather than sitting on the quote flow; Palisade runs sales agents for marketplaces, not catalog quoting.
- main risk
- Quote accuracy below roughly 95% means reps re-check everything and the time saved disappears.
concept 5
Rowanmark
An AI executor's office that settles a family estate end to end.
Rowanmark works for the individual named as executor: it inventories accounts, prepares probate court filings for the county, notifies creditors and agencies, tracks the statutory clock, and hands a distribution package to heirs. Flat fee per estate with a licensed attorney on staff for court appearances. The customer is a grieving non-professional, not a law firm.
- why now
- Legal AI in the tracked set is aimed at firms and in-house teams - 'law firms' appears in 7 pitches and 'law firm' in 4 from the last two years, plus Lightfield doing in-house intake and triage - so the consumer side of legal process work is unoccupied. Veeza AI (YC F26) shows the direct-to-consumer, document-heavy bureaucratic wedge is being funded right now, and probate is the same shape: fixed forms, county-level rules, one painful transaction per person.
- wedge
- Two or three counties in one state, sold through funeral homes and hospice social workers, starting with small uncontested estates that need no tax return.
- closest real companies
- Lightfield (AI legal intake and triage for in-house teams) and Vestris (AI-native real estate closings) both sell to professionals; Veeza AI is the closest model but works visas for weak-passport travelers. No tracked company serves executors.
- main risk
- Unauthorized-practice-of-law rules force so much attorney involvement per estate that the unit economics look like a small law firm's.
concept 6
Palewell
AI operator for small municipal water and wastewater utilities.
Palewell handles the recurring operating paperwork of a town water system: SCADA and lab alarm triage into ranked work orders, monthly discharge monitoring and state compliance reports, lead-service-line inventory updates, and consumer confidence notices. It runs alongside one or two licensed operators who approve every submission. Sold to towns and small utility districts on annual contracts.
- why now
- The 2026 cohort shows agents moving into physical infrastructure and public buyers - Edviro on AI that operates energy infrastructure, Stratum Industries on applied AI for governments, Skymerse on flight operations autopilot - but water utilities, of which the US has thousands under 10,000 connections, appear nowhere among the 473 companies tracked in the last two years.
- wedge
- One state's small systems, entered through the licensed contract-operator firms that already run dozens of towns each, starting with the monthly state discharge report they file by hand.
- closest real companies
- Edviro operates energy infrastructure and Stratum Industries sells applied AI to governments; neither addresses water compliance. Avoca Systems and Care GP show the same operations-agent pattern in healthcare rather than utilities.
- main risk
- Small-town budgets and grant-cycle purchasing cap contract sizes below what a 1-3 person team needs to survive the long sales cycle.
Fictional concepts generated 2026-08-17 by claude-opus-5 from MarkosWeb data (accelerator portfolios, funding news, US federal grants, YC RFS). Numbers reflect tracked sources only; treat every concept as a research prompt, not a plan.