Leadership by Status Request
A chief executive works with a team of C-level reports. The organization grows, and so does the number of decisions, while the way they are managed stays the same: assignments are handed out in meetings and in chat, the executive collects statuses personally, and performance is typed into a spreadsheet by hand.
Some assignments are lost between meetings. Others come back in a different shape than intended, because "done" is a matter of opinion. The executive spends hours a week finding out what is going on instead of deciding.
What a Task Tracker Misses
- Assignments arrive through seven channels. A tracker holds only what someone typed into it. A sentence said in a meeting never gets there.
- Capturing costs more than saying. Speaking takes seconds, creating a card takes minutes, so not everything gets created.
- "Done" is negotiable. Without checkable criteria, acceptance becomes a discussion in a one-to-one.
- Performance sits apart from the work. The goal sheet and the task list are not connected, so the period result is assembled by hand and contested.
The Assignment Loop, Step by Step
The solution is an AI agent that lives in the corporate messenger and runs the full life of an assignment, with no web portal or separate application to administer. It is designed to run on a dedicated machine in the company's own cloud, talks to people through a bot, and delivers any report as a file on request.
- Capture. An assignment from any source travels one route: a message, a voice note, a file, a meeting recording, a board minute, a strategy document. It lands in the owner's backlog in one format.
- Meetings. The agent is designed to announce itself at the start of a call and to record only with the participants' consent. It transcribes the recording, extracts assignments, deadlines and statuses, and keeps the supporting quote with its timestamp.
- Definition of ready. A task enters work once it has an owner, a deadline, checkable success criteria and a link to a period goal. The agent collects what is missing from the owner in short questions, without involving the executive.
- Run to the deadline. The agent requests statuses on a schedule and flags the risk of a miss before the deadline arrives.
- Acceptance. A submitted result is checked against the criteria, written as statements answered yes or no.
- Period score. The score is computed from the same tasks the team worked on. It is a proposal: a manager reviews and confirms it, and each employee can see how their score was built and contest it.
Technology Stack
Language for the Model, Scores for Code
Performance scores can feed bonuses, which makes them an employment decision. The EU AI Act treats AI that evaluates workers' performance as a high-risk use, so a manager confirms every score and the employee can see and challenge it. Everything that turns into money is calculated by ordinary code that can be tested and audited: overdue working days, the share of criteria met, the period score, penalties and multipliers.
- The model does what code cannot. It understands what was said in a meeting, checks a submitted document against the criteria and assembles the review.
- A number that affects pay is explainable. It can be tested, and it can be explained to the person whose bonus changed.
- No new interface. The work happens in the messenger the team already uses, so nobody is asked to type into one more system.
- Sensitive data stays guarded. The solution is designed to be tested on an anonymized extract in a separate environment.
Targets the Loop Is Built Around
The targets below are the ones the agent is built to meet.
- Seven assignment channels brought into one funnel and one format.
- No more than 30 minutes of executive time a week across the whole portfolio.
- No more than 10 decisions a week that require the executive.
- At least 90% of tasks with an owner, deadline and criteria within 24 hours.
- Scores and penalties calculated by deterministic code and confirmed by a manager.
Teams That Fit This Pattern
The pattern is not tied to an industry. It shows up when a company has outgrown hands-on control:
- the executive collects statuses from direct reports personally;
- assignments are handed out verbally and lost between meetings;
- managers' performance is kept in a spreadsheet unconnected to the work;
- one more system would go unused because nobody will type into it.
Manufacturers, retailers, property firms, distributors, service companies and IT businesses all reach this point, in one office or across many sites. The agent can also work in another messenger or alongside a task tracker the team already uses. The AI data analyst in Microsoft Teams splits the work between model and code in a similar way. For background, see AI agents in our glossary and our AI agent development page.
assignment channels the agent is designed to bring into one funnel
of executive time a week the loop is designed for, across the whole portfolio
of tasks designed to have an owner, deadline and criteria within 24 hours