NowMart
Engineering progress · prepared for Michael
One platform, ~11 weeks in.
A multi-surface platform, built and staging-tested by a virtual AI engineering team, and still in active development. Here is what already works, what a traditional or AI-assisted team would have at the same point, and what the work means for the company's value.
~78 days
Of active development, and counting
9 May – 26 Jul 2026 · 2,369 commits
~$2–3M
What the same build costs to outsource
Est. range $1.5M (offshore) to $4.5M (onshore)
~10–12
People a traditional team would need
Over an estimated 12–18 months
The output so far
What's built so far
A snapshot of an active build, measured from the real, git-tracked source. Hand-written lines only: generated code, built app bundles, vendored libraries and lock files are excluded, so nothing here is inflated. These counts grow every week.
323,000
Lines of application code
Backend, web, mobile, workers, shared packages
12,149
Automated tests
201,000 lines across 1,320 test files
165,000
Lines of documentation
393 docs: 51 ADRs, 45 runbooks, specs, plans
148
Database migrations
Versioned, reviewable schema history
~734,000
Total lines, hand-written
Across 3,970 source files
2,369
Commits in ~11 weeks
Small, reviewed, atomic changes
Multi-tenant backend
3 native Android apps
4 web applications
Admin control tower (11 areas)
Card payments (Stripe sandbox)
3-way reconciliation
Odoo ERP integration
Fleet device provisioning
Age & biometric verification
PDPL compliance
Offline-first sync
11 languages
Where it stands
Built and staging-tested, not yet live to customers
An honest snapshot at the eleven-week mark. A lot is built and working in a full staging environment. What remains before real customers is gated on external access, an acquirer agreement and legal sign-off, not on remaining engineering.
Built & working
- Kiosk self-service POS on the Elo tablet: 226 products, barcode scan, and an age-gate that records PDPL consent before the camera opens
- Native driver & merchandiser app: routes, PO receiving, warehouse pick and settlement count-back, offline outbox with a Sync Tray
- Admin Control Tower: an internal operations console for logistics, fleet, roster, reconciliation and access
- Backend on Cloudflare Workers behind every surface: multi-role RBAC, an Odoo write path, background workers
- Money-confirm core: one idempotent, maker-checker confirm contract, integration-tested against real Postgres
- Kiosk card payment via Stripe Terminal (sandbox, simulated reader), verified across every result code
- Isolated 24-worker staging estate on its own database, plus crash reporting and fleet heartbeat, PDPL-scrubbed
In flight now
- Odoo reconciliation feeding the money tabs: the core landed, the remaining slices are being wired
- Tenant brand-logo storage (R2) threaded through the repository and service layers
- Driver APK carrying the warehouse loop onto the physical device
- Real Elo kiosk into the fleet: heartbeat, analytics and crash reporting built, device claim pending
Gated on access, NDA or legal
- Live money posting to Odoo: needs a split poster credential, production Odoo access and queue setup
- Real guest card payments: blocked behind the UAE acquirer NDA; only the Stripe sandbox works today
- Age & face verification: engineering-done, gated on PDPL legal sign-off, a DPIA and 9-locale translation
- A closed Odoo session: per-property P&L and POS stock decrement stay unproven until one runs
- Field media upload: the R2 photo and logo buckets are not yet rebuilt in the new account
- Driver app to a real field user: roles, release keystore, error reporting and a DB migration still to harden
Production status: nothing is live to real paying customers or hotel guests yet. The kiosk shell, the catalog and some internal plumbing run on the production host, but every transaction to date has been Clint and Michael testing. The money path (Odoo posting, VAT, per-property revenue, reconciliation) is staging-only, and the changelogs are still marked Unreleased.
The team
Clint as CTO, directing an eleven-strong AI team
The one human in the loop runs a full AI engineering organisation: an orchestrator and a chronicler who coordinate the work, and nine specialists who build, test, ship and document it. Everything above was produced by this structure.
