Axioms

Constraint registry for idea evaluation. Read this file whenever brainstorming, evaluating, or comparing an idea/direction — right after index.md. Its job is to make every AI session (and every partner) apply the team’s accumulated critical thinking automatically.

Format rules:

• Two tiers per direction: Axioms = hard constraints the team formally decided (each cites its decision/meeting source and names a revisit trigger — axioms are falsifiable by a new logged decision, never silently ignored). Heuristics = evidence-derived strong defaults from research (challenge ideas with them; overridable by better evidence, cite when overriding).
• Every entry: **AX/H-<DIR>-<n> (date) — statement.** then rationale, a source wikilink to the decision/meeting/research page, and (axioms only) a revisit trigger.
• Only a team decision (logged in wiki/log.md, ideally a decision page) may add, amend, or retire an axiom. Retired entries move to the Retired section with date + reason — never deleted.

How the AI must use this file when an idea is proposed:

1. Classify the idea against each active axiom: consistent / violates / challenges.
2. If it violates an axiom — say so explicitly and require a deliberate override discussion; do not soften it.
3. If it merely challenges one — surface the tension and the axiom’s revisit trigger.
4. Apply heuristics as critical-thinking prompts, citing the underlying evidence page.

---

Insurance direction (“Insurance 2.0” / super-broker)

Axioms (decided)

• AX-INS-1 (2026-08-03) — We are not a carrier. We do not carry risk on our own balance sheet; carriers are rented infrastructure (“AWS”), we are the applicative layer (Robinhood template). Rationale: regulatory capital in the $10Ms+, ~8–10% margins, years-long loss feedback loops — “borderline a suicide mission” per two Lemonade operators. Source: 2026-08-03-enter-insurance-via-brokerage-mga-not-carrier. Revisit trigger: MGA at scale with proven underwriting data AND a partner balance sheet/foundry capital — and even then, fronting arrangements before ownership.
• AX-INS-2 (2026-08-05, amended 2026-08-15) — We are an AI brokerage serving END customers (SMB-first). We own the brand, the traffic, and the customer relationship; we optimize internal processes AND customer-facing processes. Amendments (2026-08-15, user-directed team decision): (a) B2C is the later expansion prize, not a forbidden scope — the Cover-style B2C channel was named “the biggest prize” (2026-08-13-directions-fundraising-and-validation-plan); SMB remains the entry. (b) Phase-1 execution may run through a partner brokerage that holds the licenses and executes placements while we hold the customer; BoR and execution migrate in-house as licenses land (2026-08-15-acquisition-wedge-before-workflows). Source: 2026-08-05 F2F (logged 2026-08-07) + 2026-08-07-directions-nizan-saar-ai-brokerage-thesis-and-pitch. Revisit triggers: audit→BoR conversion failing its kill-gate (<10%) in the pilot; the partner brokerage capturing the book (no contractual path to move it in-house).
• AX-INS-3 (2026-08-05) — No agency roll-ups. We build the book, we don’t buy it. Rationale: inherited mismatched books + heavy operations (the Equal Parts/AGI route is a different company). Source: 2026-08-05 F2F. Revisit trigger: a distressed-agency acquisition that is pure book purchase in our exact segment at distressed pricing.
• AX-INS-4 (2026-08-05, reaffirmed 2026-08-15) — No “AI for brokerages/carriers” tooling as the business. Small derivative TAM (~$5B, the smallest pool on the money-flow map) and it arms the channel we intend to out-compete. History: openly contested 2026-08-09 (“one of the axioms I’m not sure I agreed to” — 2026-08-09-directions-operational-ai-and-axiom-challenge); resolved 2026-08-15 when the team rejected the join-a-partner-to-optimize-op-ex path as “FDE for brokers” and chose the acquisition wedge instead (2026-08-15-acquisition-wedge-before-workflows). Source: 2026-08-05 F2F; economics in us-insurance-distribution-economics. Revisit trigger: none foreseen; a tooling wedge used purely as a Trojan for data/GTM would still require an explicit decision.
• AX-INS-6 (2026-08-07) — Premium is never TAM. All market sizing uses commission/revenue pools, waterfall haircuts, and the annually-contestable stream — never premium volume. Source: ai-brokerage-tam-model. Revisit trigger: none — this is arithmetic.
• AX-INS-7 (2026-08-15) — Differentiation = aligned incentives + real continuity, at the generalist+continuous corner. We win as the AI-native broker whose economics are independent of the customer’s premium — fixed fee + a commission-neutral comparative rater (the broker earns the same whether the plan is cheap or expensive; “we’re the angel, they’re the devil”) — and whose product is genuine post-bind, business-change-driven portfolio ownership. The competitive quadrant’s generalist+continuous corner is empty: harper owns speed, coverwatch owns auditing, cover/clark are comparison-shells without real continuity. Replaces retired AX-INS-5; the named-coverage play stays available as a GTM tactic, not the identity. Source: 2026-08-10-directions-huri-pitch-deck-build · 2026-08-15-directions-acquisition-wedge-and-deck-structure. Revisit trigger: fixed-fee economics failing against the commission-annuity math in us-insurance-distribution-economics, or discovery showing SMBs won’t pay a standalone fee (H-INS-9 pressure).
• AX-INS-8 (2026-08-15) — Prove acquisition before workflows; rent execution. The MVP is an efficient SMB acquisition channel (brandable, Cover-style front end) routing traffic to a partner brokerage — not internal-workflow tooling and not an own-license broker on day one; licenses and the internal broker are built in parallel. Joining a single partner to optimize their op-ex is explicitly rejected (“FDE for brokers” — proves nothing about an AI-native model). Brand/NPS guardrail: edge experience and pricing must be excellent even while the back end is rented. Source: 2026-08-15-acquisition-wedge-before-workflows. Revisit trigger: the channel failing CAC-vs-first-year-commission economics (us-insurance-distribution-economics) — then the operational-AI path (2026-08-09-directions-operational-ai-and-axiom-challenge) returns to the table.
• AX-INS-9 (2026-08-15) — US-first; the UK is the only second market; LatAm is off the table. The US choice is now a decision, not an inherited default: LatAm fails on substrate (~17–27% of SMBs carry any insurance — “an optimization play needs policies to optimize”), continental Europe on passporting-in-practice (~1%) plus socialized workers’ comp/health; the UK is the lone genuine alternative (AR route in 2–3 months, 94% broker share of commercial). FIDA (~2029+) is option value, not a wedge. Source: geography-alternatives-europe-latam · 2026-08-10-directions-smb-survey-verticals-and-fida. Revisit trigger: FIDA adopted with non-life scope reaching real application dates, or US diligence kill-gates failing.

