Builder Vision Handoff — St. Pierre Digital
Prepared 2026-07-13 · For: Collins (first Builder) · From: Akash · Status: source of truth for what we're building and where every system stands today.
How to read this document
- The vision is fixed. The "how" is yours. Each section below states the end state and where we are today. You create the plan, sequence the work, and execute independently.
- Definition of the whole company: an AI-agent-run agency for foundation & waterproofing contractors. Managed by scorecard, not task lists. Headcount stays ≤3 humans. Akash's time goes to sales and his own face/voice content — everything else runs on agents you own.
- Your scorecard: the system runs better every week — 3+ Akash hours removed per week by week 4; 1–2 recurring jobs automated per week. You are judged on system performance, not tasks completed.
- If this page conflicts with an older doc, this page and the operating brief (2026-07-10) win.
The two pipelines — how ALL work in the business splits TOP FRAME — set by Akash 2026-07-13
Every piece of work belongs to exactly one of two pipelines. If you can't say which one a task serves, it probably shouldn't exist.
| SALES PIPELINE (people) | PROJECT PIPELINE (outcomes) |
|---|
| Unit | A person or a business | A project / an outcome / a number |
| Goal | Move every unit RIGHT — toward active client with no churn risk. Relationship management, always. | Ship outputs that move a metric. Ideate → produce → distribute → analyze → iterate. |
| Owner | Akash — all of it | Collins + AI agents |
| Includes | Dream100 / account-based relationships · cold calling, cold email, LinkedIn, IG (cold SMS: consented form-fillers ONLY — TCPA) · closes · client relationships + churn saves · recruitment — sourcing, onboarding, and ramping outbound callers and builders is people-work and lives here | The daily content slate · ads production flow · sales-system automation · reporting systems · finance/sales/marketing projection systems · the project-management loops around all of it |
| Managed by | Pipeline stages (lead-in → booked → held → proposal → won-cash → healthy client) — the brain already tracks these person-stages | The scoreboard — lead/lag metric pairs per lane (see doctrine below) |
The delegation contract (why this doc says outcomes, not tasks)
- Akash delegates outcomes, ideally as numbers — big ones. He does not manage the middle.
- The AI floor: anything smaller than a number goes straight to AI, not to a builder. If a builder would just be a middleman between Akash and an AI agent, the builder is the wrong tool — Akash tells the AI directly. Your job is the layer AI can't do alone: running the machine, wiring the loops, QA, judgment.
- Metric doctrine — every lane owns one lead/lag pair: the lead metric is volume (objective — e.g. creatives shipped, ads staged, posts published). The lag metric is quality (conversion rate, cost per result — the market's verdict). Volume is yours to control today; quality is what volume + testing buys you. We build AI agents that optimize the lag metric using lead-metric volume and testing.
- Guardrails are format contracts: before an AI loop runs on a format, the first version gets locked as a spec — e.g. "what does tweet-style #1 look like" — then agents iterate inside it. Locking those contracts is real builder work, not overhead.
- Hook testing runs on the format grid: same hook tested across static backgrounds (white/black) AND motion backgrounds (loop B-roll) in the tweet / plain-text / notes-app / iMessage styles; carousels are the long-form of the same. Further down: Akash records green-screen reels on the same hooks — best ones get the screen removed and run again as talking-head.
- Email split: newsletter-list email is auto-send. Anyone in the SALES pipeline never gets auto-sent — pre-draft it and stage it as a manual action for Akash. This is the person/project boundary in practice.
What stays Akash-only — forever
- Money levers — ad budgets, spend activation, purchases, payroll. Agents stage; Akash flips.
- Face & voice — recording reels, YouTube, VSLs. Agents script, cut, fan out, publish.
- Human relationships — sales pushes, closes, client escalations, hiring/firing. First 10 closes are personal.
- The bar — a ~10-min weekly review: approve projections, react to misses, tighten stretch.
- No outbound is ever sent as Akash by any agent or builder. Live message tests go only to the owned test phone ending 5477.
