Use this whenYou want agents or AI tools working on your business, and nobody can say where the information they would need actually lives.

Why it exists

Most organizations can’t list the systems their work runs on. Some of it sits in licensed software, some in exports, some in a spreadsheet one person keeps, and some in inbox attachments. Put an agent on top of that and it will answer confidently from whatever it can reach. An AI is only as good as its context, and the context has to be found before it can be fixed.

The framework

  1. Step 1InventoryWhere does the information live?
  2. Step 2RankWhich sources matter most?
  3. Step 3ConnectCan it be reached from one place?
  4. Step 4OwnWho answers for it?
  5. Step 5OpenWho and what can read it?
  1. Inventory

    Where does the information live?

    Build a list of the systems, spreadsheets, and inboxes the work actually touches. Start from what people use, not from what IT thinks exists.

  2. Rank

    Which sources matter most?

    Rank each by how many decisions depend on it, how often people re-key it, whether it has an API or export, and whether anyone owns it. The top few are where to start.

  3. Connect

    Can it be reached from one place?

    Use one governed store or one access layer. The goal is reachable, not necessarily moved. Leave data where it is if a good connection gets agents and people to it.

  4. Own

    Who answers for it?

    Name an owner for each source, write down what its fields mean, and say how fresh it should be. Without this, central data turns into a swamp.

  5. Open

    Who and what can read it?

    Make it readable by people and by agents, with permissions that match who is allowed to see what. Data nobody can use is only stored.

How to use it

  1. Pick one workflow and list the sources behind it. Don’t try to inventory the company.
  2. Run discovery from several directions at once, using the list of places to look below.
  3. Rank the sources, and pick the few that decide the most or get re-keyed the most.
  4. Connect those few, and name an owner and a definition for each as you do.
  5. Open them to the people and agents who need them, with permissions in place first.
  6. Repeat with the next workflow. The inventory grows as the work grows.

Where to look for systems

  • Start with the software license and finance spend list, since anything paid for shows up there.
  • Check the single sign-on app list for what people actually log into.
  • Ask each team what they open every day, and which spreadsheet actually runs their work.
  • Follow one piece of work end to end, and list every system it touches.
  • Look for exports, shadow spreadsheets, and inbox attachments acting as databases.

In practice

At Rivian, making the underlying data readable by AI, and accessible to the people building tools, became a major engineering priority. We connected systems through custom APIs and MCP, with Slack as a place people could work across them. It is the least visible part of the work, so it is easy to skip.

Where it breaks down

Don’t boil the ocean. Start with the sources behind one workflow, or this becomes a program with no useful stopping point. Centralizing without owners makes a swamp, so a source with no owner shouldn’t be connected yet. And a system with no API or export won’t get one because you asked nicely. It needs pressure on the vendor, or a replacement.

Try it with AI

Help me find and prioritize the data behind one workflow so AI and agents can use it.

Work through five parts: inventory the sources, rank them (decisions that depend on them, re-keying, API or export, owner), connect the top few, name owners and definitions, and decide who and what can read them.

Ask me one question at a time. Start with the workflow, then ask what people open every day, which spreadsheet actually runs the work, and what a single piece of work touches from start to finish. Then give me a ranked list of sources, the ones with no owner, and the ones with no way to connect.

System inventory

Related