Skip to content

Operator Agent — physical-node business (experiment)

Field Value
Status Paused — no active node (2026-05-03). Vending sprint archived (restock mismatch); EV L2 POC killed (2026-05-03) — economics insufficient after stress-test (see projects/ev-charging-shoreline-poc). Thesis and long-term ladder (laundromat → RaaS → own hardware) remain intact; next node TBD at 2026-07-18 review.
Type Autonomous-income business thesis / portfolio direction
First node EV Level 2 — Shoreline POC (projects/ev-charging-shoreline-poc); smart vending deferred (weekly restock load)
Operator agent crusty (re-targeted from marketing to ops)

Thesis

Build an autonomous-income business by operating physical nodes (EV charging (L2) first → laundromat/car-wash/adjacencies — smart vending deferred — see projects/ev-charging-shoreline-poc) where location or hardware is the distribution (not X, SEO, or personal audience) and crusty runs ~80% of daily ops. Darcy is the guide (1–3 hrs/day): site selection, taste/design, QC, big decisions. Each node compounds; each new vertical reuses the agent-ops pattern built on the prior one.

Why this fits (vs. other paths considered):

  • Neutralizes the marketing weakness — placement and hardware do the distribution work; paying customers show up physically or via a network listing (e.g., Plugshare), not via engagement farming.
  • Activates the agent infrastructure proven on crusty for Aye Robot (Telegram gate + xurl + memory hardening). 2026-05-10: Crusty no longer runs band-mail / meal-plan dailies — those on grok.com. Future Operator ops would extend Crusty's stack when a node is active — no dependency on consumer-web X automation (projects/aye-robot-crusty-paused-x-automation).
  • Day job funds capex. No revenue urgency → can pick durable, slow-iteration bets.
  • Human role is 1–3 hrs/day curation, which matches Darcy's stated preference and rules out sales/services businesses.
  • The uniquely hard-to-replicate asset being built is a trained operator agent running a real P&L, not any single hardware choice — so learnings transfer across verticals.

Focus commitment (2026-04-18 → 2026-07-18, 90 days)

Revised 2026-04-18 (same day). Original plan treated Operator Agent as sole top-priority with projects/ayerobot-comic paused for 90 days. Revised after the 2026-04-17 X API pay-per-use pricing change (source) removed Aye Robot's specific transport + economic blocker.

Correct mental model (refined late-day): Aye Robot and Operator Agent are two waypoints on the same learning curve, not two competing priorities. Both are "Crusty runs a real automated workflow end-to-end with a Telegram human-gate." Aye Robot is the easier, lower-stakes proof-point — ship it first to prove the pattern works, then apply the same pattern to the higher-stakes operator ops (charger telemetry + settlement reconciliation, landlord comms, customer support — vending-era supplier PO skill parked). The architectural pattern — draft state-changing action → Telegram gate → execute → verify via public state — is identical; only the stakes and integration endpoints differ. Full framing: crusty Direction.

Active (2026-04-18 → 2026-07-18):

  • Waypoint 1 — projects/ayerobot-comic autonomous loop: first real automated workflow. Core stack unblocked (2026-04-21); morning image → Telegram on a schedule (2026-04-27). Done (2026-04-27): projects/crusty two-tier model routing proved then grok-4.20 orchestrator paused (2026-04-29) — steady state grok-4-1-fast-reasoning until Grok 4.3 (or successor) is cheap enough; xurl installed + OAuth 2.0 PKCE authed (dev app) — official steady-state X transport, replacing post_comic_test.py for native media; E2E brainstorm → Telegram approval → first live xurl post verified. Remaining: mentions listen path + ~1 week post-path reliability — projects/aye-robot-crusty-paused-x-automation, projects/ayerobot-comic#next-concrete-step--mentions-listen-path.
  • Waypoint 2 — Operator Agent sprint + node 1 (this page): advanced version of the same pattern — real P&L, real landlords/utilities/settlement pipelines, real drivers. Node 1 = EV L2 Shoreline POC (not vending). Ongoing months-long build.
  • crusty — runs both; the waypoints share the Telegram gate, session-JSONL discipline, and "re-read state before claiming success" rule.
  • Day job.

