A distributed system is a productivity killer

Last week I wrote about why product people need an AI operating system. This week is about one part of it: connecting your tools. Let's dig in.

The best part of product work is deciding what matters. Around it, there's a lot of keeping track. The status is in the tracker, the decisions in email and chat, the plan in a doc, the deadline on the calendar. You're the human API between them.

Every "where does this stand?" means four tabs and stitching the picture together by hand. None of it is hard, it just eats your time. And sometimes you find out the pieces don't agree, or something important is sitting with nobody.

So I built something for it, first on a real client project. Below I show it on a made-up company.

Connect first, then give it a job

The obvious move is to connect your AI to all your tools: tracker, email, docs, calendar. Do it. That alone is worth it: you get one interface to see all your work.

It gets even more useful once you give it a job, because it can compare your tools and catch things you'd miss looking at them one by one.

When I looked at what I actually needed, it came down to three jobs:

• Where does X stand? One initiative, across every tool.

• What needs me today? Everything waiting on you, across every inbox.

• Prep me for this meeting. The context you hunt for five minutes before a call.

I started with the first, because it's the one I find myself answering most often.

The use case: stay on top of a big initiative

Here's an example with a made-up company, Hivelist. Their biggest customer renews in five weeks, and they've said they need a CSV export to renew. Everything about it is spread across a proposal, a few emails, a Linear backlog and the calendar.

The proposal promises the export this month. In Linear, the ticket sits in the backlog with no owner. The customer's email asking for a date has no reply. That's the distributed system in one picture.

With the Initiative Radar, the PM goes into the next stakeholder meeting with three things to decide: staff the export now or give the customer a new date, who owns reporting, and who answers their security team.

How I built it, step by step

What I ended up with is a live artifact inside Claude, a small app Claude builds and hosts for you. I call it the Initiative Radar. Here's the build, in the order I'd do it again.

1. Map roles, not tools. Every initiative has the same four roles:

• Work: the tracker

• Conversations: email or chat

• Source of truth: the plan or the offer

• Moments: the calendar

Mine were Linear, Gmail, Google Drive and Google Calendar. Yours might be Jira, Slack and Confluence. The approach doesn't change.

2. Make it an artifact, not a one-off answer. An initiative runs for weeks or months, and you check on it again and again. With an artifact, I just open it and see where things stand. It reads your tools through Claude's connectors, with your own login. Nothing gets copied into it.

3. Ask for what needs attention, not a summary. A summary tells you what each tool says. I wanted to know what needs me. So every finding lands in one of three types:

• Deadline at risk: something date-bound isn't on track.

• Misaligned: people or records say different things.

• No owner: something was asked or agreed, and nobody picked it up.

Each one ends in a single question to settle, with a "settle by" date.

4. Write a one-page brief. Who's who. What's agreed. What's still open. And which source wins for each topic. Claude reads it before anything else.

Why bother, if the AI can already read all your tools? Because the tools can't tell it which one to believe. Take the export again. Without a rule, the AI has to guess whether the proposal or Linear is the plan, and it could guess differently on every check. One line in the brief settles it: "Linear is the truth for what's being built. The proposal is the truth for what we promised." Now if something's promised but has no ticket, the Radar flags it. 

I borrowed this from how data teams build their AI analysts. Anthropic wrote about their internal analytics agent: with short written guides to read first, its accuracy went from 21% to about 95%. My brief is the one-initiative version of that.

Build yours

You don't need my Radar. You need four things:

1. Connect the tools your work lives in. 

2. Pick one initiative that spans at least three of them.

3. Write the brief: who, what's agreed, what's open, which source wins.

4. Ask what needs your attention, and by when. Not "summarise this".

Start there, even in a plain chat. The polished artifact can come later.

See it live. I'll demo this exact artifact in a free Lightning Session on 7 October at 17:00 Berlin time. You’ll also learn more about how to build your own product AI operating system.