A
AIOS Wiki
read-only · public mirror
Open AIOS
Wikiai-supportai-support/00-state-of-the-project.md

AI Support — State of the Project

Hand-authored·12 min read·13 sections·Last edited May 12 by initial import·View history
TL;DR

_Last updated: 2026-05-08 18:00 UTC_

TL;DR

AI Support is Axiom Zen's exploration of productizing the in-house support tool built for Toby (custom UI on FreshDesk: HITL drafts, copilot, skills/corrections flywheel, predicted-next-action, deep DB integration). The May 8 meeting with Dapper's Dan Carreiro and Kenny Zamora was the most material event of the project: Dan demoed his own weekend-built CS-with-AI tool (4 runs, 75% rejected, very early), Fin's contract just renewed for another year, Retool is being sunset by end of month, and everyone agreed the cleanest first step is to connect AZ's tool to Dapper's Zendesk and run it in parallel with Dan's tool — explicitly excluding the IT brand at the API filter level. Kenny is the most bought-in operator ("I can't wait to navigate away from Zendesk"). In parallel, Peak Money (Felipe Ribeiro) is a willing internal user pending Bart's sign-off from May 1. The strategic dossier work (Phase 0 → Phase 5) has converged on a deliberate ~60-day fork by ~July 1, 2026: spin out as a venture-scale company (Path A or D), commit to a services-product practice (Path C), or drift and lose the regulatory window. No spin-out decision has been made yet — the project is exploring.

Immediate next steps

Latest meeting (May 8 Dapper / Dan & Kenny) drives the top of this list. Items further down are still-open commitments from earlier meetings or the master dossier's de-risk sequence.

  • Figure out Zendesk access path for the AZ team (likely a seat granted by Dapper; Dan to walk through) — owner: Ben/G, Dan to enable (from: ai-support/meetings/2026-05-08-dapper-dan-kenny-summary.md, 2026-05-08)
  • Connect AZ tool → Dapper Zendesk via API, excluding the IT brand at the filter level — owner: AZ engineering (from: ai-support/meetings/2026-05-08-dapper-dan-kenny-summary.md, 2026-05-08)
  • Once connected, run AZ agent in parallel with Dan's weekend tool on the same tickets so Dapper can compare drafts side-by-side — owner: AZ + Dan (from: ai-support/meetings/2026-05-08-dapper-dan-kenny-summary.md, 2026-05-08)
  • Confirm productized version can be tailored per-customer; Kenny's NBA-Top-Shot-pack-request one-click action is the first concrete request — owner: AZ product (Simon/G) (from: ai-support/meetings/2026-05-08-dapper-dan-kenny-summary.md, 2026-05-08)
  • Dan to investigate whether full threaded Zendesk ticket history (not just metadata) can be exported via the Zendesk API; possibly run an agent-driven scrape with his API key; decide where 500K+ tickets get stored — owner: Dan Carreiro (from: ai-support/meetings/2026-05-08-dapper-dan-kenny-summary.md, 2026-05-08)
  • Loop fraud and compliance teams into a later conversation — same dashboard pattern, different use cases — owner: Kenny to surface internally (from: ai-support/meetings/2026-05-08-dapper-dan-kenny-summary.md, 2026-05-08)
  • Felipe to confirm Bart's sign-off on Peak Money adopting AZ's tool (was due "by tomorrow May 5" from the May 1 meeting; status not yet recorded in the wiki) — owner: Felipe Ribeiro (from: ai-support/meetings/2026-05-01-peak-money-summary.md, 2026-05-01)
  • Scope changes for Peak Money's use case: GitHub integration, MCP interface, chat UI for the agent, report-generation back into customer repos — owner: AZ team (G/Ben/Simon) (from: ai-support/meetings/2026-05-01-peak-money-summary.md, 2026-05-01)
  • Run the de-risk sequence Step 1: buyer-side reality check across 10 named ICPs (Polymarket, Kalshi, Paxos, Ondo, Gemini, Bullish, Sorare, Fanatics Collect, Fireblocks, Bitstamp). Single kill-the-idea-if-wrong assumption. — owner: TBD (from: ai-support/briefing/00-master-dossier.md Part IX)