Explicitly paused / deprioritized for 90 days:

  • projects/lunarcast Crusty X marketing — deferred (auto-like/auto-follow removed from self-serve API 2026-04-17). Product is active — App Store live; next considering local transcription + NotebookLM-style querying.
  • projects/triviabalance — paused; revisit as possible senior-living adjacency if the first physical node succeeds.
  • projects/ai-model-distillation Phase 2–4 — weekend reading only; not a project line.

Kept as identity / life (not business):

Waypoint 1 — concluded (2026-05-03). Gate pattern proven end-to-end. Experiment stopped intentionally rather than grinding on mentions/growth path. See projects/ayerobot-comic for final status. Learnings (Telegram gate, xurl transport, cron reliability) transfer directly to future operator ops work.

Waypoint 2 — Node 1 killed (2026-05-03). EV L2 Shoreline POC did not survive economic stress-test. See projects/ev-charging-shoreline-poc and projects/operator-agent Node 1 section. Focus window direction open — next candidate node TBD at 2026-07-18 review.

Node 1: EV charging (Level 2) — Shoreline POC

Status (2026-05-03): killed. EV L2 was chosen as Node 1 after vending was rejected for restock labor. After a full economic stress-test, the net returns (~$50–$100/mo per charger at median utilization) don't justify the capital commitment, 80–150 hrs of Year 1 owner time, illiquidity, and execution risk vs preserving existing equity positions. Full analysis and corrected economics: projects/ev-charging-shoreline-poc. No active Node 1 — next candidate TBD at the 2026-07-18 review.

Crusty’s early ops surface shifts from SKU reorder + COGS POs toward telemetry ingest, anomaly alerts, session settlement checks (Stripe / network payouts / utility tariff drift) — still the same draft → Telegram gate → verify pattern.

Archived — smart vending / micro-market sprint (Feb–May 2026)

No longer Node 1 — kept so sprint notes and spreadsheets stay interpretable.

Original rationale: fastest hypothetical Crusty-training loop in physical ops (30–60 days to first dollar, low capex, daily SKU telemetry).

Unit economics (2026 PNW ballpark — reference only)

Lever Range
Hardware (smart vending, new, cashless + telemetry) $5–15k per machine
Hardware (micro-market kit — self-checkout + 2–3 coolers) $15–30k per site
Used/refurb traditional vending $2–5k per machine
Telemetry / cashless network $10–30/mo per machine + ~5–6% transaction fees (Nayax, Cantaloupe, 365 Retail Markets)
Gross monthly revenue (healthy location) $300–1,200 per machine; higher for micro-markets
COGS ~40–55% of revenue
Location commission 0–15% (often 0% at gyms/small offices/apartments; 10–15% at schools/higher-footfall)
Net cash flow per node (mature) $100–400/mo traditional; higher potential for micro-market
Payback 12–36 months

Candidate locations (PNW, Seattle-area)

Location type Fit notes
Independent gyms / CrossFit boxes / climbing gyms Captive, health-conscious; post-workout beverage + snack demand; often no existing machine
Small-to-mid offices (20–100 people) Recurring foot traffic; often under-served if not part of a corporate vending contract
Coworking spaces Aligned with "calm/no-BS" aesthetic; small operators flexible on terms
Apartment building lobbies (100+ units) 24/7 demand; property managers gate access
Martial arts / dance studios / music schools Low competition, high loyalty, parents waiting

Ops playbook (Crusty-run)

Daily (Crusty autonomous, Darcy reviews): - Pull sales telemetry; flag stock thresholds. - Draft supplier POs when items below reorder point. - Reconcile Nayax/Cantaloupe settlement → Stripe → ledger; flag discrepancies. - Monitor temperature alerts (refrigerated SKUs). - Triage customer support messages from QR contact; draft responses.

