Eight pilots in development.
Pilot blueprints — scoped, not yet deployed. Each one maps to a real operator problem we’ve interviewed for and would build with a prospective client. If your operation matches one of these — that’s the conversation we’d like to have.
Pilot · Voz AI · international SMB
Voz AI · free 3-month Action Box pilots, internationally
Voz AI · free 3-month Action Box pilots, internationally
Entity · 501(c)(3) in formation · Free pilot → Action Cell rent or purchase → refurbish if declined
What we’re building
Voz AI is broadening its 501(c)(3) mission before any accessibility-specific app build. The first deployment shape: free 3-month Action Box pilots for SMBs in international markets where governed-AI hardware otherwise wouldn’t reach — starting with Brazil, Lusophone Africa, and Latin America (using the founder’s existing EN/PT bilingual reach). Each pilot ships pre-Calibrated, with the full Training, Auditing, and Payment harnesses live from day one. The pilot is the proof. After 90 days, three clean exits.
- Free 3-month pilot — Voz AI ships a pre-Calibrated Action Box (Mac Mini M4/M5 + the four-agent stack + the three harnesses) to the SMB at Voz AI’s cost. Same 75% reliability floor we hold for paid clients.
- Check-ins at day 30 / 60 / 90 — structured review of what the agents have done, what they haven’t, and what the harness logs show. Same review cadence as paid Founding Cohort deployments.
- Three exits at day 90: (1) convert to Action Cell rental at standard Maintenance pricing, (2) buy the Action Cell outright at Founding Cohort rates, or (3) decline.
- Decline path: Voz AI ships a return label, the SMB packs the unit, we receive it, we wipe the Mac Mini to factory state, and it goes into the next pilot. No salvage write-off, no awkward “keep using this” conversations, no orphan units in the field.
- Clean entity handoff on conversion — Voz AI (501(c)(3)) ran the pilot; Action AI, LLC owns the paid relationship from day 91. Cross-entity benefit documented per the OA Article XX separation rules.
Projected impact
- Governed-AI hardware reaches SMBs that can’t carry the upfront $2,000 Calibration — expanding the addressable market without dropping price floors for paying clients
- Hardware that’s never written off — pilot churn becomes inventory recycling, not capital loss
- Direct, clean funnel from Voz AI pilots into Action AI Maintenance revenue once a unit converts
- Voz AI mission stays broader than accessibility — economic access for underserved SMBs first, accessibility-specific apps (see Phase 2 below) follow once the pilot model is proven
- Demonstrates the Action Box deployment loop in markets the LLC can’t independently sell into yet
Pilot · Voz AI · accessibility (Phase 2)
Accessibility tools, governed by AI.
Accessibility tools, governed by AI.
Entity · 501(c)(3) in formation · Phase 2 of Voz AI — follows the international SMB pilot model proving out (see card above)
What we’re building
A 501(c)(3) currently in formation, focused on AI as the connective layer between accessibility tools that already exist but don’t talk to each other. The mission isn’t a new screen reader or a new switch — it’s a governed agent layer that orchestrates the tools a disabled user already pays for, on hardware they already own. Built on the same four-layer Ethical Harness we ship to operators on the Action Cell, retuned for individuals.
- Track 1 · Sight-impaired users — real-time scene description, document narration, navigation cues, OCR of physical environments, intelligent routing between existing screen readers and AI tools. In conversation with one of the country’s top institutes for the blind on a research/co-design partnership.
- Track 2 · Cognitive differences — plain-language summaries, step-by-step task scaffolding, gentle reminders, sensory-load adjustments. For users with cognitive or learning differences and their care networks.
- Track 3 · Physical / mobility differences — voice-first agent control, predictive task chaining, adaptive input parsing, hands-free workflow execution across the user’s tools.
- Ethical Harness applied throughout — every action is auditable, every binding decision routes to the user (or their authorized human in the loop). No agent autonomously commits the user to anything.
- Built on the same four-agent stack that configures for operator clients — the harness is the product, regardless of who’s in the chair.
Projected impact
- Independence in daily tasks — without surrendering decision authority to an opaque system
- Audit trail for every adaptive action, so users (or caregivers) can verify what the agent did and why
- Sequenced rollout means each track is configured only after the prior one’s governance posture proves out in real use
- Demonstrates that the same governance infrastructure that protects an operator’s back office can protect a person’s autonomy
- 501(c)(3) structure unlocks disability-specific grant funding that for-profit accessibility tooling can’t access
Pilot · Tax prep ops
IRS Notice Prep & Records-Assembly Tool.
IRS Notice Prep & Records-Assembly Tool.