The 7 systems — end state vs. today
1 · Autonomous media production FURTHEST ALONG
- End state: a no-founder content engine. AI drafts hooks/scripts/statics from performance data, fans every recording into statics, carousels, tweets, IG, newsletter, YouTube cuts with locked UTMs, and publishes on schedule. Akash only records flagged face/voice items and approves waves.
- Today: organic FB+IG posting is live at one post/day autonomous; statics system production-hardened (latest run QA'd 12/12 PASS, cleared for paused Meta build); 11 carousel SVG templates; 3 indexed B-roll libraries (~150 clips); locked copy contract + 41 approved headlines; motion renderer + micro-VSL engine working; 44 VSL variants in flight; full skill stack exists (content-fill, static expanders, transcript-to-statics, reel/skit/video engines).
- Scoped agents: 2.1 Content-strategy agent (gate: Akash approves waves), 2.2 Repurposing agent (no separate gate), Creative-supplier delivery agent (report-only).
- Rule update (Akash, 2026-07-13): AI-UGC customer-style video IS in scope for the daily slate. Founder talking-head stays parked. (This supersedes the "real face only" line in the 07-11 Agency OS doc.)
- The daily output slate (Collins runs this): daily newsletter · daily organic posts · daily AI-UGC video · daily carousel · daily statics in tweet-style + note-style. The inputs are templatized: hooks identified (hook contract v2, 2026-07-13 — every hook names foundation & waterproofing), B-roll test library built, creative formats + messaging templatized, offer locked. The job is to RUN the machine daily and then automate the running.
- Gaps: the daily cadence isn't running yet; hooks are hand-picked into the queue (self-improving loop unbuilt); queue auto-filler approval gate in flight.
2 · Autonomous sales system BLOCKED ON TRAFFIC
- End state: AI runs demand end-to-end — media-buying agent creates/kills/scales ads inside caps; setter agent runs speed-to-lead, follow-up, confirmations, no-show detection; sales agent builds pre-call briefs and custom demos; funnel = form → custom demo → sales-managed booking ($1 gate is removed; VSL runs as feed creative, not a standalone page).
- Today: zero paid traffic is flowing — this is the single biggest constraint in the whole business. Cold email is warming (3 accounts, ready ~Aug 1). GHL sales system live. CAPI to Meta staged one approval from firing; retargeting audiences staged.
- Scoped agents: Media-buyer (auto-with-gate: reversible moves inside caps, ledgered), 2.3 Sales-projections agent (weekly close/risk calls), 2.7 Outbound + reply-triage agent (drafts gated), setter/qualification behaviors defined.
- Gaps: outbound-batch capability missing (top uncovered workstream — 10 drafts stalled on trivial gates); TCPA hard block on cold SMS (consented form-fillers only — there was a live incident); custom-demo engine not yet built.
3 · Autonomous client onboarding + delivery YOUR FIRST BUILD — pulled forward 2026-07-13
- Update (2026-07-13): Akash pulled the trigger early — this system is your first delegated build, starting Jul 14, not gated on client #1. Build SOW:
SOWs/sp-client-onboarding-automation-sow-2026-07-14.html. Week 1 = shadow one manual onboarding, then automate stage by stage. Progress posts to Slack #client-onboarding.
- End state: client signs → Provisioner agent builds the GHL subaccount from snapshot, custom values, bot persona, channel wiring, verify checklist — Akash does the kickoff call only. Delivery = 5 qualified inspections/week/client with the credit ladder auto-applied on miss. A lead-flow watchdog scores every bot conversation nightly and self-heals data errors.
- Today: all schema exists (clients, bot_qa, capi, ad_metrics, client_health — SP clients are new rows, not new tables); onboarding/GHL/QA skills exist as the quarry; overnight autonomous-execute runner exists in shadow mode.
- Scoped agents: #1 Provisioner (auto-with-gate, trigger: client #1 signs), #2 Lead-flow watchdog (auto, reversible writes only), 2.4 Client delivery + CS agent, 2.5 Onboarding agent.