Weekly (Darcy-led, Crusty supports): - Review SKU velocity; Crusty suggests price / assortment adjustments. - Approve supplier POs; drive restock route (1–3 stops). - Respond to location-manager comms; Crusty drafts.

Monthly: - P&L review (Crusty generates); decide on SKU pruning, location expansion. - Negotiate / renew location contracts as needed.

2-week feasibility sprint (2026-04-18 → 2026-05-02)

2026-05 follow-on: Pivot documents vending as mis-fit for Node 1; sprint artifacts (quotes, landlord practice, fake-data ops prototype) still train general operator-agent muscle. Next concrete diligence lives on projects/ev-charging-shoreline-poc.

Week 1 — Market + unit economics

  1. Drive 3 Seattle-area sub-markets (e.g., Shoreline / Lake Forest Park, Ballard / Fremont, Bellevue / Redmond). Count and categorize existing vending + micro-market placements. Note operator (Canteen, Aramark, independent).
  2. Get real quotes: one new smart vending machine (365 Retail Markets, Avanti, Parlevel, used via UsedVending.com); telemetry provider (Nayax vs Cantaloupe); merchant processing.
  3. Price a starter micro-market setup (AWM Smart Shelf or 365 Retail Markets Pico / NanoMarket class; 2–3 coolers; self-checkout kiosk).
  4. Build a single spreadsheet modeling one specific hypothetical location (e.g., "60-person coworking in Ballard"): revenue, COGS, commission, restock labor, Crusty ops cost, net, payback. Commit to raw/ when stable.

Week 2 — Ops + decision

  1. Cold-email or walk 10 candidate locations. Goal: 2–3 real conversations about placement terms.
  2. Prototype the Crusty ops layer against fake data: nightly sales pull, stock-threshold supplier email drafts, Stripe reconciliation, customer-support template. Reuse existing crusty skills surface (Gmail, exec, memory) — no new vendor.
  3. Go/no-go decision on 2026-05-02.

Go/no-go gate

Node 1 proceeds only if all three are true:

  • ≥ 1 warm candidate location verbally agrees to host (or will within 30 days).
  • Modeled unit economics show net-positive cashflow by month 9 after capex.
  • Crusty ops prototype handles the fake-data dry run end-to-end (sales pull → reorder draft → reconciliation).

If any gate fails: historical vending gate — as of 2026-05, primary path is EV L2 (projects/ev-charging-shoreline-poc); if EV diligence fails, re-evaluate the operator thesis entirely (vending unlikely to return as Node 1 without automation breakthrough).

Long-term plan — future verticals

Node 1 is now active EV L2 (projects/ev-charging-shoreline-poc). Below = scale + later ladders once the first chargers cashflow.

Vertical 2 (month 6–12): scale EV L2 footprint

Add sites and/or ports toward the 12–16 charger / 3–4 site band documented on projects/ev-charging-shoreline-poc. Reuse the same permitting + utility + PM playbook; diversify utilization risk across property types (multifamily first, workplaces/retail as alternates).

Optional: micro-market/vending revisit only if automation (or a hire) removes the weekly restock trap — otherwise deprioritized.

Vertical 3 (month 12–18): deepen EV portfolio or bolt-on adjacency

Treat additional L2 clusters + equipment refresh here; keep numbers on the POC page fresh — indicative legacy table retained for comparisons:

Lever Range
Hardware $2–8k per L2 port (J1772/NACS)
Install (electrical + permits + concrete) $5–15k per port — highly site-dependent
Network fees (ChargePoint, EVgo, Blink, AmpUp / EV Connect for independent) $10–30/port/mo + transaction %
Pricing $0.20–0.50/kWh retail; session fees possible
Utilization year 1 5–15% typical at new sites
Net per port (mature) $50–500/mo
Payback (L2) ~2–7 years (economies at 3–4 ports/site can hit ~2y — see POC)

DCFC (fast charging): ~$50–150k hardware per port + $50–300k install; utilization 5–40%; payback 4–8 years; out of scope until portfolio stage (post node 5+). Needs $150–500k per site.