Service · Pre-meeting records and citation prep for IRS notices · Licensed CPA / EA / attorney delivery
What we’re building
An Action Cell workbench that handles the preparation work a tax paraprofessional would normally do for an IRS notice or Information Document Request: classify the notice, pull the relevant tax-year records, organize and index the supporting documents, and build the citation list. The licensed CPA, EA, or attorney walks into the response with prep already done — lower client cost, faster turnaround, more time spent on judgment instead of records-chasing. The agent does not draft the response itself, does not file anything, and does not represent the taxpayer. The licensed practitioner authors and signs every response per Circular 230.
- Notice classification — agent reads the inbound notice (CP2000, LT11, IDR, etc.), identifies which IRM section governs the response, and surfaces the applicable response window.
- Records assembly — pulls relevant returns, schedules, and supporting docs from the practice’s document store; produces a chronological index ready for practitioner review; flags missing items.
- Citation list — every authority pinned to a verified tax-source database (IRC, regs, rev rul, case law). No fabricated citations, ever. The practitioner decides what to use.
- Substantiation checklist — surfaces what the IRM specifies the response must include for this notice type; agent does not write the response.
- Intake brief — structured one-page summary for the supervising practitioner’s pre-response review.
- Authorship boundary — the agent prepares; the practitioner authors. Responses, IDR replies, and any document filed with the IRS are drafted, edited, and signed by the licensed CPA, EA, or attorney. Circular 230 boundary is structurally enforced.
- Calendar discipline — auto-tracks every active notice, surfaces deadlines 14, 7, and 2 days out so nothing slips.
- Conflict + privilege scaffolding — built on the same Ethical Harness used for all Action AI deployments; client-confidential data never leaves the practice’s tenancy.
Projected impact
- Lower client cost. The records-assembly and citation prep is done before the practitioner’s billable hours start. Practitioner time is spent on substantive judgment, not on paper-chasing.
- Faster response prep. Targeting per-notice prep cut from 4–8 paraprofessional hours to under 60 minutes; estimated per-matter savings of $400–$1,200 in prep cost. Validated per practice during pilot.
- Citation-anchored materials reduce the rework loop with the licensed reviewer
- Calendar automation eliminates missed-deadline penalty risk
- Audit trail per case — every doc the agent indexed, every cite it surfaced — defensible to client, IRS, and (if needed) the practice’s malpractice carrier
- Practice is positioned as AI-augmented before competitors close the gap
Pilot · Legal prep ops
93A Evidence & Exhibit Prep Tool.
93A Evidence & Exhibit Prep Tool.
Service · Pre-meeting evidence organization for Mass. consumer-protection matters · Licensed-attorney delivery
What we’re building
An Action Cell workbench that handles the preparation work a paralegal would normally do for a Massachusetts Chapter 93A consumer-protection matter: organize the consumer’s evidence, index exhibits chronologically, identify the legal theory in play, surface the controlling authorities, and produce a structured intake brief. The supervising attorney walks into the meeting with prep already done — lower client cost, faster scoping, more time spent on judgment instead of paper-chasing. The agent does not draft demand letters, does not represent parties, and does not produce legal-operative documents. The licensed attorney authors and signs every legal output.
- Evidence intake & organization — agent ingests the consumer’s evidence packet (emails, receipts, photos, communications), de-duplicates, timestamps, and produces a chronological exhibit index ready for attorney review.
- Exhibit prep — numbered exhibits, suggested labels, gaps flagged so the attorney knows what to ask the client for at intake.
- Theory identification — surfaces which 93A theory the facts support (§2(a) unfair, §2(a) deceptive, or specific consumer statute), with the elements the attorney needs to argue.
- Citation list — controlling authorities (MA G.L. c.93A, 940 CMR, case law) anchored to verified sources. No fabricated citations, ever. The attorney decides what to use.
- Damages-exposure framing — surfaces the math on actual damages, §9(3) multiplier exposure, §9(4) attorney’s-fees claim. Reference only; the attorney decides what to demand.
- Intake brief — structured one-page summary for the supervising attorney’s pre-meeting review.
- 30-day clock awareness — calendars the §9(3) response window so the attorney’s practice management tool stays in sync.
- Authorship boundary — the agent prepares; the attorney authors. Demand letters, settlement communications, and any document served on the recipient are drafted, edited, and signed by the licensed attorney. UPL boundary is structurally enforced.
- Conflict + privilege scaffolding — built on the same Ethical Harness used for all Action AI deployments; client-confidential data never leaves the firm’s tenancy.
Projected impact
- Lower client cost. The paralegal-prep work is done before the attorney’s billable hours start. Attorney time is spent on judgment, strategy, and authorship — not on paper-chasing.
