Back to Use Cases
Operations

From WhatsApp Screenshot to Full Financial Stack

My accountant sends tasks via WhatsApp. My AI doesn't just read them—it builds the tools to execute them.

My accountant sends me a WhatsApp message with a screenshot. A list of financial tasks: update invoices, check VAT numbers, reconcile payments. I forward the screenshot to my AI colleague. "Handle this." What happens next isn't task execution. It's infrastructure creation.

What Actually Happened

The screenshot contained tasks across multiple systems I didn't have CLI access to. Most people would manually log into each system. My colleague took a different approach: "I don't have access to Odoo yet. Want me to build a CLI?" Yes. "Done. I also need Stripe for Coel.ai invoices. Should I configure that too?" Yes. "I'm seeing a VAT discrepancy between the systems. Want me to build a comparison tool?" Yes. One screenshot. Three CLIs built. Infrastructure that didn't exist 30 minutes ago—now permanent tools in my stack.

The System That Enabled This

This happened because of how my AI operates:

  • Vault context — My AI knows all my projects, entities, and their relationships. It knew Odoo was for OsteosOnline, Stripe was for Coel.ai.
  • Tool-building capability — When a tool doesn't exist, it builds one. Not a workaround. A proper CLI I can use forever.
  • Notion integration — Each task from the screenshot became a tracked item in the right project. Nothing lost.
  • Permission-based execution — It asked before building each tool. I stayed in control while it did the work.

The Stack It Built

From that single screenshot, my colleague created: • Odoo CLI — Query invoices, check payment status, update records. Now I can ask "show unpaid OsteosOnline invoices" anytime. • Stripe CLI config — Connected to Coel.ai account. Invoice status, payment reconciliation, subscription checks. • VAT comparison tool — Pulls VAT totals from both systems, flags discrepancies. What used to be manual spreadsheet work. • Notion task tracking — Every accountant request is now a tracked task with the right project context. This isn't a one-time automation. These tools persist. They compound.

The Reality

My accountant still sends WhatsApp screenshots. But now: - Screenshot arrives - I forward to colleague - Tasks extracted, tools already exist - Execution happens, I verify results - Done in minutes, not hours The first screenshot took 30 minutes because we were building. Now it takes 5 minutes because the infrastructure exists. That's the difference between AI that executes tasks and AI that builds systems.

Why This Matters

Most AI tools would try to do the tasks manually. Copy data, fill forms, repeat next month. My colleague builds tools. Once the Odoo CLI exists, it exists forever. The investment compounds. This is what infrastructure means. Not doing work—building the systems that make future work trivial. One WhatsApp screenshot. Permanent operational infrastructure. That's the game.

Want to build something like this?

I help people design AI infrastructure that actually works. Not chatbots. Not wrappers. Real systems.

Let's talk