Skip to content

Riptide Roadmap

ROADMAP | DERIVED | Updated 2026-04-08 | Owner: AI Team / Riptide

Key Timeline Anchors

Event Target Date Status Dependency
M1 — The Loop March 27 (Day 7) In progress Navid + Forte testnet
M2 — The Conversation April 3 (Day 14) Blocked on M1 M1 Go decision
M3 — Pinky and the Brain April 18 (Day 24) Blocked on M2 M2 Go + Dete review
flow.com/ai landing page April 2026 Parallel track Marketing team
AI Showrunner (Peak Money) April 2026 Directive issued Alan Carr + Yusang

Now Shipping

M1: The Loop

Cadence contract on testnet with Forte-native scheduling. Each cycle reads storage from 3+ accounts, reads its own prior state, emits structured observation event. Off-chain oracle catches event, calls LLM, submits response transaction.

Criteria Threshold
Must 24h+ autonomous runtime
Must Oracle round-trip < 60s
Must < 5% dropped cycles
Must Deployed to testnet (not emulator)
Should Compute deltas between cycles
Should LLM shows visible adaptation
Should Basic retry logic
Kill if > 20% dropped cycles or > 5 min latency consistently

Owner: Navid (full-time). Roham + Navid decide Go/No-Go.

flow.com/ai (Parallel)

  • Landing page with "Where Agents Live" positioning
  • Forte differentiator explained
  • Links to existing dev tutorials (Eliza, AgentKit, MCP, Claude Code)
  • "Why Flow Is Agent-Native" blog post
  • awesome-flow-ai GitHub repo
  • Owner: Marketing. Not Navid.

AI Showrunner — Applied AI for Peak Money

  • CEO directive (April 7, 2026): autonomous experiment loops for waitlist rollout
  • Agent designs email copy, sends invites, tracks conversion data, designs next experiment based on results
  • Peak Money architecturally ready: Docker sandboxing, backwards/forwards-compatible APIs
  • Owner: Alan Carr + Yusang Lee

Next Up

M2: The Conversation (blocked on M1 Go)

Second agent contract on testnet with independent Forte schedule and independent LLM persona. Shared state object (bulletin board) both agents read and write.

Criteria Threshold
Must 24h+ both agents running
Must Distinct LLM personas
Must Interleaved events readable as back-and-forth
Must At least one genuine adaptation instance
Should Agents build toward something together
Should Explicit cross-agent references
Kill if Two independent monologues, or state conflicts corrupt > 5% of cycles

Go/No-Go: Roham + Navid + Dete. Day 14.

M3: Pinky and the Brain (blocked on M2 Go)

Single web page: live conversation feed, Flowscan links on every entry, "How This Works" in 3 sentences, running time counter and cycle stats.

Criteria Threshold
Must Running 7+ days at launch, zero manual intervention
Must No wallet connection or Web3 friction
Must Open Graph tags for social sharing
Should Auto-refresh, visual distinction between agents
Should Historical scroll, mobile-responsive

Narrative Milestone

M3 cannot kill the project. If the tech works (M1+M2), iterate the wrapper until the narrative lands.

Go/No-Go: Roham + Dete. Day 24.

Post-M3: External Launch

  • Post publicly. Pin in #ai-riptide-flow
  • Begin flow.com/ai integration
  • Stranger test: show URL to 5 non-technical people before launch
  • This becomes the link in every investor/partner/developer conversation about Flow + AI

Blocked / At Risk

Navid is the sole engineer on M1-M3. Alex Hentschel provides 2-3 hours of architecture review. Design support arrives at M3 only. No backup plan documented.

Flowscan Scheduled Tx Visibility

Navid flagged: "super hard but doable." Needed by M1 Day 5 for the event log to be readable. Owner: Michael/Jordan.

Alex Architecture Review — NOT YET SCHEDULED

Needed before M1 Go/No-Go decision. Not on calendar.

Dete Review — NOT YET SCHEDULED

Needed before M3 external launch. Not on calendar.

flow.com/ai Design + Content — NOT YET ASSIGNED

Marketing team owns. No assignment visible.


Recently Shipped

Item Date Notes
x402 (agent payments on Flow) March 2026 Agents can pay through Flow. Confirmed production.
Forte scheduled transactions Production FYV-EVM Worker running 1+ month autonomously on mainnet.
Agent contracts (emulator) Pre-Riptide Navid prototype: agents as Cadence contracts, multi-agent pipeline orchestration, oracle bridge, output as Resources.

Kill Criteria Are Pre-Committed

Each gate has explicit kill criteria. Pre-commitment: if kill criteria are hit, stop and rethink. Do not rationalize "one more iteration" without a clear hypothesis for what changes.

Gate Kill If
M1 Oracle cannot close the loop (> 20% drops or > 5 min latency)
M2 Agents do not interact meaningfully through shared state
M3 Cannot kill — iterate the wrapper