Safe Lane

Add intake, fanout, and review gates when the source of truth is unclear, so agent work is checked before it acts.

For leaders

The problem in one sentence. Some agent tasks arrive with unclear requirements or high stakes, and acting immediately commits the work to a direction nobody confirmed.

What the pattern changes. Uncertain work pauses for a written scope, parallel read-only exploration, and an approval before anything acts.

The failure it prevents. A confident change built on the wrong source of truth, discovered after it landed.

What to ask your team. Which of our agent tasks can act before anyone confirms the source of truth, and what would a required pause cost us?

The Problem

Some agent work arrives without a clear source of truth. The requirements are ambiguous, the right file or system is unknown, or the task touches something costly to undo. An agent that acts immediately on this kind of work commits to a direction before anyone has confirmed it is the right one. When the direction is wrong, the cost is a bad change that has already landed, or an outward action that has already been sent. The work needed a pause for grounding and review, and the fast path gave it none.

The Pattern

Route uncertain or high-risk work into a separate lane with three stages before it is allowed to act. The pathway routing pattern is what sends work here.

1. Intake. State the task, the unknowns, and what a good result looks like, before any action. Intake turns a vague request into a written scope that a reviewer can hold the result against.
2. Fanout. Explore in parallel across sources and options rather than committing to the first idea. Each probe is read-only and returns what it found. Fanout surfaces the real source of truth and the competing approaches while nothing has been changed yet.
3. Review gate. A reviewer, a person or a checking step, approves the grounded plan before any action runs. Only approved work proceeds to act.

The lane is deliberately slower than the build path. It is reserved for the small share of work that earns the extra care, so the cost is paid only where uncertainty or stakes are high.

Diagram

flowchart TB
    entry["Uncertain or high-risk task"] --> intake["Intake: scope and unknowns"]
    intake --> fanout["Read-only probes"]
    fanout --> plan["Grounded plan with sources"]
    plan --> gate{"Review gate"}
    gate -->|approved| act["Act"]
    gate -->|needs work| intake
    act --> final["Final scope check"]
    final --> done["Result with sources"]

Template Or Small Example

A safe-lane intake can be captured in a short, fixed form that travels with the task through the lane:

Task:        one sentence on what is being asked
Unknowns:    the specific things not yet confirmed
Sources:     where the source of truth is expected to live
Good result: how a reviewer will judge the output
Blast radius: what is affected if this is wrong
Approval:    who must approve before any action runs

The intake form is filled first. Fanout fills in the sources and confirms or corrects the unknowns. The review gate reads the completed form and either approves the plan or sends it back. The final check compares the result against the "good result" line written at intake, so success is measured against a bar set before the work began.

Failure Mode It Guards Against

The failure it prevents is premature commitment: an agent acts on an unclear task, picks the wrong source of truth, and produces a change or an action that is confidently wrong and already done. The safe lane forces grounding and approval to happen before the action, so a wrong direction is caught while it is still a plan.

What This Pattern Does Not Prove

A review gate confirms that a plan was approved. It does not guarantee the approver judged well, and it does not guarantee the fanout found every relevant source. The safe lane raises the chance that uncertain work is grounded and checked before it acts. It does not make the underlying answer correct on its own. Important claims still need a human check.

Proof Note

The intake form above is reproducible on its own. A reader can adopt it for their own uncertain tasks and see the effect on how work is scoped and reviewed, with no private system involved. The module offers the form as a reusable method and claims no measured result.