/ AI Roadmap Internal
AI Roadmap

What we're building

Current state and what's coming — internal systems and client-facing products. For the full technical breakdown of what's live, see Infrastructure Status.
1

Status now

System Status What it does What's not working
Internal
NHH — hilda-scheduler Live Always-on Fly.io scheduler. ClickUp task prep, Gemini notes → ClickUp tasks, Source of Truth updates, Slack approval flow, data passing to the Burrow. 12 cron jobs commented out (daily prep, weekly work plan, planning reports, etc.) — confirmed off, not coming back. post-meeting-email, mid-month-checkin, and patch-source-of-truth task files exist but are never imported. Source of Truth cron is live (Friday 12pm).
Cowork (Claude) Live Primary tool for internal deep work — briefs, strategy, anything that needs full context and active thinking. —
Google Ads MCP Live ~95% coverage. Platform queries working well. Ownership confusion — gets mixed up across client accounts.
Meta Ads MCP Limited Connected and functional for basic queries. App approval pending for full API access. ClickUp task open.
GA4 / GSC / Sheets / WhatConverts / Semrush Live All connected. GA4: all 17 client IDs populated. —
GBP MCP Building All client IDs mapped. GCP API approved June 2026 after 4-month wait. Integration testing to follow.
Client-facing
Slack — data access Building Clients ask Hilda about their campaign data in Slack. Powered by MCPs. Dependent on MCP polish. Not yet rolled out to clients.
MM Autopilot Live Keyword review and planning for mechanics. Feedback loop functional. Keywords pausing too early. Missing data points (impression share, QS, LPE). Campaign build out of scope.
HH Autopilots (Google + Meta) Live Parallel approval cards to internal + client Slack. Race guard dedup working. HH Autopilot not reviewed at Byron — needs dedicated session.
The Burrow Live (core) Client portal — 14 clients seeded, 3 months of planning reports and quarterly plans ingested, autopilot approval panels live, settings tabs (briefing, brand files, global rules) wired to Supabase, persistent memory per client. 36 API routes, 18 migrations deployed. Chat wiring is current priority. Chat routes exist (/api/chat/[slug], /api/hilda/chat) but backend wiring unconfirmed — needs verification. NHH deliverable pipeline (PRs #102, #103) open but mid-month-checkin task file not yet imported in NHH. /preview/burrow/* demo pages are dead fixture data (Klaylife), not live.
2

Decision rule

NHH ClickUp prep, planning session support, scheduled automations. Runs always-on.
Cowork Internal deep work, briefs, strategy. Needs full context and active thinking.
Burrow Client-facing deliverables — HTML docs, structured assets. Anything the client sees built by AI.

Decision tree complete — see How we use AI. Summary: Cowork does all thinking and analysis. NHH handles data passing and known automations only.