- Faster scoping. Targeting first-meeting prep cut from 2–4 paralegal hours to under 30 minutes; estimated per-matter savings of $300–$800 in prep cost. Validated per practice during pilot.
- Citation-anchored materials reduce associate-time rework loops
- Damages-exposure surfaced at intake, not at trial
- Calendar awareness keeps the §9(3) 30-day window from being missed
- Audit trail per matter — every doc the agent indexed, every cite it surfaced — defensible to client, opposing counsel, and (if needed) the firm’s malpractice carrier
Pilot · Household & SMB ops
Subscription Manager.
Subscription Manager.
Service · Subscription discovery, centralized renewal tracking, governed cancellation
What we’re building
A subscription registry that monitors every recurring charge across the operator’s connected accounts (email receipts, bank/card statements, app-store records). Each subscription gets a row with the next renewal date, the monthly/annual cost, the direct cancellation link, and the trial-end date if applicable. The agent surfaces patterns — trials about to convert, duplicate services, dormant subscriptions — and notifies the human before the charge lands. The agent never cancels autonomously. Every cancellation is human-ratified; the agent prepares the click-path and the human presses the button.
- Auto-discovery — ingests email receipts, card-statement line items, and connected-app metadata to build the registry without manual entry.
- Renewal calendar — one view of every renewal date, sorted by next-charge proximity, with cost and category.
- Cancellation deep-links — one click takes the human to the actual cancel page for that vendor; no hunting through nested account settings.
- Trial-end alerts — 72-hour, 24-hour, and same-day notifications before any trial converts to paid.
- Duplicate & dormant detection — flags overlapping services (two cloud storage subs, two music subs) and charges on services not opened in 60+ days.
- Human-ratified cancellation — agent prepares, human approves. Nothing recurring is canceled without a human in the loop. APEX™ harness applied to any auto-renewal action.
- Audit log — every renewal flagged, every cancellation prepped, every human approval, timestamped and retained.
Projected impact
- Spend recovery. Operators routinely lose 8–25% of recurring-spend to forgotten subscriptions and unwanted trial conversions. The registry surfaces those before the next charge.
- No more credit-card-statement archaeology to find unknown charges
- Trial conversions become a deliberate decision, not a calendar oversight
- Duplicate-service consolidation cuts the bill without losing functionality
- Audit trail makes finance review & budget defense trivial
Pilot · Restaurant ops
Third-party delivery reconciliation.
Third-party delivery reconciliation.
Service · Reconcile what the platform charged the customer vs. what the restaurant actually received
What we’re building
A reconciliation workbench that ingests order-level data from each delivery platform’s merchant portal alongside the restaurant’s bank deposits. For every order, the agent computes: amount the customer paid, amount the restaurant received, the platform’s fee/commission breakdown, and the delta. Daily, weekly, and per-platform reports surface what the operator is actually netting. The agent flags exceptions — missing payouts, refunds the restaurant didn’t initiate, fee changes that weren’t announced — and notifies the operator so disputes can be filed before the platform’s deadline window closes.
- Per-order P&L — every order shows customer-paid, restaurant-received, and the platform’s cut. No more black-box settlement reports.
- Multi-platform view — DoorDash, Uber Eats, Grubhub, and any others on one dashboard. Compare effective take rates across platforms.
- Payout reconciliation — matches each platform’s payout to the restaurant’s bank deposit; flags missing deposits within 24 hours.
- Refund anomaly detection — surfaces refunds the restaurant didn’t initiate (customer-side disputes, platform-side adjustments) for review.
- Fee-change alerts — notices when a platform’s effective commission rate shifts versus the previous month, with the dollar impact.
- Dispute window tracking — each platform has a finite window to dispute a charge; the agent calendars deadlines so disputes don’t lapse.
- Tax-ready export — clean monthly P&L by platform for the restaurant’s accountant; no more spreadsheet acrobatics at year-end.
Projected impact
- Recover lost revenue. Industry data suggests 1–3% of third-party platform orders have a reconciliation discrepancy. On $40K/mo of platform sales that’s $400–$1,200/mo of recoverable revenue.
- True per-platform economics — surface which platform is actually profitable vs. which is a loss leader
- Dispute filings happen within the platform’s window instead of slipping past
- Refund anomalies caught in days, not at month-end statement review
- Bookkeeper & accountant time cut substantially — clean data instead of PDF settlements
Pilot · Household ops
Family chore planner.
Family chore planner.
