What changed
Three things landed at once: EPA posted a grant program funding TA intermediaries to serve thousands of rural/small/tribal wastewater systems (grants.gov 362798 — FACT); OfficeCLI now lets a headless server manipulate the Word/Excel forms this world runs on with one binary (FACT); and Gemini 3.5 Flash computer use makes per-document browser/portal automation economically viable for a solo builder (FACT).
Why now
The cost of producing a compliance packet by machine collapsed in the same window that federal money was appropriated to pay intermediaries whose deliverable IS compliance packets. The spread between a loaded staff-hour and an agent-hour is the margin, and it did not exist 12 months ago.
Converging signals
(1) EPA TA/training grant for rural/small/tribal wastewater systems — a funded intermediary class with paperwork deliverables; (2) OfficeCLI — headless Office-document assembly on any server; (3) Flash-tier computer use — cheap agentic portal navigation. Rule-money, document tooling, and portal automation meet at one point: the packet.
Customer pain
TA grantees have fixed staff and a mandate to serve thousands of systems; each served utility means hand-built monitoring summaries, sampling schedules, asset inventories, and funding applications. The utilities themselves face recurring permit-reporting obligations (HYPOTHESIS as to exact volumes — the input names 'chronic compliance and paperwork pain' but no counts). Note honestly: the demand_evidence array is EMPTY — no complaints or job postings from grantees asking for this were retrieved, so grantee appetite is unproven.
Who pays
The TA grantee organizations (RCAP-style nonprofits, state rural water associations) out of their federal award — a pre-funded budget line. The beneficiary (small utility) and the buyer (funded grantee) are deliberately different; this sidesteps the small-utility poverty problem. CRITICAL UNPROVEN ASSUMPTION: that their award budgets permit software/subcontracted tooling as an allowable cost rather than exclusively staff labor (2 CFR 200 rebudgeting rules apply) — this is the gate to validate before building.
Solved today
Grantee staff labor: Word/Excel templates filled by hand, manual data entry into state and EPA portals, spreadsheets. Larger utilities use lab/compliance suites (WIMS-class) that are priced and designed above the small-system tier.
Why current solutions are bad
Per-packet labor cost caps how many systems each grantee can serve against its deliverable targets; incumbent software targets big utilities, not the intermediary-serving-100-tiny-systems workflow. EPA is effectively paying consultant hours for work an agent pipeline can draft in minutes — that appropriated spend is the existing-spend evidence (cited), even though it is not yet software spend.
Proposed product
An intake-to-packet factory sold to grantees: structured intake form per utility → OfficeCLI assembles the state's Word/Excel forms → computer-use agent prepares portal submission DRAFTS → human-in-the-loop review screen → grantee staff (or the utility's certified operator) reviews and signs/submits. E-signature/certification stays with the legally responsible signatory — the product prepares, it does not sign (HYPOTHESIS: federal e-reporting signature rules likely prohibit fully automated submission, so draft-prep is the ceiling; this caps but does not kill the value).
MVP version
One state. Corpus of its small-system forms; three packet types (discharge-monitoring summary, sampling schedule, asset inventory); pipeline + review UI; per-packet invoice. No multi-state generality, no direct portal submission — drafts only.
30-day build
Pick one state; assemble its paperwork corpus; pull the prior/current grantee list from the NOFO program page and USAspending; email 10 grantees a free 2-utility pilot offer AND ask each point-blank whether their award budget allows paying a software subcontractor (this doubles as the kill test).
60-day build
Build the intake→OfficeCLI→draft-submission pipeline with human review; run the pilot on 5 real utilities with one regional TA provider; measure hours saved per packet vs. their staff baseline.
90-day revenue plan
Convert the pilot to paid per-completed-packet pricing; expand within the grantee's service territory (they hold the utility relationships — you hold the factory). First revenue realistically days 90–150 given nonprofit decision speed; founder's runway covers this per the capital lesson (confidence 0.90).
Distribution path
Enumerable buyer list: grantees are publicly named on the EPA program page and USAspending — direct outreach plus state rural water association events. Demonstrated-value sale (pilot metrics), which matches the founder's selling style. Weakness: the national pool is likely only dozens of organizations, so a handful of rejections materially closes the channel — the 2-week kill test is the right gate.
Pricing hypothesis
$75–250 per completed packet, or $1–2k/month per grantee region; anchor against the loaded staff-hours a packet currently costs them. Per-transaction pricing mirrors the founder's proven FMCSA per-upload model.
Technical difficulty
Moderate. Form variance across states and permits, data extraction from messy utility records, and accuracy QA (an erroneous monitoring report is a permit problem) are the hard parts; OfficeCLI + Flash-tier computer use make the mechanics cheap. Human-in-the-loop is non-negotiable in v1.
Legal / regulatory risk
Low-to-moderate. No licensure required of the founder; compliance domain is the moat, not a burden. Signature/certification authority must remain with the utility's responsible official (draft-only design). Liability for draft errors handled by review-and-sign workflow plus contract language. The real legal question is grant-side: allowable-cost treatment of the tool under the grantee's award.
Platform dependency
No platform owner can deplatform a tool that prepares government filings. Dependencies are ordinary: model-vendor pricing and state form/portal changes (which are also the recurring-revenue reason customers stay).
Founder fit
Near-maximal. This is the FMCSA ELDT shape one level up the chain: public money flows (EPA award) → a funded class must produce documented deliverables → he builds the paperwork layer and charges per transaction. Industrial-operations credibility plus government-portal automation plus AI-workflow skill all load-bearing. Docked one point because the buyer is a discretionary funded intermediary, not a compelled filer.
Breakout potential
High if the wedge lands: 50 near-identical state markets, the drinking-water TA twin of this program, USDA rural-development and FEMA cost-recovery paperwork classes, and white-labeling the factory to engineering/consulting firms serving the same systems.
Final recommendation
PURSUE WITH A HARD GATE. The convergence is real and the founder fit is exceptional, but the two load-bearing unknowns (allowable-cost treatment and grantee appetite) are both testable in two weeks for the cost of 10 emails. Run the kill test before writing pipeline code; build the state corpus in parallel since it is reusable for adjacent plays even if this buyer dies.
Next action
Today: pull the grantee list from the grants.gov/EPA program page and USAspending, and send 10 pilot-offer emails that explicitly ask (a) will you pilot on 2 utilities free, and (b) does your award budget permit paying a software subcontractor per packet. Two weeks, no acceptances → kill.