New startup ideas · AI and software · Agent infrastructure

startup idea

Portico

Agents that work state legacy systems through their own screens, priced per case cleared.

Portico runs computer-use agents that log into the mainframe terminals, county portals and 1990s case management systems that state agencies still process benefits, permits and licences in, and complete the transaction the way a clerk would.

AI agent as a service
Government and public sector
$1-10B market
Collapses a cost 10x
US first

3/5

venture judge

similar startups, last 2 years (0 all-time)

98%

of 4 nearest real companies still alive

sector

36 public grants in this sector, none matching the idea's terms

Test it before you build it

$1,000 · 6 weeks · 15 prospects

Prove a county will issue a purchase order and system credentials for agent-cleared backlog cases at $15 each, for $1,000 in 6 weeks, with no production code written.

Riskiest assumption · A county director with small purchase authority will grant an outside vendor supervised credentials to their state eligibility system and buy cleared cases on a PO, rather than burying the request in a year of state IT security review, which is the access the risk analyst flags as unlicensed screen work.

1Focus group: who and where

The deputy director or operations chief who owns the published SNAP redetermination backlog number at a county agency in a county-administered state, holds small purchase authority, and is under a state corrective action plan for timeliness this quarter.

where to find 15 · The CalSAWS consortium's public committee meetings and county roster, the user group of the eligibility system all 58 California counties already run; USDA FNS timeliness reports naming counties under corrective action, to rank the worst 15 backlogs; NACo's Health and Human Services policy community, plus warm intros through 18F and Code for America alumni now inside state digital service teams.

2Sell first, build later

A 250-case redetermination pilot: Portico's agents clear backlog cases on the county's existing screens within 30 days of credential issuance, county QA accepts or rejects every case, and the county pays only for accepted cases.

the ask · $15 per cleared case, $3,750 for the 250-case pilot, against the $40 to $80 of clerk time each case costs today.

a real yes · A signed purchase order plus a credential request submitted to state IT is a yes; an enthusiastic call, an offer of an unpaid demo on live data, or a director who defers to state IT without a date is a no.

3Small experiments

The first one attacks the riskiest assumption; each ends with a number that says whether to run the next.

1. Access-and-PO gut check calls
$750 · 14 days
Call 15 county ops directors from the FNS-ranked list, attending one CalSAWS public committee day in Sacramento to meet several in person. The ask on every call is concrete: a 250-case paid pilot at $15 per case, $3,750 total under their small purchase threshold, with a supervised credentialed worker account issued to Portico. Count who names a path to credentials versus who deflects to state IT with no date.
keep going if · 4 of 15 directors agree to route the quote to procurement as a small purchase and name how the account gets issued
2. Recipe proof on training screens
$100 · 10 days
Ask the two friendliest counties for a screen recording of a clerk working 5 redeterminations, then demo an agent replaying the flow in the system's training environment or a faithful mock, with per-case correctness checked against the clerk's outcome. Show the recording-to-recipe demo live to every director who took the first call.
keep going if · The agent matches the clerk's outcome on 5 of 5 training cases, and 3 of 4 directors who see it advance to the PO step
3. PO under the threshold
$150 · 21 days
Send each advancing county a one-page scope, a W-9, vendor registration, and a quote for 250 cases at $15 per case, payment only for cases their QA accepts. Then count purchase orders, not verbal yeses.
keep going if · 2 POs issued within 21 days of the quote

4Collect a deposit up front

Tesla took $1,000 refundable reservations for the Model 3 and $100 for the Cybertruck before building either: the deposit is the measurement, not the revenue.

$3,750

per prospect, refundable

how · A county purchase order for the 250-case pilot at $15 per case, issued under the existing small purchase threshold with no new procurement vehicle, paired with the credentialed account request submitted to state IT; counties commit through POs rather than prepayment, so the signed PO is the money moved.

what it reserves · A slot in the first pilot cohort, the $15-per-case price locked for a year, and a go-live date within 30 days of credential issuance.

refund · The county may cancel the PO in writing at no cost any time before go-live, and afterward pays only for cases its own QA accepts.

target · 2 signed POs with credential requests submitted, from 15 director conversations, within 6 weeks

before taking money · SNAP casework runs on federally governed systems: obtain written county authorization and state-issued credentials before touching any production screen, and never let the agent finalize a determination without the county's QA sign-off.

