What Zeno Household taught us about building AI-first finance tools
If you build financial software today, the temptation is to bolt a chat sidebar onto every page and call it AI-ready.
While building Zeno Household, our sibling app for personal and household budgeting, we learned that a bolted-on chat box is the wrong abstraction. People already have an assistant they use every day, such as Claude or ChatGPT, with their own context, memory, and preferences. They do not want another isolated chat window inside every SaaS tool.
What they want is for their own assistant to be able to read their numbers, explain what is happening, and prepare changes cleanly.
No embedded chat, full MCP coverage
Zeno Budget adopts the architecture we proved in Zeno Household:
- Bring your own assistant. There is no chat widget inside Zeno Budget. The web console is for looking at the grid, typing numbers, and approving changes.
- Every screen has a matching MCP tool. Anything a person can see or do in the console, their assistant can read or prepare over MCP, calling the exact same underlying service functions.
- Every screen has its own card. When an assistant fetches your cash flow, checks forecast completeness, or previews a change request, it can return a structured interactive card right inside your chat.
Enforcing the rule with tests
It is easy to promise “AI-first” on day one and then ship three new console screens without updating the MCP server.
In Zeno Household, we solved that with an automated test that inspects every API route in the application and requires it to map to a matching MCP tool, or be explicitly listed as console-only with a documented reason. We brought that same gate into Zeno Budget.
The result is that both apps reinforce each other: patterns for MCP cards, tool descriptions, and route parity tested in Zeno Household carry straight into Zeno Budget, and improvements to recurring cash flow planning flow back the other way.