Answer
How should a main agent delegate work?
With a self-contained brief, a defined output, and everything needed to answer, because the worker cannot come back and ask.
With a brief that stands alone. State the question, what to work from or how to find it, what the answer must contain, and what evidence it must carry. The worker cannot ask, so an incomplete brief becomes a guess.
The defining constraint is the absence of a back-channel. Everything a person would clarify mid-task has to be anticipated, and anything not anticipated is decided by the worker without the context that would inform it. This makes brief quality the dominant variable in delegated work, more so than in ordinary agent work where a misunderstanding surfaces in a turn or two.
The question comes first and should be answerable rather than open. Find where the retry interval is configured and what sets it is answerable. Look into the retry behaviour is not, and it returns an essay whose relationship to the actual need is a matter of luck. The narrower the question, the smaller the returned answer and the higher the value of the delegation.
The output specification comes second, and it should describe the shape as well as the content. What sections, in what order, how long, and critically what evidence each claim must carry. A returned answer with file references or quotations can be spot-checked; one without cannot be used without repeating the work, which is the failure that makes delegation pointless.
Third is what the worker may and may not do. It is operating with less context and therefore with a worse basis for judgement, so the bounds should be tighter than the main run's: read this, do not change anything, do not follow the problem into other systems. This also keeps the returned result interpretable, since a worker that took unexpected actions returns a summary that mixes findings with side effects.
Fourth is what to do when the question cannot be answered. Without an instruction, a worker will produce the best answer it can rather than reporting the obstacle, and the best answer it can produce from insufficient material is exactly the confident plausible thing you do not want. Saying explicitly that reporting an obstacle is an acceptable outcome changes what comes back, and it is one line.
Finally, the main run should keep the brief. When the result arrives, the useful question is whether it answered what was asked, and that comparison requires the question in its original form rather than as remembered. This sounds trivial and is the difference between accepting a result because it is coherent and accepting it because it is responsive.
A delegated brief is a letter, not a conversation, and every ambiguity in it will be resolved by somebody who cannot check with you.
Siddharth Sharma, Context Theory
Related questions
Should the brief include the reasoning behind the task?
Enough of it to disambiguate, and no more. A sentence saying why the question matters helps a worker choose between readings; a paragraph of background is context it has to process without needing. The test is whether a line would change the answer to a plausible ambiguity, and most background does not.
How much material should be attached versus fetched?
Attach what the first step needs and let the rest be fetched, for the same reason as anywhere else. The exception is any constraint that governs the work, which must be attached because the worker will not know to look for it. Governing material in front, situational material behind a search.
METHOD
Every figure below carries its source and the date it was verified. Nothing on this page is asserted.
The numbers on this page.
| What | Value | Specific to |
|---|---|---|
| Sub-15-minute compliance — automated routing vs manual only | 62.5% vs 39.1% | Category-wide |
| Close rate — response under 5 minutes vs over 24 hours | 32% vs 12% | Category-wide |
2026 speed-to-lead benchmark · verified
Optifai speed-to-lead benchmark · n=939 companies · Q2 2025–Q1 2026 · verified
What is specific to this page.
| Kind | Claim | Check it against |
|---|---|---|
| Workflow | The absence of a back-channel makes brief completeness the dominant variable in delegated work, because every unanticipated ambiguity is resolved by a worker without the context that would inform it and without the chance to surface it. | Comparing outcomes on the same delegated task with a complete brief and with one containing a known ambiguity. |
| Response | An answerable question returns a bounded answer while an open one returns an essay whose fit to the need is incidental, so question narrowness determines both the size of the result and the value of the delegation. | Comparing the length and usability of results from an open and a narrowed version of the same delegated question. |
| Software | A worker operating with less context has a worse basis for judgement, so its permitted actions should be narrower than the main run's, which also keeps its report interpretable by preventing findings from being mixed with side effects. | Comparing reports from delegated runs permitted to act against those restricted to reading. |
| Constraint | Without an explicit statement that reporting an obstacle is acceptable, a worker returns the best answer available from insufficient material rather than the obstacle, producing a confident result with no indication of the deficiency. | Delegating a question whose answer is not available in the supplied material and observing whether the obstacle is reported. |
Each row would be wrong on another industry's page. Where a sourced figure exists it is in the table above instead; these are the constraints that shape the work and do not happen to be numbers.
Start with the measurement.
Reading about a benchmark is not the same as knowing your own number. The audit produces yours, measured rather than estimated.
$497 · delivered in 5 business days · credited against month one