Incentives to stack (WA):

  • NEVI (federal) — corridor-eligible sites.
  • Washington Climate Commitment Act programs.
  • Seattle City Light / PSE utility rebates for commercial EV charging installs (POC stacks City Light grants aggressively — during diligence now, not “month 9–12”).
  • Meaningful install offset when eligible — modeled on projects/ev-charging-shoreline-poc.

Moat: Long commercial leases + utility interconnect + PlugShare reputation stay the barrier. Relationships with multifamily operators matter whether the first SKU was soda or electrons.

Aesthetic alignment: Strong — Musk / moonshot / clean-grid fit already surfaced in projects/lunarcast starter library tone.

Vertical 4 (month 18–30): laundromat / car wash / self-serve acquisition

Lever Range
Acquisition price (small/mid laundromat) $100–400k
Self-serve car wash single-bay $50–200k to build new; less for acquisition
Typical cap rate 15–25% (well-run boring biz)
Autonomy level Very high (no staff most days)
Marketing needed Near-zero (location-based)

Entry point: acquisition, not greenfield. The boring-business / SMB-acquisition path (Codie Sanchez-style) rewards operators who can replace 60–80% of bookkeeping, customer support, and supplier comms with agents — which is exactly what the EV (+ any future SKU) portfolio should have trained Crusty to do. The multiple on an AI-augmented boring business is the core edge.

Why not earlier: Capex jumps an order of magnitude vs a few L2 ports. Need cashflow proof and operator confidence from prior nodes.

Vertical 5 (month 24+): robotics-as-a-service (deploy someone else's hardware)

Deploy autonomous or semi-autonomous robots owned by you, billed as a recurring service:

  • Robotic lawn care (Husqvarna Automower / Worx Landroid / Mammotion / Electric Sheep partnerships) — residential subscription or commercial property contracts.
  • Robotic floor cleaning (Avidbots, SoftBank Whiz, Brain Corp-powered fleets) for small retail / gyms / offices.
  • Drone roof / solar / infrastructure inspection — FAA Part 107, insurance-paid work.
  • Autonomous mowers on commercial sports fields / campuses — multi-unit contracts.

Why deferred: Category requires fleet dispatch + customer relationships, which is closer to services-business texture than the EV / laundromat line. The operator-agent pattern helps but doesn't fully eliminate the customer-management surface. Revisit once a mature Crusty can handle complex dispatch + billing cycles.

Honest risk: This is also where most solo operators get stuck doing labor. Evaluate carefully before committing capex.

Vertical 6 (year 3+, exploratory): own hardware / robotics build

The ambition end-state: design and manufacture a narrow hardware product — not a generalist robot, but a tightly scoped physical device serving one of the verticals Crusty already operates (e.g., a better vending telemetry box, a purpose-built inspection drone, a niche EV-adjacent device, an elderly-activity device adjacent to projects/triviabalance).

Why this is year 3+, not year 1:

  • Hardware R&D cycles are 2–5 years to revenue; no realistic solo path to market without prior operator learning.
  • The best hardware ideas will come from operating the prior verticals and finding genuine gaps — not from top-down ideation.
  • Capital requirement is serious ($100k+ just for a credible prototype round); only responsible after nodes 1–4 cashflow.
  • The projects/ai-model-distillation learning track (including Boxy on-device Core ML hands-on) is pre-training for this end-state — keep it alive as a weekend/background input, not a primary line.

Criteria to greenlight a hardware build:

  1. ≥ 18 months of operator reps across ≥ 2 verticals.
  2. Identified a specific hardware gap your prior operations would pay for (dogfood-able market of 1).
  3. Cashflow from nodes 1–N covers R&D without touching day-job savings.
  4. A concrete manufacturing partner (PNW hardware ecosystem: Boeing alumni network, Glowforge-adjacent makerspaces, overseas partner).

