I now have access to Microsoft Copilot. Task: draft a meeting agenda. What I get: agenda items from an email three weeks ago. Topics that have long been discussed and resolved. A budget item that's been off the table since the last steering committee.
You might say that's a Copilot problem. It's not. It's a project management context problem.
Stale data is the problem
Look at what AI sees in code and the difference becomes immediately clear. Files are versioned, changes appear as diffs, and many assumptions can be checked through types and tests. A repository still does not contain every decision or automatically provide complete context. But it gives AI a shared, inspectable work object and fast feedback loops. That is why coding agents can plan, change, and verify work far more reliably there.
And what does AI see in project management? Email chains from two weeks ago. Meeting notes that haven't been updated in months. Budget agreements from lunch that were never documented. Scope changes from hallway conversations. Knowledge that only exists in Sarah's head.
More data is not the solution
The industry's current answer? More data. More integrations. Connect Jira, search Confluence, index Slack, scan emails.
But more data doesn't solve the problem. It makes it worse. More sources that are no longer current mean more agenda items that are no longer relevant.
The right approach is not to have fewer input channels. Emails, tickets, chats, and documents remain valuable signals. But they need one confirmed reference: a living project context that reflects today's reality. Sources provide evidence; the jointly maintained context carries project truth.
Distillation, not aggregation
But how do you keep this context fresh when we're exposed to a flood of information every day that has nothing to do with the project? AI can filter, compare, and propose changes. What it should not decide alone is which statement becomes binding project truth. That requires confirmed reference context and human authority.
But it can help here too. Just differently than most people think.
MCP connections, Jira integrations, Slack connections are valuable. But they're not a suitable source for the end result. They're the right source for a distillation process: AI reads project updates like emails, meeting summaries, tickets, and checks each piece of content for project relevance. If relevant, the content becomes a candidate for the project context. The project manager as human in the loop then approves it.
Not just collecting. Distilling the right information.
Then AI generates an agenda with topics that are relevant today. Not with what was once written in some email.
This principle is at the core of Context Driven Project Management. TensorPM now puts it into practice as an AI project agent: current, human-confirmed context is the foundation, analysis and guidance are the core, and concrete execution and artifacts are the extension. Explore the framework, start collaborating on GitHub, or read why people and AI should build on shared artifacts.
Sources
- Introducing the Model Context Protocol (Anthropic)
- Context-Driven Project Management Whitepaper (CDPM)
- TensorPM: The AI Project Agent for Complex Projects (TensorPM)
- Codex, Claude and Co.: The Wrong Approach (TensorPM)