Human · CTO & Founder
Clint Browne
Chief Technology Officer & Founder
Sets product direction and architecture, makes the calls, and directs the AI team. The one human in the loop.
Reports to Clint · directs the team
Bob
Orchestrator / Engineering Lead
Turns Clint's direction into work, delegates to the specialists, and keeps nothing falling through the cracks.
Coordination
Margo
Workstream Chronicler
Maintains durable, human-readable status for every in-flight workstream.
The nine specialists · reporting up through Bob
Engineering
Erik
Backend Engineer
Cloudflare Workers, Postgres, tenancy/RLS, queues, Odoo integration.
Jade
Frontend Engineer
React web applications, including the admin control tower.
Kayla
Android Engineer
Kotlin/Compose kiosk, front-desk and driver apps, offline-first, camera.
Quality & Architecture
Karen
Code Reviewer & Architect
Security review, architecture decisions, the quality gate.
Nadine
QA & Testing Engineer
Cross-surface test strategy, end-to-end and regression testing.
Sophie
Evidence Specialist
Proves a fix is real with a falsifying test before anything ships.
Platform & Operations
Tanya
DevOps & Infrastructure
Cloudflare, CI/CD, release engineering, Android build & deploy.
Devon
SRE, Observability & Data
Metrics, alerting, probes, reconciliation audits, incident response.
Documentation
Lauren
Technical Writer & Spec Author
Architecture records, specs, runbooks, changelogs, release notes.
The comparison
Three ways to build the same thing
The fair modern benchmark is not just a traditional team, but an AI-assisted human team, which is the realistic alternative today. Even with full AI tooling, a human team still carries the coordination, review, onboarding and project-management overhead that AI does not remove. The first row marks where each team would realistically be at NowMart's current eleven-week point; the rows below are the full build. Cost and time are estimates with stated assumptions; the code, test and documentation counts are exact.
Traditional team · no AI
By week 11hiring & discovery
Calendar time12–18 months
Team size~10–12 people
Effort140–190 p-mo
Cost to build~$2–3M
Tests & docsusually far fewer
AI-assisted human team
By week 111–2 surfaces
Calendar time~7–12 months
Team size~6–9 people
Effort85–130 p-mo
Cost to build~$1.2–2.2M
Tests & docsless, if disciplined
NowMart · AI team
By week 11several, in staging
Calendar time~11 weeks in
Team size1 founder + 11 AI
Effortn/a
Cost to builda fraction
Tests & docs12,149 & 393
Even against the realistic modern alternative, a top agency given the same brief and full AI tooling would still need most of a year and well over a million dollars to reach where NowMart already is. The reason is that the compression here came from removing the team-coordination overhead entirely, one founder with complete context directing specialist agents, not from AI alone, which a multi-person team cannot escape. AI speeds up writing code; it does not remove the meetings, handoffs, onboarding and review cycles that dominate a human team's time and cost.
Why it matters at exit
From a services business to an IP business
The real prize is not the cost saved. It is that owning the platform changes which multiple the company is valued on. A services business is priced on people and projects; a product business is priced on recurring revenue and defensible IP.
| Business type | How it is valued | Typical multiple |
|---|
| Services / agency | Multiple of revenue or profit; people-based, project revenue | ~0.5–1.5× revenue |
| Product / SaaS / IP | Multiple of recurring revenue; scalable, high margin, defensible | ~4–10× ARR |
Owning this IP is what values NowMart as a product company, not a services one. A services business is priced near 1× revenue; a product business at 4–10× ARR. On the same revenue that is a 3–5× swing in enterprise value, and at a three-year exit it is the difference between a modest number and $$$.
Every dollar of recurring revenue is worth several times more inside a product company than a services one. Get the product story right and the exit is valued on the product line, not the services line, and the unusual depth of testing and documentation is exactly what defends that multiple when a buyer's technical team looks under the hood. The platform is also a defensible, rebuildable asset the day the conversation starts. That is where the $$$ is.