What changed
FACT: CMS issued an interim final rule (Federal Register, 2026-06-03, doc 2026-11094, corrected 2026-06-29) implementing a Medicaid community-engagement (work) requirement — millions of expansion-population adults must periodically document qualifying activities or lose coverage, and every affected state must stand up verification/reporting machinery. FACT: Google shipped computer use in Gemini 3.5 Flash, putting browser-driving agents at low-cost model pricing. The convergence: per-member per-month portal filings that were never worth human caseworker time are now economical to automate.
Why now
The rule is final-ish (interim final, in effect ahead of comment resolution) and state compliance dates land over the next 6-12 months (statutory implementation is expected in the quarter after Dec 31, 2026 — HYPOTHESIS on exact per-state timing; some states like Georgia already run live work-requirement portals). The window is the gap before Gainwell/Maximus-class MCO vendors ship equivalent retention tooling. History (Arkansas 2018) shows most coverage losses come from documentation failure, not ineligibility — that is the pain this automates away (HYPOTHESIS grounded in widely reported studies, not in the provided sources).
Converging signals
(1) FORCED-BUYER regulation: CMS community-engagement rule compels a defined class — expansion adults — to file recurring verifications into state systems (federalregister.gov). (2) Cheap capability: Flash-tier computer use makes per-filing automation cost pennies (deepmind.google). (3) Ecosystem alarm: Georgetown CCF, Norton Rose Fulbright, HME Business, and Paragon all published analyses/comments within weeks — proof the regulated ecosystem is scrambling.
Customer pain
Beneficiaries lose coverage over paperwork they qualify to pass: confusing portals, monthly cadence, document uploads, low digital literacy. MCOs lose the member's monthly capitation the moment the state disenrolls them, and reacquisition is expensive or impossible. CBOs and legal-aid groups are about to drown in manual verification assistance. Georgetown CCF explicitly frames the rule as threatening coverage for eligible people — the paperwork IS the pain (cited).
Who pays
PRIMARY: Medicaid MCOs (per-member-retained or PMPM fee on the at-risk segment) — each avoided disenrollment preserves capitation. SECONDARY: community health orgs, FQHCs, and legal-aid groups with retention/outreach grant funding; state agencies are a distant third (procurement-gated, avoid). CRITICAL HONESTY: the COMPELLED party (the member) is not the PAYER — the MCO's purchase is discretionary, so this is NOT a pure forced-buyer sale; it is a strong-ROI enterprise sale.
Solved today
Members self-file on state portals or call caseworkers; MCOs run mail/SMS reminder campaigns and community health worker outreach; CBO navigators sit with members one-at-a-time; states attempt ex parte data-matching (payroll databases) to auto-verify some members. Incumbent MCO vendors (Gainwell, HMS, Maximus) sell eligibility/redetermination outreach services.
Why current solutions are bad
Reminders don't complete filings — the member still faces the portal alone. Human-assisted filing costs more per member than the capitation at risk justifies at scale. Ex parte matching misses gig workers, volunteers, caregivers, and students — exactly the people who must upload proof. Nobody today closes the loop from 'member has proof' to 'submission confirmed in the state system' automatically.
Proposed product
Consent-based filing agent: member enrolls once (consent + credentials/authorized-representative designation), texts photos of pay stubs/volunteer logs to an SMS number; a Flash-tier computer-use agent completes the state portal attestation/upload monthly; every run produces an audit log and confirmation screenshot; exceptions route to a human review queue. Sold to MCOs as retention infrastructure with a compliance-grade audit trail, or white-labeled to CBOs.
MVP version
One state, one portal (Georgia's live work-requirement flow is the obvious testbed — HYPOTHESIS that it remains the first/most accessible live portal), SMS intake via Twilio, agent runbook for the single filing flow, human fallback for every failure, HIPAA-eligible hosting (AWS/GCP BAA tiers are commodity). Explicitly run the input's kill test: 10 consenting members, 3 monthly cycles, require ≥8/10 unblocked end-to-end completions.
30-day build
Legal review FIRST (authorized-representative rules, state portal terms, assister-certification requirements in the target state — this is the real gate, not the tech). Build the agent against the live portal in a test harness. Recruit one CBO/legal-aid partner to source 10 consenting pilot members.
