FarmAgent + FieldAgent · Product Review
v3 · post-session · 2026-06-28
Re-evaluation against the Jun-27 baseline

Two product families, not two apps. The Labor MVP is closer than it looked — because scouting isn't in it.

FruitScout is two independent product families — Field and Labor — each climbing the same ladder (instrumentation → manual management → agentic ops). The apps are delivery surfaces, not products. This session hardened the Labor family's instrumentation + wedge (the 4W's and the onboarding agent) against messy real data. The key correction to the earlier read: scouting is a Field-family product bundled into the FieldAgent app — it is not in the 4W's (Labor) MVP, so its unbuilt screens don't gate the labor product at all.

The product structure — corrected

Two families, one ladder each — apps cut across as delivery surfaces

FruitScout is two independent product families — Field and Labor — each climbing the same three rungs. FieldAgent (mobile) delivers the instrumentation capture for both families; FarmAgent (web) delivers the management and agentic rungs. Status pills = commercial readiness, not code quality.

FamilyInstrumentation (capture)Manual schedulingAgentic scheduling
Labor The 4W's — Who / What / When / Where, captured as executed (clock-in/out, tasks, roster).
MVP · capture path built
Build the schedule by hand — the console's schedule grid, crews, LFT-conflict checks.
Coherent · largely built
The agent builds & optimizes the schedule.
Onboarding wedge built · agent-scheduling nascent
Field Drone + phone data pipelines (CV inventory, piña sizing → Digital Twin); scouting (phone capture).
Drone mature · phone scouting unbuilt
Plan field ops by hand — flights, harvest windows, Register Fields, DT viewer.
Built
Agentic field ops — GIS co-pilot.
Real tech · early

Delivery vs product: FieldAgent (mobile) bundles Labor's 4W's capture and Field's scouting — but they remain separate products on separate roadmaps. The 4W's MVP does not include scouting, so the unbuilt scouting screens are not a labor-product blocker — they're Field-family instrumentation on its own timeline.

The strategy — what the ladder is actually for

Scheduling is a trojan horse. The product isn't a scheduler — it's the Chief of Staff that ends up running the entire business.

The schedule is the one object that holds the whole operation — the who/what/when/where that ties labor and field together. So the agent that earns its way in by building the schedule is positioned to expand into running the business: compliance, payroll, procurement, field ops. Agentic scheduling is the beachhead, not the ceiling. Which reframes the original review's "the persona is thin, it goes quiet after onboarding" note from a minor gap into the gap — the distance between a templated onboarding bot and a persistent operating agent is the distance between the wedge and the actual product.

What changed this session

The product delta

Concentrated on the wedge — the review's #1 leverage point

  • The wedge now eats mess. Feature 038 reconciles a roster CSV + a whiteboard photo + sticky notes into one schedule instead of double-counting — confidence-banded, with a side-by-side review queue and a batched "ask for the gaps."
  • A measured result. The session eval moved naive multi-artifact accuracy 0.915 → 1.000 — the first quantified evidence the "magic trick" survives multi-source input.
  • First real marketing artifact for the wedge — a hosted, on-brand landing page (cos-onboarding.farmagent.fruitscout.ai) built from the discovery research. The whiteboard demo finally has customer-facing collateral.
  • Upsell clarified. The crop-intelligence models (drone YOLO26-OBB detector, phone segmentron, size→weight curve) are now documented — so the upsell tier can be scoped, not hand-waved.
The three commercialization gaps — progress

Where the business is still held back

Gap 2 · Proven on mocks, not on mess

Real progress
Then: the wedge was unvalidated end-to-end — a real Sauza file extracted zero workers with no graceful recovery; "all tests pass" was green-on-mocks.
Moved: 038 hardens exactly this — multi-artifact reconciliation, confidence bands, and a gap-ask that recovers instead of stranding the user; eval 0.915→1.000. Still pending: a real-customer POC against a live, messy file (the eval is synthetic/curated, not a buyer's actual mess).

Gap 1 · The pitch sells the wrong company

Small dent
Then: the live 21-slide deck pitches the deprecated "Farm OS"; labor and the onboarding wedge are absent; pricing is per-hectare, not per-worker.
Moved: the wedge now has a hosted marketing page — the "strongest asset not in the pitch" got a surface. Still pending: the core deck rebuild (lead with the whiteboard demo), the per-worker re-pricing, and retiring the Fleet-Command slides — untouched.

