Publish Date :
What an AI Agent Should Never Do Without Approval
An agent should wait when a proposed action exceeds its permitted spend, changes protected terms or lacks the evidence required by the retailer. Yukti supports rules and proposals; the useful starting point is a clear decision owner, a defined action and a record of what was authorised.

THE MERCHMIX DECISION / Yukti
A sensible proposal still needs authority
Yukti needs an explicit authority boundary for spend, price and supplier commitments, even when the recommendation is commercially plausible.
Independent retail planning example. Financial amounts are US dollars. Forecasts, quantities and proposals are assumptions for this decision, not customer results.
Cream knit / purchasing authority
| Authority check | Position |
|---|---|
| Purchase commitment | Person required above limit |
| Price change | Proposal only |
| Supplier terms | Separate commercial acceptance |
The limits are retailer-selected assumptions, not standard Yukti settings.
Name the permitted action before testing its financial limit.
THE KEY IDEA
Make every proposed action show its commercial scope, applicable authority and actual outcome. A plausible recommendation cannot grant itself permission.
Define authority before discussing autonomy
A reorder, a stock transfer and a price change create different obligations. Permission to recommend all three does not mean permission to execute all three. Set the action type, merchandise scope, locations, financial basis and rule owner before deciding which work may proceed within a limit.
The public Yukti description distinguishes proposals, bounded action and actions requiring a person. Treat those as capabilities to configure and verify for your deployment. A policy is incomplete if it names a dollar limit but leaves the supplier, affected products, expiry or evidence requirements unspecified.
Make the commitment visible in the same units as the rule
Consider an independent apparel retailer reviewing 80 cream knit tops at $28 each. The proposed goods commitment is $2,240. Its chosen rule allows goods commitments only up to $2,000 without further review. The proposal exceeds that particular limit by $240 and should wait.
The figures exclude tax, freight and other charges. A retailer might instead control total landed commitment, a weekly supplier total or cumulative spending. Yukti needs the applicable policy basis; a goods-cost calculation cannot establish compliance with a rule defined on a different amount.
Do not turn a lower quantity into an approval shortcut
A 70-unit alternative would cost $1,960. That amount falls below the illustrative limit, but it is a different stock decision. The ten-unit reduction may leave a size short or fail to cover demand before the next receipt. A smaller number alone does not make the alternative suitable.
Repeated small orders can also exceed the commercial intent of a limit. Define whether related requests are assessed together and who checks the total. Present the original requirement to the authorised person rather than dividing it into smaller actions simply to make each appear permissible.
THE FIGURES BEHIND THE DECISION
| Record | Position |
|---|---|
| 80-unit reorder | $2,240 / awaiting review |
| 70-unit alternative | $1,960 / not selected |
| Price and supplier terms | No change authorised |
80 × $28 = $2,240. The $2,000 limit is exceeded by $240. An unselected 70-unit alternative would cost $1,960.
Treat price and supplier changes as separate authority
A recommendation to reduce a selling price affects a different control from a recommendation to buy more units. Pricing scope includes eligible products, channels, start and end dates, overlapping offers and the person accountable for the commercial decision. A purchasing rule does not settle those questions.
The same applies to changing a supplier or agreeing new delivery terms. An agent may organise the evidence and prepare a proposal. The retailer must still establish the supported approval and execution path. The animation keeps all such changes unexecuted instead of showing a reassuring tick as proof of authority.
Give Yukti a decision that can be reviewed
For the 80-unit proposal, retain the quantity, $28 cost basis, $2,240 total, selected rule and reason for escalation. Include the demand window and relevant stock position. A reviewer should be able to understand what accepting the proposal would actually commit.
Yukti can present that bounded action with its reasoning. If the reviewer requests 70 units instead, retain the changed scope and its justification. If the input stock or price changes before action, revisit the comparison rather than assuming that an earlier approval covers a different commercial position.
Separate approval from a confirmed outcome
Approval records a decision by the appropriate authority. A purchase order identifier, successful system response and supplier acknowledgement describe later states with different meanings. Keep them distinct so a proposed reorder is not reported as available supply.
For this example, the conclusion remains awaiting review. There is no claimed transaction or invented approver. A real implementation should reconcile the authorised instruction with the supported operational result, including any failure or partial completion, before presenting the action as finished.
| Signal | Merchmix workspace | Decision to review |
|---|---|---|
| Decision scope | Yukti / evidence | Name the permitted action before testing its financial limit. |
| Spend test | Yukti / comparison | 80 units at $28 create a $2,240 goods commitment. |
| Review record | Yukti / decision review | No purchase order, price change or supplier commitment is executed in this illustration. |
Review the rule as trading conditions change
A limit that made sense for an established continuity line may be unsuitable for a seasonal test or a new supplier. Assign responsibility for reviewing the scope, data freshness and permitted actions. Make any change to authority deliberate and visible.
Start with the retail decision you need to control, then test whether the supported Yukti workflow applies that boundary. The goal is a recommendation that saves preparation while preserving a meaningful decision, rather than an approval button that hides the commercial consequence.
BEFORE THE NEXT DECISION
- Planning owner
Tax, freight and other charges are excluded; define which costs count in the actual policy.
- Commercial owner
The alternative needs its own commercial case and any applicable source, category and timing checks.
- Decision owner
No purchase order, price change or supplier commitment is executed in this illustration.
Common questions
Should every agent action require a new approval?
The retailer sets the supported rules. Actions outside a permitted scope or limit should wait; bounded authority still needs a named owner and review.
Why does the 80-unit reorder need review?
Its $2,240 goods commitment exceeds the example’s $2,000 limit by $240. Other costs are excluded from this specific calculation.
Can Yukti simply order 70 units instead?
That is a different proposal costing $1,960. Demand, cover and all other applicable rules still need review.
Does approval mean stock is available?
No. The order, supplier response, arrival and receipt are separate operational states.
Are these standard Yukti spending limits?
No. The quantities and limits are independent assumptions illustrating a retailer-defined authority boundary.
FROM THE ARTICLE TO YOUR OWN RETAIL PLAN
Review the decision.
See the workflow.
Bring your planning questions to a demo, or explore a proof of concept on your own data.