Phase progress

  • Phase 0 — First-Principles Thesis: complete. Pre-research conviction documented; HITL pattern + skills/corrections flywheel + crypto vertical wedge identified before any external data. (see: ai-support/briefing/01-phase-0-thesis.md)
  • Phase 2 — Market + Unit Economics + Crypto Vertical: complete. Three dossiers cover the competitive landscape ($15B → $48B TAM, Sierra/Decagon/Fin/Forethought-acquihire dynamics), unit economics (HITL ~$2/ticket vs $4-5 human; 60-70% blended GM, NOT 80%), and crypto buyer universe (~125-160 globally, ~40-50 qualified ICP, zero horizontal logos). (see: ai-support/research/03-phase-2-market-competitors.md, ai-support/research/04-phase-2-unit-economics.md, ai-support/research/05-phase-2-crypto-vertical.md, briefing recap at ai-support/briefing/02-phase-2-brief.md)
  • Phase 4 — Multi-Perspective Reasoning: complete. Five-persona stress test (Sarah Chen skeptical seed VC, Marcus Tan visionary founder, Mark Kingston power user, Decagon CEO, Priya Bose CMO/GTM) with verdicts ranging from SKEPTICAL → CAUTIOUSLY EXCITED, all converging on "spin out cleanly or don't bother." (see: ai-support/research/06-phase-4-perspectives.md)
  • Phase 5 — Scenario Planning: partial. Only Scenario B (Vertical Squeeze, bear case, p≈33%) has been written up as its own dossier; Scenarios A (Compliance Lock-In bull, $850M), C (The Practice, services-product hybrid), and D (Acquisition Bait, $210M strategic exit) are summarized inside the master dossier but lack standalone documents. (see: ai-support/research/07-phase-5-scenario-vertical-squeeze.md and ai-support/briefing/00-master-dossier.md Part V)

Roadmap

Next 2 weeks

  • Get AZ-tool-to-Dapper-Zendesk connection working with the IT brand filtered out at the API level. (see: ai-support/meetings/2026-05-08-dapper-dan-kenny-summary.md)
  • Stand up parallel-run mode against Dan's weekend tool so Dapper can compare drafts on the same tickets. (see: ai-support/meetings/2026-05-08-dapper-dan-kenny-summary.md)
  • Close out the Peak Money decision: Bart's go/no-go, then scope MCP + chat UI + GitHub-integration changes for Felipe's pipeline. (see: ai-support/meetings/2026-05-01-peak-money-summary.md)

Next month

  • Ship Kenny's first concrete tailored action (one-click NBA Top Shot pack request) so the "tailor per customer" claim is real, not aspirational. (see: ai-support/meetings/2026-05-08-dapper-dan-kenny-summary.md)
  • Loop fraud and compliance teams at Dapper into a follow-up — same dashboard pattern, different use cases — to broaden the in-house reference beyond the CS team. (see: ai-support/meetings/2026-05-08-dapper-dan-kenny-summary.md)
  • Begin the de-risk sequence in earnest: buyer-side interviews across the 10 named ICPs, adjacent-vendor threat model (Chainalysis/TRM/Persona/Sumsub/Jumio/Fireblocks), Lorikeet head-to-head, MiCA pre-sales test on 5 EU CASPs. (see: ai-support/briefing/00-master-dossier.md Part IX, Part X)
  • Clarify with Justin what "we could work together some way" means commercially — vendor at $300-400K ACV, $0 partnership, equity, or contractor staffing. (see: ai-support/briefing/00-master-dossier.md Part X #8)

Next quarter / beyond

  • Make the Stage 1 → Stage 2 commit decision by ~July 1, 2026: spin-out (Path A/D) with separate cap table + external CEO + Compliance Engineer hire, OR commit to Path C (Practice, cap at 6 customers, no SOC 2 push). The wrong answer is drifting between paths for 8 months and missing the regulatory window. (see: ai-support/briefing/00-master-dossier.md "Strategic Recommendation Synthesis", Part VI)
  • If Path A/D: ship MiCA Travel Rule + OFAC + KYC-aware module before July 1, 2026 deadline; hire Compliance Engineer (ex-Persona/Chainalysis/Fireblocks) and a Customer Success Compliance Lead by month 4; pursue SOC 2 Type II + ISO 27001 + ISO 42001 certifications. (see: ai-support/research/06-phase-4-perspectives.md (Marcus Tan), ai-support/research/07-phase-5-scenario-vertical-squeeze.md §11, ai-support/research/05-phase-2-crypto-vertical.md §10)
  • Sign one non-Dapper anchor at >$200K ACV by Q4 2026 (target: mid-tier CEX, stablecoin issuer, or prediction market — Bitstamp / Gemini / Paxos / Polymarket / Kalshi shape). Without this, every scenario's exit value compresses. (see: ai-support/briefing/00-master-dossier.md Part VIII)
  • Watch the leading indicators that flip Scenario A→B: Decagon shipping a vertical pack or acquiring Lorikeet; Anthropic shipping a first-party claude-support-agent; Sierra adding ISO 42001 + a regulated-verticals lead. (see: ai-support/research/07-phase-5-scenario-vertical-squeeze.md §8)