Go: build it if

2 signed POs with credential requests submitted to state IT and at least 4 of 15 directors having named a working credential path.

Kill: stop if

0 POs after 15 conversations and every director deferring to state IT security with no date, which means the credentialed-access assumption is false and the wedge does not exist.

5 Scripts to run itoutreach message, landing copy, deposit terms · click to open

outreach message

Your redetermination backlog is a published number, and every case sitting in it costs you $40 to $80 of clerk time. I clear cases for $15 each: my agents work your existing screens exactly the way a clerk would, your QA accepts or rejects every single case, and a 250-case pilot lands at $3,750, under your small purchase threshold with nothing to install. Can I get 20 minutes this week to walk through how the supervised credentialed account works?

landing page

Backlog cases cleared for $15, on the screens you already run $15 per cleared case; the 250-case pilot is $3,750, under most county small purchase thresholds, and you pay only for cases your QA accepts. Request the one-page pilot scope and quote

deposit terms

Your purchase order for the 250-case pilot at $15 per case, $3,750 total, reserves a first-cohort slot and a go-live date within 30 days of credential issuance, with the $15 price locked for a year. You pay only for cases your QA accepts, invoiced net 30 after delivery. Cancel the PO in writing at no cost any time before go-live.

Would you run this test?

One tap. The yes-share feeds the Demand pillar of this idea's score; nobody sees who answered.

Budgets are out-of-pocket estimates for a team of one to three, US market. Size the deposit to the deal, and check the terms before taking money in a regulated line.

Scorecard

One score that balances how trendy the idea is, the demand for it and its potential for 100x, with competition measured relative to every other idea in the catalog. Recent startup trends first, government priorities second.

Idea Score, 0-100 · raw 41.3 x 1.61

Open

competition: more crowded than 6% of ideas · headwind x0.97

+0.0

government priorities, secondary (0 matching grants)

Trend

Is the wave forming now? 2025-26 entrants vs 2023-24, rounds since 2025, the sector's live-batch direction, the 2026 trend analyst.

Entrants 2025-26 vs 2023-24 (similar companies)23
Rounds announced 2025+ in the sector4
Sector direction (live batch)100
2026 trend analyst50

Demand

Does anyone want it? YC's current RFS, companies already paid for something similar, the operator judge, founders' yes-rate in decks, readers who would run the test.

YC asks for it (current RFS: idea / sector)30
Someone already pays (similar companies, recent / all-time)0
Operator judge: real pain100

100x potential

Can it return a fund? The venture judge (double weight), market-size and moat axes, neighbours still alive, the technologist judge.

Venture judge50
Market size axis33
Moat axis20
Technologist judge50

Score = 100 x cbrt(Trend x Demand x 100x) x (1 - 0.5 x crowding) + government bonus (max 5), calibrated so the 95th-percentile idea scores 90 (order never changes). A geometric mean: a weak pillar cannot be papered over. Percentiles are among the 272 ideas in the catalog; the terms matched were state, legacy, computer-use, log, mainframe, terminals, county.