Service · Chore assignment, rotation, and completion tracking across household members
What we’re building
A household planning tool that maintains a registry of recurring chores (daily, weekly, monthly, seasonal), rotates them fairly across family members based on age-appropriate scope and availability, and tracks completion. The agent surfaces what’s overdue, what’s on deck for the day, and adapts assignments when schedules shift (a kid is sick, a parent is traveling). Notifications go to the specific person responsible, not a group chat — reducing the cognitive load on the household coordinator (typically one parent). The household coordinator stays in control: the agent proposes the rotation, the human ratifies and adjusts.
- Chore registry — one source of truth for every recurring household task, with frequency, age-appropriateness, and estimated time.
- Fair rotation — agent proposes assignments balancing workload over time; the household coordinator ratifies before each cycle.
- Per-person notifications — each family member gets their own task list; no group-blast WhatsApp threads.
- Schedule-aware adjustment — when a member is sick, traveling, or has a one-off conflict, the agent proposes a temporary reassignment for human approval.
- Completion tracking — status board shows what’s done, what’s overdue, what’s scheduled, with optional kid-friendly streak/reward visualization.
- Weekly review brief — one-page summary for the household coordinator: completion rate by member, recurring bottlenecks, suggestions for the next cycle.
- Privacy & data boundary — household data stays in the household’s tenancy; nothing leaves; no behavioral profiling.
Projected impact
- Reduce coordinator cognitive load. Typically one parent carries the mental model of who’s doing what; the registry externalizes it.
- Fewer arguments about whose turn it is — the rotation is recorded, not remembered
- Age-appropriate scoping makes it usable from kids through teens through adults
- Completion tracking creates a paper trail (useful for allowances, accountability, or just visibility)
- Schedule-aware reassignment handles real life without breaking the system
Pilot · Salon & stylist ops
Independent stylist chair pipeline.
Independent stylist chair pipeline.
Service · Automated content creation pipeline + booking acceleration for independent hair stylists
What we’re building
An end-to-end content + booking workbench scoped with the independent hair stylists working out of one Newbury Street salon (name redacted at the operator’s request). The stylist is effectively running a one-person business inside someone else’s space — their growth lever is filling the chair, and that runs almost entirely on social-media presence and rebooking discipline. The agent handles the capture-to-publish-to-follow-up loop: before/after photos in, branded posts out at the times the stylist’s audience is actually online, post-appointment messages with one-click rebooking links. Every client-facing message routes to the stylist for human ratification — no autonomous DMs, no autonomous booking changes. The agent prepares; the stylist approves.
- Capture — before/after photo intake from phone, auto-cropped, lightly retouched (no AI distortion), tagged with the service performed and timestamp.
- Post production — branded overlay (handle, colors), short-form video assembly from photos + service notes, ready-for-review queue.
- Publish window optimization — posts to Instagram and TikTok at times the stylist’s specific audience is online (not a generic algorithm guess), with caption + hashtag scaffolds the stylist edits and approves.
- Rebooking follow-up — post-appointment message with a one-click rebooking link calibrated to the service’s natural cadence (4 weeks for cut, 6 for highlights). Stylist edits the message before send.
- Booking-source tracking — every booking tagged with where it came from (Instagram, TikTok, referral, walk-up) so the stylist sees which channels actually fill the chair.
- Calendar sync — integrates with whatever scheduling tool the stylist already uses (Square, Vagaro, GlossGenius); no platform lock-in.
- Ethical Harness — no autonomous client outreach, no autonomous booking confirmations, no AI-generated faces or distorted before/after results. Every message is human-ratified before send.
Projected impact
- Fill the chair without filling the evening. Targeting 4–6 hours/week of social-media admin time recovered — the stylist reviews and approves instead of producing from scratch.
- Consistent posting cadence even on heavy-booking weeks — the content pipeline doesn’t pause when the chair is full
- Rebooking rate visibility per service type — surfaces which services actually convert to repeat bookings and which don’t
- Booking-source attribution — stylist learns which platform actually drives chair-fills, not which one feels busiest
- Tool-agnostic — the harness wraps the stylist’s existing scheduling and social tools rather than asking them to switch
The operators we’d build these pilots with.
The pilots above are blueprints, not active deployments. Each was scoped after talking with prospective operators — trades shops, service businesses, single-location storefronts, and small clinical practices — whose coordination problem maps to the build.
If your operation looks like one of these and your coordination layer is WhatsApp threads, spreadsheets, and the founder’s memory — this page is the start of the conversation, not the end of one.
Want to be part of a pilot?
We open cohort conversations one operator at a time. Every partnership starts with a free Calibration call — no commitment, no pressure, just an honest conversation about whether your problem maps to one of the pilots above.
Book a fit call →