Key decisions to date

  • Lead with HITL + skills/corrections + deep DB integration as the architectural bet. Validated against post-Klarna-reversal market evidence and EU AI Act Article 50 / Setzer products-liability rulings. (see: ai-support/briefing/00-master-dossier.md, ai-support/briefing/01-phase-0-thesis.md)
  • Crypto/web3 is the wedge vertical, not horizontal mid-market. Sierra/Decagon/Fin have zero crypto logos publicly; MiCA July 1, 2026 + GENIUS Act + EU AI Act Article 50 (Aug 2, 2026) create a forced-buy window. (see: ai-support/research/05-phase-2-crypto-vertical.md, ai-support/briefing/02-phase-2-brief.md)
  • Dapper is lighthouse, not market. Real ICP is ~40 mid-tier CEXs, stablecoin issuers, prediction markets, and collectibles peers. Pricing should be compliance-grade ($300-400K ACV), not Dapper-shape ($150-200K). (see: ai-support/research/06-phase-4-perspectives.md — Marcus Tan and Priya Bose perspectives; ai-support/research/05-phase-2-crypto-vertical.md)
  • Pricing model: hybrid (floor + meter + ceiling), not pure per-resolution. Pure per-resolution backfires above ~5K resolutions/mo. (see: ai-support/research/04-phase-2-unit-economics.md)
  • Next concrete step with Dapper: parallel Zendesk integration excluding the IT brand, running AZ's agent side-by-side with Dan Carreiro's weekend tool on the same tickets. (see: ai-support/meetings/2026-05-08-dapper-dan-kenny-summary.md)
  • Fin is locked in at Dapper for another year (cancellation notice deadline ~May 5 was missed); mitigation is to stop sending tickets to Fin so the only cost is the sub floor. (see: ai-support/meetings/2026-05-08-dapper-dan-kenny-summary.md)
  • Retool is being sunset at Dapper by end of month; Ralph is building a vibe-coded in-house replacement. AZ's tool will need to coexist with that moving surface and stays read-only by design (writes hand off to a human). (see: ai-support/meetings/2026-05-08-dapper-dan-kenny-summary.md)
  • Peak Money is a willing second internal user pending Bart's sign-off; Felipe Ribeiro is the operator-of-record. Likely first deployment after Toby itself. (see: ai-support/meetings/2026-05-01-peak-money-summary.md)
  • The unhired role that determines the outcome is Customer Success Compliance Lead (Marcus Tan named it; nobody at AZ has drafted the JD). Required by month 4 for any venture-scale path. (see: ai-support/research/06-phase-4-perspectives.md, ai-support/briefing/00-master-dossier.md Part III)

Open questions / blockers

  • Spin-out commit, by ~July 1, 2026 (60-day fork). Are AZ partners willing to cede control with separate cap table, separate brand, external CEO? The "drift between paths" failure mode is the highest-probability bad outcome across all scenarios. (see: ai-support/briefing/00-master-dossier.md Part VI)
  • Compliance Engineer (ex-Persona/Chainalysis/Fireblocks) hired by July 2026? Single highest-leverage hire to keep MiCA forced-buy window viable. (see: ai-support/research/07-phase-5-scenario-vertical-squeeze.md §11)
  • Dapper "could work together some way" — what does that actually mean commercially? $300-400K ACV vendor relationship, $0 partnership/equity discount, or contractor staffing? Not yet clarified with Justin. (see: ai-support/briefing/00-master-dossier.md Part X #8)
  • Zendesk ticket-history export feasibility for Dapper. Dan Carreiro to investigate whether full threaded history (not just metadata) can be pulled via Zendesk API; storage cost at 500K+ tickets is non-trivial. (see: ai-support/meetings/2026-05-08-dapper-dan-kenny-summary.md)
  • What replaces Retool day-to-day for write actions (2FA reset, score adjustments, etc.) once it's sunset? Ralph's in-house tool is moving but timeline is unclear; AZ's tool is read-only by design. (see: ai-support/meetings/2026-05-08-dapper-dan-kenny-summary.md)
  • Commercial framing of the AZ → Dapper handoff — internal tool, customizable product, or guided build? Not decided in the May 8 meeting. (see: ai-support/meetings/2026-05-08-dapper-dan-kenny-summary.md)
  • Adjacent-vendor threat model (Chainalysis, TRM, Persona, Sumsub, Jumio, Fireblocks). Red team flagged these as the most credible substitute threat — they already own the buyer relationship and have AI roadmaps. Not yet de-risked. (see: ai-support/briefing/00-master-dossier.md Part IV, Part X #2)
  • Anthropic / OpenAI first-party support agent risk (~45% probability by end of 2027). The platform-eats-the-app-layer dynamic is the trigger event for Scenario B. (see: ai-support/research/07-phase-5-scenario-vertical-squeeze.md §10)
  • Buyer-side reality check across 10 named ICPs not yet run. Step 1 of the de-risk sequence and the single kill-the-idea-if-wrong assumption. (see: ai-support/briefing/00-master-dossier.md Part IX)
  • Three of four Phase 5 scenarios lack standalone dossiers. Only Scenario B (Vertical Squeeze) has been fleshed out; A (Compliance Lock-In), C (The Practice), and D (Acquisition Bait) live only as summaries in the master dossier. Worth deciding whether to invest in the missing three.
  • Note on doc pairings (not duplicates, but flagged for the operator): each meeting has a verbatim transcript and a distilled summary. They are intentional pairs, not duplicates — the summaries link back to the transcripts. No deletion recommended. (2026-05-01 Dapper transcript ↔ workflow summary; 2026-05-01 Peak Money transcript ↔ summary; 2026-05-08 Dapper transcript ↔ summary)

Doc index

Briefing

Meetings

Research