Skip to main content

Mitra recommends the next step; a person decides

Continuing from A draft payroll contract, reviewed and explicitly activated: a draft payroll contract stays a draft until a person activates it. This page shows what Mitra does while that is still undone. It reads the company's records, says what it found, and recommends a next step. Taking that step stays with a person.

Illustrative demonstration prepared using publicly available company context and synthetic data. This page does not represent a production deployment, customer endorsement, or use of real employee data.

The company this page follows

Same illustrative demo structure as the rest of this book: one fictional company, Meridian Interiors. Its name, its people and every figure shown are synthetic; none of it represents a real customer, a real headcount or a real payroll. Some of its people still hold draft payroll contracts, which makes them a good example of something Mitra can notice and cannot fix.

Ask Mitra from the Control Center

Mitra opens as a side panel from the top of the Control Center. The panel has two tabs, Ask and Actions & audit, and a Voice control that reads the briefing aloud. Under the question box, the panel's own footer says: "Mitra reads only what your permissions allow, and explains rather than acts."

Asked what needs attention first in payroll, Mitra answers from the company's own records and marks the answer "Computed from your records". The answer carries a suggested priority and a link that opens the payroll screen. The suggestion is a sentence for a person to act on. Nothing is changed by asking.

The Actions & audit tab is for asking Mitra to look something up, or to do something small on your behalf. Its own text says: "Anything that writes data is only ever a proposal: nothing changes until you confirm it."

Mitra Operations: what Mitra is allowed to do

Administrators see a separate area, Settings, then Mitra, with seven screens: Autonomous Operations, Policies, Activity, Action Review, Simulation, Engine and Audit. The first screen shows a status line and the operational controls. In this company the status reads: "Mitra is watching 10 scopes and recommending. It is not changing anything." The Autonomous execution control reads OFF.

Autonomous execution is a switch an administrator controls. This page describes the company as recorded, with the switch off. It does not say that every company runs this way, and it does not say Mitra acts on its own anywhere. Where an administrator does turn execution on, the scope cards on the same screen show it, and the list of autonomous actions shows what Mitra may do. Those actions concern Help Desk cases, notes, notifications and re-checks. Activating a payroll contract is not on that list.

The Payroll scope card describes its job in one line: "Identify employees payroll cannot pay: no active contract, a contract about to lapse, or a payslip a run missed." Its detection and recommendation switches are on and its execution switch is off.

A finding, and its recommendation

The Activity screen lists every situation Mitra has detected, what it decided and the evidence behind it. One entry reads "No active contract: Meera Iyengar". Opening it shows:

  • A one-line summary: the employee is active, has no contract in the Active state, and the contracts on file are draft.
  • Impact: payroll computes wage, pay frequency and filing status from the contract, so this employee will not be paid by the next payroll run.
  • Recommended action: "Create or activate this employee's contract before the next payroll run."
  • A decision label, "Recommend for human review", and the reason "Autonomous execution is switched off globally."
  • What was done: "No action has been taken on this situation."
  • Evidence: each line names the database record it came from. The screen says that nothing in the evidence was written by a language model.

Action Review: a person decides

Action Review is where Mitra's proposals wait for a decision. Its subtitle reads: "Mitra proposes; a human decides; the approved action then runs as that human." When nothing is waiting, it says "Nothing to review." A note at the bottom adds that an approved action runs as the person who approved it, and that Mitra has no account of its own to run it with.

The person's side

The employee's own record shows the same facts from the other side. For the employee in this example, the Compensation and Compliance block still reads Contract: Draft and Bank account: Not set. Mitra pointed at the gap. It is closed by a person, on the screens described in A draft payroll contract, reviewed and explicitly activated.

Detections follow a schedule

Mitra's scan runs on a schedule. In this company the Engine screen shows it every 900 seconds, which is every 15 minutes. A finding therefore appears some time after a record changes, and a finding can stay listed for a while after the record behind it has been fixed. The contract and the employee record are correct the moment someone saves them. Mitra's list catches up on its next pass.

What this page does and does not claim

  • Mitra reads and recommends. In the screens shown, no action was taken by Mitra.
  • Autonomous execution is a switch an administrator controls. It is off in the company shown.
  • No claim is made that Mitra acts on its own in any scope, and none is made that Mitra can activate a payroll contract.
  • Detections are not instant. Plan for the scan interval.

No comments to display