What gets killed outright (not in long-term plan)

  • DCFC-first entry — wrong scale for solo first operator; $300k+ per site.
  • Pure services / consulting — fails the "marketing is weak" and "1–3 hrs/day" filters.
  • Dropship / digital product store — marketing-engagement-dependent, same trap that killed prior projects' engagement.
  • Crypto / web3 operator plays — orthogonal to the thesis; skip.
  • Franchise purchase (robot-service franchise, vending franchise) — fee structure eats the AI-leverage edge; prefer independent ops with Crusty as differentiation.

Crusty — re-targeted ops surface

Crusty's primary direction on this project (2026-04-18 positioning; EV Node 1 as of projects/ev-charging-shoreline-poc): operator-agent for the physical EV/L2 footprint (telemetry, settlement checks, landlord PM drafts), replacing the earlier "marketing for @ayerobotcomic + LunarCast" posture that stalled on consumer X web automation (projects/aye-robot-crusty-paused-x-automation).

Day 1 Crusty additions (sprint prototype — fake data first; EV-aligned):

  • Nightly charger telemetry pull (sessions, uptime, faults — mocked in sprint; real once EMS/network API keys exist).
  • Settlement / Stripe mismatch flags (network payout vs Stripe vs expected kWh tariffs).
  • Landlord PM template drafts (renewals, access, insurance COI reminders) — same Gmail surface as vending-era supplier PO drafts.
  • Customer / driver triage for fault codes (“stall blocked”) — inherits QR/support template discipline from vending notes.
  • Daily Telegram digest (already proven surface) summarizing revenue, offline ports, tariff anomalies.
  • (Legacy vending skill) Supplier PO + SKU reorder drafts — parked unless vending returns.

Reuses: memory hardening, Brave search for tariff/rebate research, exec/JSONL audit trail, Google Calendar for contractor visits, existing Telegram.

Does not need: unattended X posting, browser automation of consumer web UIs, xAI image generation. All the fragile pieces stay out of the critical path.

See crusty Direction section for the updated pointer.

Known risks

  • Location acquisition risk — first warm lead may fall through; have 5+ candidates in play, not 1.
  • Scope-creep risk (from the two-waypoint framing) — the late-day 2026-04-18 revision reframed Aye Robot as "the easier proof-point for the same pattern Operator Agent will use," not as a competing priority. That framing is correct only if Aye Robot stays strictly at "daily post + mentions triage." Core port landed 2026-04-21 — if future-you starts thinking "maybe I should also have Crusty reply to trending threads / quote-tweet news / run LunarCast posting off the same skill," stop — those are scope creep, not waypoint refinement. Use the guardrail in Focus commitment above.
  • Operator-as-hobbyist risk — "supplement" timeline means zero external pressure to ship. The 2-week sprint deadline exists specifically to create it. Public accountability (wiki + log) is the intended forcing function.
  • Crusty ops quality — if the agent mis-reconciles payout telemetry, drafts careless landlord clauses, or mis-handles tariff assumptions at scale, it erodes the core advantage. Every state-changing skill must re-read public state before claiming success (same rule as recurring-issues "ground truth" notes).
  • Installer / interconnect drag — L2 still faces months-class permitting + utility sequencing even if faster than DCFC.
  • (Legacy) PNW vending market saturation — only relevant if vending ever re-enters; incumbents own corporate; indie wedge was gyms / apt / coworking.

Evaluation checkpoints

Date Gate
2026-05-02 Vending-era sprint deadline (historical). EV L2 POC inherits next concrete gates — keep 2026-07-18 review meaningful.
2026-07-18 90-day focus-commitment review. Is node 1 live? Cashflow trajectory? Continue / pause / kill.
2026-10-18 6-month review. Node 1 cashflowing? Ready for node 2 site hunt? Crusty ops hardened?
2027-04-18 12-month review. Portfolio scale (3 sites / 12+ ports directional goal)? Crusty EV ops hardened?

Raw sources

  • (None yet.) Drop sprint-output docs (quotes, location model, cold-email tracker) under raw/operator-agent/ as they're produced. EV POC spreadsheets / bids → annotate “ev-charging-shoreline” when saving.