SCNET · Enterprise IT · Ankara, Türkiye

Sanal Çekirdek

Ask an agent not what it can do, but what it cannot undo.

Agentic workflows choose their steps from context and reach enterprise systems through tools. This piece is not about what such an agent can do; it is about where the permission boundary should sit.

Short answer: the threshold follows reversibility, not difficulty. Every step that cannot be undone belongs to a human.

The wrong question: 'can the agent do this?'

Technically an agent can call any tool it is wired to, so the capability question never shapes the design. The question that does: if this step goes wrong, can it be undone, who notices, and how long does the repair take?

  • The capability question always answers yes
  • The real criteria are impact and reversibility
  • Time to notice is part of the risk
  • Cost of repair pulls the threshold down

Three buckets: free, logged, approved

Sorting actions into three buckets settles the question faster than debating them one by one. Reading and drafting can run free; state changes that can be undone run with a log; financial, legal or customer-facing steps wait for approval.

  • Free: reading, searching, drafting
  • Logged: state changes that can be reversed
  • Approved: money, contracts, outbound messages
  • Moving an action between buckets is its own decision

Approval is context, not a button

An approval screen does more than ask for a click. It shows which data the agent read, why it chose this step, and what the path back looks like. Approval without context transfers responsibility on paper only.

  • The sources consulted are visible
  • The reason for the chosen step is written down
  • The rollback path is known at approval time
  • Who approved, and when, is recorded

Test the boundary in production

A permission threshold is tuned against real cases, not at a desk. In the first period the agent only proposes; its proposals are compared with human decisions. Where agreement is high the threshold moves up; where it is not, the workflow is corrected.

  • The first period runs in suggest-only mode
  • Proposals are compared against human decisions
  • The threshold moves on measurement, not on guesswork
  • Every threshold change is recorded with its date

How we work

  1. List every tool the agent can call
  2. Score each tool on reversibility
  3. Sort the tools into free, logged and approved
  4. Put the reason and the rollback path on the approval screen
  5. Measure the threshold in suggest-only mode, then move it

How success is measured

  • No irreversible step runs without approval
  • The approval screen shows reason and rollback path
  • Agreement between proposal and human decision is measured
  • Threshold changes carry a date and a reason

Frequently asked questions

Does approval on every step stall the workflow?

It would, which is why approval belongs only on steps that cannot be undone. With reading and drafting left free, most of the flow runs without interruption.

What happens if the agent calls the wrong tool?

Crossing a defined boundary halts the flow, the action log is reviewed, and the previous version is restored if needed. This behavior is designed in from the start, not added after an incident.

How detailed should the approval record be?

Detailed enough to reconstruct the decision later: which data, which reason, who and when. 'The system said so' is not an explanation.

Share the list of tools your agent will call and we will mark which ones belong behind approval.

Let's draw your agent permission threshold