- Gaps: deliberately unbuilt until triggers fire — nothing gets rebuilt before its client-count trigger. QBI-qualification agent is net-new with no predecessor. QBI definition per client is written once by Akash at onboarding, then enforced by agents forever.
4 · Autonomous reporting + client experience DRAFTS-ONLY BY DESIGN
- End state: weekly per-client health tier computed daily, client updates drafted every day and auto-sent once trusted, complaints auto-routed, Loom outlines drafted, one roll-up Review Panel where every department's target/actual/plan is browsable. Acceptance test: Akash off-grid 7 days and everything still runs.
- Today: cockpit + briefing + capture live; 74 dashboards registered; client-health, csm-inbox, complaint-capture, escalation, scorecard skills all live from the EyeFly era.
- Scoped agents: #6 Account manager (drafts only — sending stays Akash's), #7 Scoreboard (report-only, includes hire-trigger capacity math), Phase-3 Review Panel.
- Gaps: dashboard sprawl (99 CF pages vs. target 13; Cloudflare 100-page limit hit); client-facing dashboards deliberately deferred; client-health should become a computed daily score instead of hand-assembled.
5 · Autonomous data tracking + attribution CORE LIVE
- End state: person-level attribution rivaling Hyros — any lead traced ad-click → VSL watch → application → booking → payment across devices; every test result feeds a learnings table that every drafting agent reads before generating; server-side CAPI closes the Meta optimization loop.
- Today: attribution core live — 7,352 touches stitched into 3,993 persons with a live timeline dashboard; CAPI staged (match quality proven, one approval from firing); brain = Postgres, 254 tables, sole source of truth, 982 ledger rows in 30 days.
- Scoped agents: #3 Attribution keeper (auto, mostly deterministic — absorbs capi-backfill/health/test-fire), 2.8 CRM-hygiene agent (no gate).
- Gaps: email engagement tracking missing entirely; per-lead organic engagement missing (aggregate only); CAPI v2 written but unbuilt; n8n Cloud hit its execution quota (runaway webhook fired 1,933× — needs a dedupe guard + Akash billing fix); n8n has zero registry/monitor (its watchdog died silently for 26 days — fleet-health now catches this).
6 · Autonomous project management THE CORE MISSING LOOP
- End state: the company is managed by scorecard, not task lists. Every department agent sets weekly targets with confidence + rationale (Sunday), actuals auto-compare (Friday), learnings get logged, hit-rate auto-tightens from 100% toward 75% to force stretch. Akash is pulled in only on misses. Founder ops time ≤10 min/week.
- Today (corrected 07-13 against the live brain): capture/TaskOS live (527 handoffs, 417 tasks completed, 982 ledger rows in 30d) — and the Loop Engine Phase 1 is BUILT and running: 8 department lanes registered with lead/lag metrics, weekly targets set (wk 07-13), the Sunday-projections rite logged, and 12 daily metric keys flowing fresh. What's missing is the rendered scoreboard surface and history (~1 week of actuals so far). See the Daily Operating Rhythm doc for the full state.
- Gaps: capture still mostly manual; ~1,050 not-started tasks in backlog (hygiene debt); session-launcher skill needed (26 handoffs/30d hand-assembled); loop acceptance test = two consecutive weeks of locked targets + logged actuals + exceptions-only brief.
7 · Builder pipeline YOU ARE HERE
- End state: builders own the agent layer — prompts, workflows, dashboards, QA, rollback, data quality — with an idea→scoped→built→QA→deployed→documented pipeline and an AI-agent registry. Rule: automate before you hire — a function moves to a human only after its backend runs reliably.
- Today: board + 6-week Gantt live at team.stpierre.ai/builderpipeline; access model designed (per-builder identities, one shared brain, never one shared identity); three autonomy tiers defined (see rules below); 37 EyeFly skills mapped into 7 St. Pierre agent loops (the "quarry" — graduate one-by-one, delete nothing).
- Gaps: approved build order for new capabilities: fleet-health → recruiting-pipeline → outbound-batch → n8n registry+build skill → session-launcher (all human-gated on sends); dual skill-tree collision partially merged; n8n invisible to the registry.