Heuristics (evidence-derived defaults — challenge ideas with these)

• H-INS-1 — Thin accounts kill brokerages. Target accounts ≥~$25Kpremium(\ge$3K commission); micro-accounts ($200–500 commission) cannot carry CAC or service — Verifly’s corpse, CoverWallet’s dismantling. Evidence: us-insurance-distribution-economics · insurance-beachhead-segment-sprint.
• H-INS-2 — Policy-parsing/document AI is not a moat. It is commoditized broker tooling (Coverflow, Qumis, FurtherAI, Patra). Moat = BoR ownership + live business-change data + the category name. Evidence: smb-insurance-competitive-landscape.
• H-INS-3 — BoR flips are renewal-window campaigns, not a rail. 5–10-day incumbent rescission window, no mid-term commission transfer, appointment prerequisites, >90% incumbent save rates. Any GTM assuming “one-click switching” is wrong. Evidence: us-insurance-distribution-economics.
• H-INS-4 — The beachhead is never the market. Segment pools are $25–300M/yr commission — chapter one of a sequence (vertical → adjacencies → MGA ladder), presented as such in every pitch. Evidence: ai-brokerage-tam-model · insurance-beachhead-segment-sprint.
• H-INS-5 — Avoid Corgi’s home field. Startup/AI-vendor-flavored segments are owned by Corgi’s YC/VC-network distribution; prefer industrial/main-street buyers it can’t reach. Evidence: corgi.
• H-INS-6 — Embedded distribution is an experiment, not a plan. No SMB embedded program has public attach rates; NEXT scaled on ~15 performance marketers. Evidence: smb-insurance-portfolio-brokerage · coverdash.
• H-INS-7 — Segment choice runs the five-part filter. New-risk salience × supply-to-curate × premium density × reachability × name whitespace — scored, not vibed. Evidence: insurance-beachhead-segment-sprint.
• H-INS-8 — Competitor claims are unproven until a renewal cycle. No AI brokerage or AI carrier has published retention/loss data through a full cycle; treat all traction (including Corgi’s) as momentum, not validation. Evidence: ai-native-brokerages-capture-smb-distribution · corgi.
• H-INS-9 — Soft-market discipline. Since Q1 2026 rates are declining: savings pitches weaken, incumbents can match quotes — prefer coverage-confidence/requirements wedges over price wedges. Evidence: us-insurance-distribution-economics.
• H-INS-10 — In the US there is no policy-data rail; portfolio ingestion is a funnel gate, not free acquisition. Cover/Clark give the audit away because a regulator or mandate supplies the data; the US does not, so the “free insurance check-up” funnel pays a document-handoff tax (canopy-connect/InsurGrid pull or customer upload) that deserves its own conversion kill-gate. Evidence: policy-ingestion-requires-a-data-rail. (Proposed 2026-08-09; adopted 2026-08-15 with the axiom update.)

Retired

• AX-INS-5 (2026-08-07, retired 2026-08-15) — Differentiation = category creation, not speed or auditing. Original statement: we win by being the category-defining broker for a new risk — a NAMED coverage packaged from existing carrier paper for a sharply chosen segment (“the Corgi lesson”). Source: 2026-08-07-directions-nizan-saar-ai-brokerage-thesis-and-pitch · named-coverage-category-creation-wins-gtm. Retirement reason: superseded by AX-INS-7 — the 2026-08-10/15 “Ainsure” deck sessions repositioned differentiation as incentive alignment (fixed fee, neutral comparative rater) plus genuine post-bind continuity at the generalist+continuous quadrant corner; the named-coverage play is demoted to an available GTM tactic. Contributing pressure: the robotics cross partly pre-empted (RiskyTix, 2026-08-09-directions-operational-ai-and-axiom-challenge) and the 2026-08-10 research read that category creation is an entry tactic rather than a moat. Honesty note: the original revisit trigger (discovery calls disproving the named-coverage pull) was never formally run — this is a strategic repositioning, decided by user direction 2026-08-15 and logged in wiki/log.md.