Case study 05 · Pintu
Internal tools built with agentic AI
Moving Pintu’s budgeting and procurement off spreadsheets and email, with internal tools I built alone with Claude Code and MCP in 3–4 months, now used by every department that works with a budget.
The challenge
Budgets and purchases ran on spreadsheets and email
Company budgeting at Pintu, and procurement from purchase to payment, ran manually on spreadsheets and email. Every department spends money, so every department was part of that process: Finance, Procurement, Engineering, Marketing and Product.
Internal tools rarely make it to the top of a product roadmap, yet the work they replace is manual and repeats all year. The question was whether agentic AI could close that gap without creating new risks.
Small tools, built quickly, that people outside engineering actually use.
- 01Everyone is a user. Five departments work with budgets, each with its own needs, and all of them had to move off spreadsheets and email.
- 02Real money. Purchases and payments move company money, so every money movement needed a person’s approval, and every step needed an audit log.
- 03Built by one person. There was no team: the AI wrote the code, so the requirements and the testing had to carry the quality.
My role
Building it alone, with AI
I built the tools myself, fully with AI, using Claude Code and MCP. It took 3–4 months from the start in June 2026 to launch, including testing.
- 01Requirements. Wrote detailed requirements for every flow, because vague ones led the AI to produce code that looked right but wasn’t.
- 02Building. Built the budgeting, purchase and payment flows with Claude Code and MCP.
- 03Testing. Tested every flow by hand before it went to the departments.
Approach
How I built it
1. Start from the manual process
The tools replaced the spreadsheets and email behind two processes: company budgeting, and procurement from purchase to payment.
2. Detailed requirements first
Claude Code writes code very fast, but sometimes it hallucinates. Writing the requirements in real detail kept it on track.
3. Human approval and an audit log
Every money movement needs a person’s approval before it goes ahead, and every step is recorded in an audit log.
4. Test by hand
Every flow was tested manually before release. Testing was part of the 3–4 months, not an afterthought.
Outcome
What changed
Finance, Procurement, Engineering, Marketing and Product now work in the same tools, and every money movement is approved by a person and recorded. Some teams still use spreadsheets this year while the transition finishes.
What I learned
What I’d keep, and what I’d change
AI makes coding very fast, but its output sometimes hallucinates. What made it work was writing requirements in real detail and testing everything by hand. The engineer is still the architect; the AI is only the executor.