Gap 3 · Pre-launch gates & server-side enforcement

Mixed
Then: billing + FeatureGate off by design; the item that "shouldn't wait" was server-side LFT compliance enforcement (a liability for a compliance product).
Moved: billing is now genuinely Stripe-integrated (real payments, OXXO/SPEI, webhooks). Still pending: entitlement is enforced server-side only in Labor — everything else behind FeatureGate is UI-only; and server-side compliance enforcement (validate still a stub) is not done.
Distance to commercial release — by product family

Labor (the 4W's MVP) vs Field

Readiness is per family, not per app — FieldAgent (mobile) and FarmAgent (web) each deliver pieces of both. The Labor 4W's is the near-term commercial product; Field is a separate, earlier-stage product on its own timeline.

Labor family — the 4W's

Capture → manual scheduling → (agentic). The near-term commercial product. Delivered by mobile capture + web management.
  • Capture path built — the mobile 4W's (clock-in/out, tasks, roster) is the MVP instrumentation, and it exists.
  • Manual scheduling largely built — web schedule grid, crews, LFT-conflict checks, STPS export (030).
  • Real compliance depth + a dated trigger — dynamic LFT limits; the Jan-2027 Art. 132 mandate is a forced-buying event. Billing is real (Stripe); security floor rose this session.
  • Entitlement enforcement server-side only in Labor; compliance validate is a stub (client-side only) — the real-now pre-launch items.
  • Activation bridge missing — no in-app QR/crew enrollment to consume the onboarding agent's codes; no real-customer POC yet; deck/pricing still wrong.
  • Agentic scheduling is nascent — the wedge is built, but the persistent operating agent (the actual thesis) is not.
Path: a focused pre-launch punch-list — enforce entitlements + compliance server-side, build the QR activation bridge, one real POC, fix the deck. Scouting is not on this list.

Field family — pipelines + scouting

Capture → manual scheduling → (agentic). Separate product, earlier stage. Delivered by drone pipeline + the FieldAgent app.
  • Drone instrumentation is the mature half — the CV inventory / piña-sizing pipeline (Agave Inventory) is real, productionized tech; models now documented.
  • Manual layer built — Register Fields + the Digital-Twin viewer.
  • Phone scouting unbuilt — placeholder screens; backend scouting model not even registered. This is the Field-instrumentation gap (and it's bundled into the FieldAgent app, not the labor MVP).
  • The next-gen mobile app is pre-1.0 — strong offline-first bones (Flutter + FastAPI/PostGIS, outbox sync), but blank prod API URL, no CI/CD, manual store uploads. Production still rides the legacy app.
  • Agentic (GIS co-pilot) is early; intelligence tier/pricing unclear.
Path: a real 1.0 effort — ship phone scouting, fix the mobile prod config + CI, mature the intelligence tier. Its own roadmap, decoupled from the labor MVP.
The recommended moves — done vs remaining

Scorecard against the original five moves

Move (from v2)StatusDetail
3 · Harden the wedge vs real dataSubstantial038 multi-artifact reconcile + gap-ask + banded confirm shipped; eval 0.915→1.000. Needs one live-customer POC to fully close.
1 · Make the pitch match the productStartedWedge marketing page shipped. Core deck rebuild, per-worker pricing, Spanish finish — not done.
2 · Pre-launch gates + server-side compliancePartialBilling now real (Stripe). Entitlement breadth + server-side LFT enforcement still open.
4 · Make the mobile hand-off realNot startedThe onboarding→crew QR activation bridge gates the 4W's MVP (hand-off to crews) — untouched. The mobile prod-config/CI + phone scouting are the Field family's 1.0, a separate timeline.
5 · Crop-config layer for "omni-crop"Not startedStill agave-coupled in the visible product; models documented but no per-customer config layer.
★ The thesis — persistent COS agentBuild-aheadAgentic scheduling → the Chief of Staff running the business. Today's COS is a templated onboarding bot; the persistent operating agent is unbuilt — and it's the actual product.
Bottom line

Where the product sits on the path to commercial release

The session moved the highest-leverage thing the first review named — hardening the Labor instrumentation + onboarding wedge against real, messy, multi-source data — and it worked (eval 0.915→1.000). Reframed correctly: the Labor 4W's is the near-term commercial product, and it's close — capture built, manual scheduling built, real billing and compliance depth — held back by a finishable pre-launch list (server-side entitlement + compliance, the QR activation bridge, one real POC, a corrected deck). Scouting and the next-gen mobile 1.0 are the Field family's separate, earlier-stage roadmap — not labor blockers. And the real prize sits one rung up: agentic scheduling is the beachhead for the Chief of Staff that runs the entire business — building that persistent operating agent is what turns a good scheduler into the category-defining product.