The idea in full

What: Portico runs computer-use agents that log into the mainframe terminals, county portals and 1990s case management systems that state agencies still process benefits, permits and licences in, and complete the transaction the way a clerk would. Agencies buy cleared cases, not software: eligibility redeterminations processed, permit files advanced, backlog worked down. A drift detector re-derives the operating recipe whenever a screen changes, so the integration maintains itself instead of breaking on the next vendor patch.
Why now: YC's Fall 2026 Request for Startups names Self-Maintaining APIs, and Rindler in S26 is building the translation layer between AI agents and the web while CoArena is building the largest crowdsourced computer-use benchmark; NSF also funded an I-Corps project on evaluating foundation model outputs at ~$50k, so the measurement side of reliability is being built in public.
Wedge: first customer and entry point: One county unemployment or SNAP redetermination queue with a published backlog number; work it on a per-case fee under an existing small purchase threshold, no new procurement vehicle needed.
Path to 100x: State and local case processing labor for these queues is a $1-10B addressable market where a cleared case costs $40 to $80 in staff time, and Portico prices at a fraction of that on software margins because there is no new system to install. The 100x requires the recipe library across the handful of vendor systems that most states share to become reusable, so the fiftieth agency onboards in days and the cost per case keeps falling as volume grows.
Ceiling: If each state's configuration turns out to be genuinely bespoke, delivery cost tracks headcount and the business becomes a modernisation contractor with better tools.
Closest real companies, as the generator saw them: Rindler translates between agents and the web as infrastructure for developers, and Decawork sells an agent control plane to IT teams; Portico sells the completed government transaction and carries the reliability risk. Propolis attempted hands-off QA via browser agents and is now inactive, which is the failure mode Portico must avoid by measuring per-case correctness rather than screen actions.
Main risk: A single wrong benefits determination produces a news story and a statewide suspension, and no agency reinstates the vendor.

Five judges

Each judge scores every idea in the catalog with a named rubric; the venture judge decides whether a card is shown at all (4-5 is venture-grade).

Venture investor
3/5
Software margins on $40-80 cleared cases scale well, but bespoke state configurations would turn it into a modernisation contractor.
Bootstrapper
5/5
Per-case fees under a county small purchase threshold, no new procurement vehicle, software margins against $40-80 of clerk time.
Operator
5/5
A published county backlog priced at $40 to $80 per case, bought under a small purchase threshold with no new system installed.
Technologist
3/5
The drift detector that re-derives recipes is the only real engineering, and generic computer-use models are commoditising underneath it.
Risk
2/5
Screen-scraping state mainframes with no licence and no moat, where one wrong SNAP determination produces a statewide suspension and no reinstatement.
trends
3/5
Computer-use agents becoming reliable is a genuine current shift and per-case pricing fits it, but the Self-Maintaining APIs RFS tie is a stretch for screen scraping.

Similar startups in the directory

Companies whose pitch matches most of the idea's terms (state, legacy, computer-use, log, mainframe, terminals, county): 0 all-time, 0 from the last two years. Same matching as Idea Check.

Nothing in the directory matches this idea's terms: an open space, or a wording nobody uses yet.

Run this as an Idea Check →

The generator's reference companies

Real companies the model named as closest when it wrote the card, with their fate. A check mark is a company the radar could verify in its directory.

Public money in this direction

US federal grants, SBIR/STTR awards and open opportunities from the radar's public-money feed, matched to the idea's terms; the sector totals give the context.

grants and programs matching the idea

startup-relevant grants in Agent infrastructure

$16M

awarded in the sector, tracked

All public money by sector →

Market signal

What the radar sees in Agent infrastructure: new companies by cohort year, the forming YC batch, and outcomes since the February snapshot.

Agent infrastructure · 28 → 64 → 108 → 150 → 178 new companies 2022 → 2026 · 94% aliveYC F26 live: 23 in this cluster, 19% of the batch (was 16% in S26)Since February, of 239 YC companies here: 9 acquired, 4 shut down, 110 rewrote their pitch

Agent infrastructure: companies, trend and grants →

Design attributes

The card is one cell of a designed set: every axis below was chosen before the text was written, and the text had to realize it.

Buyer: Government and public sector
Business model: AI agent as a service
Path to 100x: Collapses a cost 10x
Market size: $1-10B market
Capital intensity: Capital-light (software margins)
Speed to revenue: Revenue in 1-3 years
Technical depth: Deep tech: ML, hardware, bio
Go-to-market: Founder-led sales
Moat: No moat yet
Geography: US first
Regulation: Some regulation
Vibe: Hot space