3

Roadmap

31 May 2026
AI roadmap doc
This page. Single source of truth for everything we're building — reviewed and updated at the start of each month.
Done Teddi Roadmap
NHH vs Cowork decision tree
Clear written rule agreed in AI WIP meeting, 1 June 2026. Cowork = all thinking, analysis, and complex deliverables. NHH = data passing, known automations, client-facing tools. See How we use AI page.
Done Josh NHH
PR #77 MCP init — merge ASAP
Pending merge. Blocking MCP init improvements.
Pending Teddi NHH
June 2026
Planning session support workflow
After every client planning session: Hilda generates ClickUp task list → Teddi or Josh reviews same day → marks High priorities → automated client email summarising what's in their hours and what costs extra.
  • Human review step is the new gate — must happen same day as session
  • Future: integrate with Burrow for richer output
Building Teddi NHH
MCP polish — step out all MCPs clearly in NHH
All MCPs need to be explicitly spelled out in NHH so Hilda knows what to call and when. If it's not spelled out, she won't ask for it.
  • Brief #6: add missing data points — impression share, QS, landing page experience
  • Enable GBP API on GCP project 791805910145 (My Business Account Management API)
  • Meta Ads Library API access — resolve app approval, unblock full Meta MCP capability
Building Teddi MCP
NHH open PRs and briefs
Clear the open build backlog before July rollouts.
  • NHH PR #102 — trigger-deliverable scaffold (open)
  • NHH PR #103 — mid-month checkin Phase 2 (open) — note: task file also needs importing in app.js before it will fire
  • HH Google Ads PR #11 — clickup-subtask-id-capture
  • HH Meta Ads PR #20 — clickup-logging
  • Brief: cache diagnostics — ready to run
  • Brief: ClickUp llms.txt update — ready to run
  • NHH PR #60 — Meta TOFU awareness (merged)
  • NHH PR #61 — Meta action specificity (merged)
In progress Teddi NHH
MM Autopilot — June test month
2–3 test clients running Autopilot through June. Fix blockers first, validate, build feedback loop. Full rollout to all 8 MM clients by 1 Jul.
  • Fix keyword pausing threshold before test starts
  • Add missing data points (impression share, QS, landing page experience) via brief #6
  • Feedback loop: inline Slack reactions (✅/❌) + monthly text check-in
  • Confirmed out of scope: campaign build — that's a separate campaign builder project
Starting Jun Teddi Autopilot
HH Autopilot — dedicated review session
HH Autopilot wasn't reviewed at Byron. Needs its own session to assess current state, what's working, and what needs fixing.
Schedule Josh + Teddi Autopilot
Burrow — wire chat to functional backend
The Burrow chat UI exists but is connected to dummy/placeholder data. Wire it to the real Supabase backend and NHH orchestration so clients can chat with their actual campaign data.
  • Front end is built — no backend config exists yet, building from scratch
  • Supabase holds all client config, channel data, and persistent memory
  • Data warehouse (daily 6am Supabase sync) is the data source — not live MCP calls
  • Once wired: clients can chat with their Google Ads, GA4, and SEMrush data in the Burrow
  • Slack chat integration (persistent memory parity) deferred to phase 2
Building Teddi Burrow
1 July 2026
MM Autopilot — all 8 clients live
Full rollout after June test month. All clients live, feedback loop operational, June issues resolved.
  • Review June test data before rolling out to remaining clients
  • Client-facing explainer ready: what Autopilot does, how to read recommendations
  • Longer-term opportunity: package and sell standalone to mechanics ~$200/month
Building Teddi Autopilot
Client Slack data access — rollout
Clients can ask Hilda about their campaigns, spend, and results directly in Slack. Dependent on MCP polish being complete in June.
  • Meta app approval must be resolved before Meta Ads queries work for clients
  • All MCPs stepped out in NHH first — so Hilda knows what to call
Building Teddi Slack
31 August 2026
The Burrow — client-facing asset building MVP
Client portal is live. Internal deliverables (planning reports, quarterly plans) are manually created in Cowork and pushed into the Burrow — this workflow stays manual for now. The build focus has shifted to client-facing chat and deliverables.
  • Wire Burrow chat to functional backend orchestration — current top priority
  • Test data warehouse integration (Supabase, daily 6am sync) end-to-end in the Burrow
  • NHH → Burrow deliverable pipeline (PRs #102, #103 open)
  • Persistent memory per client now live (Supabase-backed)
  • Slack persistent memory deferred to phase 2
  • "New report" button (one-click report trigger from Burrow UI) is the end-state goal — not yet built
Live (core) Josh Burrow
Parked
Daily prep via NHH
Confirmed off in code audit (June 2026). Includes: daily-project-briefing, teddi-day-ahead, teddi-week-ahead, josh-day-ahead, josh-week-ahead, monday-funday-prep, friday-wrapup-prep, weekly-work-plan, weekly-campaign-flag, weekly-client-report, weekly-data-snapshot. All commented out in app.js — not coming back to NHH.
Parked NHH
Hilda PM + Hilda Review automations
Not working well enough. Revisit Q3 2026.
Parked NHH
MM Autopilot as standalone product
Strong opportunity — package and sell to mechanics for ~$200/month. Not actionable until core autopilot is solid. Revisit Q3.
Parked Autopilot
Killed
NHH templated document generation
Not a use case for NHH. Complexity without payoff. This belongs in the Burrow. Killed May 2026.
Killed NHH
Rogue monthly report generator
Disabled May 2026.
Killed NHH
Competitor Watch
Discontinued June 2026. Quality was not good enough to justify the build — outputs were not useful enough for internal or client use. AI Digest remains (Cowork). Competitor analysis handled ad hoc in Cowork when needed.
Killed NHH