Most writing about agentic project management stays at the level of definitions: agents that act, not assistants that answer. That is true and not very useful on a Monday. What makes the idea concrete is the work itself. Below are six workflows almost every team runs, described twice: once as they usually happen, with a person carrying the coordination, and once as they run in Kavanah with an AI employee doing it inside the limits the team set. None of them is exotic. That is the point.
1. 1. Monday Planning
Before: the lead spends the first hour of the week reading the backlog, checking who is out, guessing capacity and moving cards, then the planning meeting spends another hour sorting. After: a scheduled planning run reads the backlog, availability and past estimates on Monday morning, drafts the sprint with owners and dates, and posts it for the lead to edit. The meeting becomes twenty minutes of deciding instead of two hours of arranging. The agent runs in safe mode, so it can create and assign work but cannot send or delete anything while it does.
2. 2. Request Triage
Before: requests arrive by email, chat and form, and someone sorts them into projects when they get a moment, which is usually not the same day. After: new requests are triaged on a schedule into the right project, with a proposed owner chosen by skill and capacity and a due date. The time from request to owner drops from days to minutes, and it is measurable, because Kavanah records when a request arrived and when it was assigned.
3. 3. The Standup
Before: fifteen people say what they did yesterday, mostly to the one person who needs to know. After: the agent writes the standup from what actually changed in the workspace, tasks moved, commits merged, blockers raised, and posts it before the meeting. The meeting, if it survives at all, is about the two blockers that need a decision. Many teams find the written standup is enough and stop meeting daily.
4. 4. Chasing a Blocker
Before: a task sits waiting on a reply from another team, and nobody notices until the deadline. After: the agent watches for tasks that have not moved past their expected time, nudges the owner, and if the blocker is an outside reply, notes it in the standup so a person can decide whether to escalate. It does not escalate on its own, because the team put escalation behind an approval.
5. 5. The Client Update
Before: the account lead writes a weekly email per client on Friday afternoon from memory and three dashboards. After: the agent drafts each client's update from the project record, with every sentence traceable to a task or a thread, and parks it in the approval queue. External email is a category the team set to approve, so nothing reaches a client until a person reads it and presses send.
6. 6. The Weekly Report
Before: someone spends Friday assembling a status report from five tools, and it is out of date by Monday. After: a scheduled report goes to the people who need it, built from the workspace's own data, and the executive version flags the projects slipping against their dates. The person who used to write it now reads it, checks the two numbers that look wrong, and spends the rest of Friday on the work.
Notice what stays human in all six: deciding what matters, approving anything that leaves the building, and making the call on a blocker. The agent carries the coordination; the team keeps the judgment. The full model, with the autonomy ladder and the controls behind each of these, is on our agentic project management page.
