Accounts, Groups, and Task Lists — and how AI agents turn them into an Activity Intelligence Platform

The big idea, in one breath

Minecart already knows where your business is — every account, where it sits on the map, what kind of customer it is. The opportunity is to stop being a place where reps look things up and become a place that tells them what's worth doing — and then lets their own AI agent build the actual plan around their real life.

That shift has a name: an activity intelligence platform. Minecart supplies the structured truth (who your accounts are, who's grouped together, who's overdue for a visit). The rep's agent — Claude Code, Codex, Copilot, whatever they use — brings the context only it has (their calendar, where they're driving this week, their judgment). The two meet, and out comes a prioritized, trackable plan for the day.

The four building blocks

Each layer below sits on the one beneath it. The bottom is raw reality; the top is a decision you act on. They build up — and they are not interchangeable.

The funnel narrows on purpose: thousands of accounts → a few relevant lenses → one short list you can finish today.

Group vs. Task List — the crux

These two get conflated constantly, so here they are side by side. The cleanest test: you can't "finish" a group. You finish a task list.

Which accounts actually matter this week

A group can hold hundreds of accounts. Nobody visits hundreds in a week. So the platform needs an opinion about which ones are due — and that's the job of visit cadence & availability (the work in PR #1960).

This is what turns "all my Wisconsin accounts" into "the six that are overdue and sit near where you'll be Thursday." Managers also get a weekly breach report — accounts slipping past their cadence — so coverage gaps surface before they cost a deal.

The doorway for agents — the MCP

Here's the move that makes this an intelligence platform rather than another dashboard. PR #1901 adds an MCP server — think of it as a clean, secure doorway that lets an outside AI agent read from and write to Minecart on the rep's behalf.

Why that matters: the most useful context for planning a day doesn't live in Minecart. It lives with the rep — their calendar, the trip they're planning, the customer who just called. An MCP lets the rep's own agent combine the two worlds:

The agent can read accounts, groups, and what's due, and it can write a finished plan back — creating and ordering an Account Task List, and later reconciling whether the work actually got done. The platform provides the raw material; the agent assembles it into something specific to one rep, one week, one route.

The payoff: a trip to Chicago

This is the example to remember — it shows all four building blocks and the agent working together.

Who decides what

The reason this works — and stays trustworthy — is a clear division of labor. The platform never pretends to be the human.

Minecart provides the menu of what's worth doing. The agent drafts the plan. The rep chooses. The choice lands back in an Account Task List — and the loop closes.

Where this stands today

So the picture is honest: the foundation already exists in the product; the agent layer is the work in flight.