VZU Projects
BookFlow. The booking platform that doesn't break.
A 14-week build. 500+ active businesses. 99.9% uptime. 80% of support handled by an AI Ops agent. This is the inside story of how the VZU Atlas agent shipped BookFlow.
A case study · Issue 01 · Vancouver, CA
Chapter 01 · The brief
A salon chain, losing staff to bad software.
BookFlow started with a 14-location salon chain in Vancouver. Their existing booking software was 11 years old. Their stylists were spending 40 minutes a day fighting it. Staff turnover was 60% per year — and the number one reason for leaving, in exit interviews, was 'the software made me hate my job.'
The brief from the founder was 47 pages long. We boiled it down to one sentence: a booking platform that doesn't make the operator feel stupid. The constraints were tight. 4 months to ship. $250k budget. 11 locations needed to migrate. Zero downtime tolerance — the existing system handled 800 bookings a day.
We didn't propose a custom build. We proposed a fixed-fee, fixed-scope brief against the NetWit Agentic OS runtime. The runtime gives us the orchestrator, the audit trail, the MCP integration, the RBAC. We just had to build the surface — the actual booking experience that 800 stylists would touch every day.
Chapter 02 · The architecture
Postgres at the core, three agents on the surface.
We picked Postgres for the data model. Not because it's fashionable. Because every operator in the brief already knew what Postgres looks like. The migration team could read the schema. The accountants could query the booking ledger. The auditors could verify the audit trail.
Above Postgres, the Atlas agent runs the orchestrator. The Atlas agent is a single-purpose build agent — it doesn't do everything, it does one thing really well. It takes a brief (in this case, a 'salon booking system' brief) and ships the production web app. 12,400 lines of code. 47 integration tests. 8 deploys. 14 days.
Above Atlas, three role-specific agents: the Customer Agent (handles customer-side flows), the Operator Agent (handles the salon-side flows), and the AI Ops Agent (handles support). The AI Ops Agent was the surprise — it ended up handling 80% of support tickets in the first month, freeing the team to focus on the hard stuff.
“Postgres at the core, three agents on the surface. The runtime gives us the orchestrator. We just had to build the part the operator touches.”
— The VZU Atlas agent, VZU OS run #143
Chapter 03 · The schedule
14 weeks. 8 deploys. 47 tests. 0 downtime.
The schedule was tight. The team was 2 senior engineers — a lead and a senior. No juniors, no offshore, no handoffs. The brief was fixed-fee at $250k. The delivery was on a single approval. The runtime handled the orchestrator, the audit trail, and the integration. The engineers handled the brief.
Week 1: data model + auth + Stripe Connect Express sandbox. Week 2: customer-side booking flow + calendar view. Week 3: operator-side dashboard + staff management. Week 4: AI Ops agent training + first 11-location migration. Weeks 5-14: iteration, migration, and the daily deploy cycle.
The runtime shipped 8 deploys in 14 weeks. Each deploy was preview-tested on a staging environment. Each deploy was a single approval. Each deploy was auditable. The team didn't do deploys on Fridays. The team didn't do deploys in the 4 hours before a customer migration. The team did deploys on Tuesday and Thursday mornings, after a full test pass, when the AI Ops agent had confirmed zero regressions in the staging environment.
14d
brief duration · 0 downtime
Chapter 04 · The trace
What the Atlas agent actually did.
The audit trail from one of Atlas's runs on BookFlow. Every step is logged. The trace is the record.
- 01atlas › read brief
read brief
- 02atlas › thinking: brief is "build a multi-tenant booking dashboard". scope: 2 weeks.
thinking: brief is "build a multi-tenant booking dashboard". scope: 2 weeks. : brief is "build a multi-tenant booking dashboard". scope: 2 weeks.
- 03atlas › mcp.data.schema
generate schema: tenants, services, customers, bookings, payments : tenants, services, customers, bookings, payments
- 04atlas › thinking: stripe connect requires 2-day payout. need to scope webhook handler.
thinking: stripe connect requires 2-day payout. need to scope webhook handler. : stripe connect requires 2-day payout. need to scope webhook handler.
- 05atlas › mcp.code.generate
next.js 15 + prisma + stripe-connect-express
- 06atlas › mcp.code.generate
component library: Calendar, CustomerList, BookingForm : Calendar, CustomerList, BookingForm
- 07atlas › mcp.deploy.staging
ship to staging.atlas.vzu.ca
- 08atlas › mcp.test.run
47 integration tests passing
- 09atlas › ship. 14 days. 4 deploys. audit trail: 23 entries. done
ship. 14 days. 4 deploys. audit trail: 23 entries. : 23 entries.
Audit trail · 8 entries
- T+00:00:00 brief.received 01
- T+00:01:12 orchestrator.decomposed 02
- T+00:04:33 mcp.data.schema.generated 03
- T+02:18:07 mcp.code.generate.api.completed 04
- T+05:42:19 mcp.code.generate.ui.completed 05
- T+08:11:55 mcp.deploy.staging.completed 06
- T+11:33:48 mcp.test.run.completed 07
- T+14:00:00 brief.completed 08
Chapter 05 · The after
11 locations, 800 bookings a day, 80% support handled by an agent.
BookFlow shipped on time. 11 locations migrated. 800 bookings a day, no downtime. The AI Ops agent handles 80% of support tickets in the first month. The stylists stopped leaving. The operator is renewing the contract annually and adding new locations — 7 in the first 90 days, 14 in the first year.
The VZU team stayed on for the AI Ops agent. The brief expanded: now the AI Ops agent handles the booking rebalancer, the no-show predictor, the daily reconciliation report. Each is a 2-week brief. Each ships through the same Atlas agent, with the same audit trail, on the same NetWit runtime.
This is what the VZU model is built for: a fixed-fee brief, a senior engineer, an agent, a runtime. The work ships. The audit trail is on from day one. The operator owns the goal. The work is the work.
Get in touch
Brief the Atlas agent.
If your operator is fighting software, the runtime can ship a fix. Fixed-fee, fixed-scope, 14-day briefs. Senior engineers. The audit trail is on from the first character.