60-day build
Run monthly cycle 1-2 of the kill test. Instrument everything: completion rate, CAPTCHA/identity-wall hit rate, minutes of human fallback per filing. Package results as a one-page outcome study ('we kept N of 10 members enrolled at $X each').
90-day revenue plan
Two parallel monetization tracks: (a) paid CBO/FQHC pilots at $500-2,000/mo per org (reachable, grant-funded, no procurement committee) — realistic first revenue; (b) open MCO conversations with the pilot data, targeting a paid pilot with a regional/provider-owned Medicaid plan (not Centene-class) — realistic at day 120-270, not 90.
Distribution path
CBO/legal-aid networks and Medicaid-focused conferences for the wedge; regional and provider-sponsored MCO plan innovation teams for the real revenue; the pilot outcome study is the sales asset. This is demonstrated-value selling, which fits the founder — but the MCO leg still requires BAAs and security review (see penalties).
Pricing hypothesis
CBO white-label: $500-2,000/mo per org. MCO: $10-20 PMPM on the flagged at-risk segment or $25-75 per verified filing/member-month retained — trivially ROI-positive against $300-800/mo capitation (capitation range is HYPOTHESIS/industry-common-knowledge, not from provided sources).
Technical difficulty
Moderate. The agent itself is now cheap (that's the epiphany). The hard parts are non-model: portal bot-detection and identity-proofing walls (ID.me-class flows may be unautomatable by design), per-state portal variance, evidence-quality triage of member photos, and running a HIPAA-grade audit-logged pipeline solo.
Legal / regulatory risk
HIGH and load-bearing — three distinct layers, all HYPOTHESES needing counsel before build: (1) whether automated credentialed access to a state eligibility portal on a member's behalf is permitted (portal ToS, state computer-access statutes; 'authorized representative' designation likely helps but was designed for humans); (2) several states certify/register application assisters — if that regime covers this activity, it's a licensure gate; (3) the rule itself faces litigation risk (advocacy groups call it unlawful in the cited coverage) — if enjoined or softened, the market shrinks or vanishes. Also absolute duty of non-deceptive conduct toward a vulnerable population.
Platform dependency
State portals are government systems — no commercial deplatforming, but bot-detection/identity-proofing is functionally equivalent and is the named kill-test risk. Model dependency is low (any computer-use-capable tier works).
Founder fit
Split verdict. The core motion — read a mandate, find the compelled filer class, automate the government-portal submission, charge per transaction — is EXACTLY his proven FMCSA ELDT edge, and the lesson base (confidence 0.65) says this shape fits him best. But the paying channel breaks his pattern: ELDT customers were small training providers paying by card; here the money is in HIPAA-gated MCO enterprise sales to Centene/Molina-class plans, plus custody of a vulnerable population's PII — both on his avoid list. The CBO wedge partially rescues fit; it does not fully repair it.
Breakout potential
Large if it works: ~40 states must implement, the same member-side agent extends to SNAP recertification, redeterminations, and childcare-subsidy reporting — a general 'benefits paperwork agent' with recurring monthly volume. Also a defensible data asset: per-portal automation runbooks.
Final recommendation
CONDITIONAL GO — as a cheap, staged validation, not a committed build. The convergence is real and the founder's portal-automation edge is squarely relevant, but the buyer channel (MCO enterprise), the PII burden, the not-yet-live portals in most states, and rule litigation push first meaningful revenue past the 180-day window. Spend limited capital now on (1) a legal opinion in one live-portal state and (2) the 10-member/3-cycle kill test via a CBO partner. If the agent clears ≥8/10 and counsel clears the access model, pursue the CBO white-label wedge for first revenue and use its data to approach regional MCOs. If identity-proofing blocks the agent, kill without regret.
Next action
Commission a fixed-fee legal memo (Medicaid authorized-representative + assister rules + portal access, in the first live-portal state, Georgia being the likely candidate) while simultaneously building the agent against that portal and lining up one CBO to source 10 consenting pilot members — kill test starts the first monthly cycle after both clear.