Rules of the road (non-negotiable)
Autonomy tiers
- Auto — reversible + ledgered. Act, log, move on.
- Auto-with-gate — do 100% of the work, queue exactly ONE human task. Never an approval queue.
- Report-only — read and recommend; no writes.
- Every external write goes through the change ledger with a rollback path.
Locked surfaces — do not touch
- SMS worker (live TCPA compliance incident)
- Meta ad account (frozen; budgets are Akash-only)
- Command-center launcher & scorecard tables
- Nothing customer-facing changes outside an approved test path; tests only to the …5477 phone
Current-doc map (what wins over what)
| Read this | Status |
|---|
| Operating brief — team.stpierre.ai/builderpipeline/operating-brief/ (2026-07-10) | CANON — wins over everything older |
| This handoff (2026-07-13) | CANON — vision + current state per system |
| Daily Operating Rhythm (2026-07-13) | CANON — how every day runs: 1–2 PM daily meeting with Akash, 2–4 PM independent block, 4 PM EOD report to Slack #daily-reports, the live scoreboard + the metrics you own, logoff hygiene |
| Custom Audit Engine SOW + Onboarding Automation SOW (builderpipeline/sow/) | Active scope |
| Builder board (team.stpierre.ai/builderpipeline/) | Live, but FOUND-1 scorecard card is stale — agents build that now, not you. Flag, don't build. |
| 6-week Gantt / Jun-30 content-operator packet / phase-1 assignments (Jul 4) / projects-pool (frozen) | STALE where they conflict — $1-gate phasing retired; content-operator framing superseded |
Where the gated work stands — 2026-07-13 (status only, nothing approved)
Akash reviews each of these with Collins directly. None are approved as of 2026-07-13. Do not build, render, send, or activate any of them until his feedback lands. Full scope + context: 01-Projects/St_Pierre_Digital/SOWs/pipeline-first-onboarding-sow-2026-07-13.html.
| Item | Where it is today | Status |
|---|
| B-roll look-lock | 3 looks built; 48-video batch + AI pilot (2,000-credit ceiling, Higgsfield ~1,210 credits) both held on this call | Waiting on Akash |
| Carousel pilots | 3 pilots staged (Receipt→Reframe→Rule→Proof→Belief), on the approve/revise page; no scale until sign-off | Waiting on Akash |
| Newsletter first send | "The Backlog Brief" wired to send gate; 4 Akash-only checklist items open (list verify, canon name/cadence fix, domain auth, cadence confirm) + explicit send go | Waiting on Akash |
| Week-1 framing | Daily-slate vs talking-head conflict across older docs; unresolved | Waiting on Akash |
| Instantly handoff | 3 inboxes warming, no campaign sent; handoff trigger unconfirmed | Waiting on Akash |
| Voice-note loop | Proposed: voice note → Claude → brain tasks + builder-board card; landing surface unconfirmed | Waiting on Akash |
Collins' focus — set by Akash 2026-07-13 (supersedes all earlier week-1 missions)
- Focus 1 — Run the daily production machine: daily ad production flow + automated content production. Ship the full daily slate (newsletter, organic, AI-UGC video, carousel, tweet-style + note-style statics) every day using the templatized hooks/offer/formats/messaging, then automate the running so it needs no hands.
- Focus 2 — Automated sales: automate the sales system around the funnel (speed-to-lead, follow-up, no-show recovery, pre-call assets). Akash keeps the relationship and the close; live message tests only to the …5477 phone.
- Earlier week-1 candidates (custom demo engine proof-of-life per the 07-10 brief; funnel-path debug) move to follow-on unless Akash re-prioritizes them on the kickoff call.
Compiled 2026-07-13 from: ai-agent-run-business-vision (07-09) · autonomous-agency-os executive summary + SOW (07-11) · sp-delivery-agents blueprint (07-04) · ai-os-audit blueprint (07-12) · talent/hiring outline (06-02) · operating brief (07-10) · master SOW orchestration (07-10) · Postgres brain (handoffs, signals, tasks).