Strategy note, updated 2026-07-31. The "why this exists and why it can work" argument behind CRYPTO_REVIEW.md, CODE_GRAPH.md, and REPUTATION_MODEL.md.
Thesis
Email is the last major internet protocol still running its 1982 core. It froze not because nobody tried to fix it, but because its topology forbids in-place evolution. The only strategy that has ever moved infrastructure this entrenched is the coexistence overlay — run a strictly-better parallel protocol and bridge down to the legacy swamp. AfterSMTP is that play, and three forces in 2024–2026 have, for the first time in 40 years, made the adoption window real.
Why email froze while HTTP evolved and FTP died
Same era, three outcomes — and the difference is topology, not protocol quality.
- HTTP got SPDY → HTTP/2 (2015) → QUIC/HTTP/3 (2022) in a decade. It is point-to-point: a client and a server negotiate per connection (ALPN, in-band), and a handful of players control both ends (Chrome + Google servers, Cloudflare at the edge). Upgrade one client and one server, fall back gracefully, done.
- FTP was quietly euthanized — browsers removed it in 2021. Also point-to-point, and every FTP session was an island, so each could migrate to HTTPS/SFTP independently. No global graph depended on FTP-to-FTP interop, so the islands could be abandoned one at a time.
- Email is an N-to-N federated mesh where the slowest participant sets the protocol version for everyone, forever. Your MTA must interoperate with a Sendmail box from 2003 that will never be patched, so the lowest common denominator wins by default.
SMTP's only evolution mechanism is ESMTP (EHLO extensions), and it is a one-way
ratchet: you can add STARTTLS, SIZE, SMTPUTF8, but you can never remove or
change the core, because the next hop might be ancient. Every security property
since 1982 — SPF, DKIM, DMARC, DANE, MTA-STS, ARC, BIMI — is a bolt-on graft on a hull
never meant to carry it. The freeze is a mathematical property of federation without
coordinated version negotiation, not inertia.
Proof point: JMAP (2019) is a genuinely better IMAP successor that still can't get traction even on the less-federated client side. The freeze reaches everywhere the network effect does.
Why better filtering / AI will never close it
The architecture externalizes the cost of sending onto the receiver: sending is free, filtering is expensive. Gmail's filter is extraordinary and spam still lands — because AI drove both curves down at once. The same models that filter also let an attacker generate fluent, personalized, context-aware phishing at zero marginal cost. You cannot win an arms race whose structure guarantees the attacker's costs fall as fast as your defenses improve. That asymmetry lives in the protocol, not the filter — which is why a better filter was never the answer, and why touching the foundation is the only real move.
Why past redesigns died
Not bad tech — network effects. Email's federation is its moat: four billion interoperating mailboxes are an incumbent nothing dislodges by merely being better. Every "email killer" either balkanized into a walled garden (the messengers carved off use cases but never replaced the universal federated inbox) or died waiting for a critical mass that the incumbent's interop inertia never allowed.
The only playbook that works: coexistence overlay
You cannot upgrade the mesh in place — so don't. Run a parallel protocol that is strictly better, advertise it, and bridge down to legacy only when you must. This is literally how HTTP/3 won: QUIC didn't replace HTTP/1.1, it runs alongside it, advertised via Alt-Svc and now SVCB/HTTPS DNS records, with graceful fallback.
AfterSMTP maps onto this exactly:
| HTTP/3 coexistence | AfterSMTP equivalent |
|---|---|
| QUIC overlay alongside HTTP/1.1 | AMP-over-QUIC alongside SMTP |
| Alt-Svc / SVCB DNS discovery | _aftersmtp._tcp TXT record |
| Graceful fallback to HTTP/1.1 | Legacy SMTP on/off-ramp bridge |
| Opt-in, per-origin | Opt-in, per-domain |
Current implementation reality
The thesis is ahead of the implementation. The repository currently supports a controlled-pilot path: native AMP ingress, SMTP-to-local-inbox bridging, durable admission controls, and idempotent legacy retries. It does not yet provide a public HTTP/3 API, recipient request/approval APIs, or a wired outbound SMTP relay. The bridge is strategically central, but it must be fail-closed on TLS and tested as an Internet-facing MTA before it can carry the adoption argument in production.
Strategic consequence: the least glamorous component in the repo — the legacy bridge — is not a compatibility feature, it is the entire adoption strategy. You don't replace email; you wrap it, run alongside it, make the crypto-native path strictly better for whoever opts in, and onboard one domain at a time. Prioritize and harden the bridge accordingly (it's also the most attacker-exposed surface — see CRYPTO_REVIEW.md next steps).
Why now (the window that wasn't open before)
For 40 years the answer to "redesign email" was "you can't move the ocean." The ocean is, for the first time, sloshing:
- Post-quantum forces a crypto migration anyway. Everyone has to touch their key infrastructure for ML-KEM/ML-DSA regardless. Riding that migration is far cheaper than asking operators to change for change's sake. (AfterSMTP already has the hybrid primitive — see CRYPTO_REVIEW.md.)
- AI phishing makes authentication + reputation existential, not optional. When anyone can mass-produce flawless targeted phishing, strong sender identity stops being hygiene and becomes survival.
- Operators are already conditioned. The 2024 Gmail/Yahoo bulk-sender mandates trained the entire ecosystem to accept "adopt cryptographic auth (SPF+DKIM+DMARC) or your mail doesn't deliver." The muscle memory for "deliverability requires crypto compliance" now exists — AfterSMTP extends a reflex operators already have.
- Identity cost is now cryptographically enforceable. PoW/stake-bonded DID registration makes the Sybil-resistant, reputation-enforcing identity layer (see REPUTATION_MODEL.md) practical in a way it wasn't in the Hashcash/Penny Black era.
One-paragraph version (for the pitch)
Email is the only major internet protocol still frozen at its 1982 design, because unlike HTTP or FTP it's a federated mesh where the oldest server sets the rules for everyone — so it can only be grafted onto, never redesigned. Better filters can't fix it because the cost asymmetry that makes spam profitable lives in the protocol, and AI made both sides of that race faster at once. The only way to move infrastructure this entrenched is the HTTP/3 playbook: a strictly-better parallel protocol with a legacy bridge, adopted one domain at a time. Post-quantum migration, AI-phishing pressure, the 2024 deliverability mandates, and cryptographically enforceable identity cost have opened that window for the first time in 40 years. AfterSMTP is the overlay.