16 weeks, 225 features, 3 releases a day
We are building a program management and profitability system for a global automotive parts maker entirely through vibe coding and Agile SI.
- Client
- Global automotive parts maker
- Domain
- Program management & profitability
- Approach
- End-to-end vibe coding · Agile SI
- Timeline
- Kicked off June 2026 · ongoing
The project in numbers
16 wks
Since kickoff
225
Features · 12 business areas
700+
Screens
~3,000
APIs
10,000+
Commits
2,800+
Merge requests
3 / day
Releases
500+
Requirements delivered
Measured as of October 8, 2026
Challenge
The challenge
- Twelve business areas — orders, cost, profitability, quality and schedule — and 225 features had to come together in one system.
- Business teams, the PMO and developers moved together, so requirements kept arriving and changing.
- The usual SI approach — freeze requirements, go live months later — could not handle the schedule and the change at the same time.
Vibe-Ops
An AI-native delivery process
When a requirement comes in, AI drives it in a fixed order — from documents to code, release and progress records. People request, approve and check the result on screen.
How one requirement flows
- 01Submit requirement
- 02Analyze · write PRD
- 03Approve scope
- 04Dev spec · test scenarios
- 05Break into work items
- 06Build · verify locally · open MR
- 07Review · merge
- 083 releases a day · release notes
- 09Record each stage
- 10Check on screen · sign off
Who does what
| Stage | AI does | People do |
|---|---|---|
| Requirement | Analyzes it, drafts and files the PRD | Submit requirements, approve scope |
| Design | Writes dev specs and test scenarios, splits work items | Review specs |
| Build | Writes and refactors code, writes tests, verifies locally, opens MRs | Review, decide merges |
| Release | Writes release notes for each release | Approve production rollout |
| Records | Logs stage start/finish, dates and comments automatically | Final check on screen, sign off |
Three principles
Progress is never typed by hand
Records are created when a stage actually finishes, and progress is calculated from them.
Scope grows only by approval
Approving a requirement approves its scope; work items exist only inside it.
Procedures live in AI skills
From branching to checks, MRs, merges and release notes, 16 skills make everyone follow the same order.
Ops CLI
A door that connects AI to the work
Ops CLI opens the API of the Ops system — where requirements and work items live — on the command line. People use the web screens; AI uses the CLI on the same data. With a door made for AI, AI became a participant that reads requirements and records progress itself.
# 1. Read the requirement as development context$ ops req get REQ-101# 2. File the PRD and move the requirement to approved$ ops req artifact save REQ-101 prd prd.md$ ops req status REQ-101 APPROVED "PRD filed"# 3. Split it into work items$ ops dev add --process PRC-01 --title "Quote list screen" --type SCREEN$ ops dev add --process PRC-01 --title "Quote history API" --type API# 4. Record each stage — progress is calculated from here$ ops dev stage step DEV-201 BUILD DOING$ ops dev stage step DEV-201 BUILD APPROVE
How we opened it safely
Irreversible actions stay human
Deleting and revoking approvals are not in the CLI — people do them on the web.
Only within assigned work
AI can write only to the items assigned to its user.
Tokens never exposed
Auth tokens are passed so they never appear on the command line or process list.
Order enforced by the server
A stage can't start until the previous one is finished.
Open a CLI on a business system and AI can take part in that work. The Reboot Agent demo applies the same idea to sales, service and finance.
Agile SI
What was different from traditional SI
| Traditional SI | Reboot Agile SI | |
|---|---|---|
| Requirements | Frozen before kickoff | Taken one at a time and delivered right away |
| Design docs | Written by people over weeks | Drafted by AI (PRD · specs), approved by people |
| Development | Hand-written code | End-to-end vibe coding with AI agents and 16 skills |
| Releases | Phased go-lives months apart | Three times a day |
| Progress tracking | Manual spreadsheet reports | Calculated from stage records |
| Handling change | Change request → renegotiation | Absorbed one requirement at a time |
| Business involvement | Testing just before go-live | Daily checks and feedback in the dev environment |
The same idea across the business
Try an AI agent that handles several business systems from a single sentence.
Transform your organization
with AI — starting now
A simple consultation can reveal where to automate and how to boost productivity across your business.
support@reboot.im