A deliverable comes back wrong. Not slightly off. Wrong problem, wrong shape, wrong everything.
Your first instinct is to blame whoever built it. Did they even read the brief. You start drafting the message in your head, the one where you explain, again, more slowly this time.
Before you send it, check the brief. Nine times out of ten, the person did exactly what you gave them. The problem is what you gave them was never actually the plan. It was a fragment of the plan, typed fast, with the part that made it obvious to you still sitting in your head.
Why does delegation fail even when I explained the task clearly?
It fails because “clearly” almost always means the steps, not the goal or the reasoning behind them. You handed over a checklist. You kept the why. The person executing had no way to make a good call once reality didn’t match the checklist, which is most of the time.
Here’s what everyone tells you when a handoff goes sideways: find better people. Hire someone with more initiative. Write a more detailed prompt. All of that treats the failure as a listening problem, on their end.
That’s the conventional wisdom, and it’s wrong. Or at least incomplete enough to keep costing you money.
It’s not a listening problem. It’s a documentation problem.
You had the whole plan mapped out in your head. You could walk through every step of it out loud and it made sense. That’s not the same as a plan someone else can execute, and the gap between those two things is where the failure actually lives.
When you hand a task off, it gets compressed to whatever you had time to type into a Slack message or say on a five-minute call. The reasoning that made the plan obvious to you, why you ruled out three other approaches, what “done” actually looks like, almost never survives that compression. You know it. Nobody else does.
There’s decent evidence this isn’t a soft claim. A meta-analysis of role ambiguity and job performance across dozens of studies found a real negative correlation between how unclear a role is and how well someone performs in it (Tubre & Collins, 2000). It’s correlational, and it’s not the same as blaming the founder specifically, but it lines up with what you already suspect: the gap isn’t intelligence or effort. It’s clarity that never got written down.
And the person on the receiving end isn’t experiencing it as “my founder trusts me with autonomy.” They’re experiencing a guessing game with real stakes attached.
What should I actually write down before I hand something off?
Four things, in order: the goal, the why, the steps, and the specific decision points where you’d want to be looped in. Most people write the goal and the steps. They skip the why, because it feels like the part that doesn’t need saying. It’s the part that matters most.
Two operators have independently landed on close to the same fix, from different angles.
Matt Gray, in “The Art of Delegation,” tells founders to start a doc for every task and “write down the goals, the steps, and most importantly the why behind each task.” His point: the why is what lets someone make a good call when the situation doesn’t match the plan exactly, which it never does past step two.
Dan Martell runs a sharper version of the same interrogation. In his framework on building SOPs, he calls it the vacation test: if you had to hire someone to do your job for two months because you were going on vacation, what would you need to walk them through? Not a polished manual. Just everything they’d need to know so they wouldn’t have to call you.
Ask that question honestly and you surface the stuff that never makes it into a normal handoff. What counts as done. What to do if the client pushes back on price. Why the third step happens before the second one. Those are the gaps that sink a delegation, and they stay invisible until you force yourself to answer for them in writing.
There’s research behind why steps alone are the wrong lever here. A study of Ghana’s civil service tracking 3,620 tasks across 45 government organizations found that granting bounded discretion, not tighter monitoring, was what improved completion on tasks that were unclear going in but measurable after the fact (Rasul, Rogger & Williams, 2021). Define the outcome and the boundaries. Don’t try to script every step of something that can’t be fully scripted.
Most founders never run this interrogation on themselves. Not because it’s hard. It takes ten minutes. Because a Tuesday with six fires burning doesn’t leave room for it, so the instructions ship half finished and the gaps go straight into the handoff.
This is exactly where an AI thinking partner earns its keep, before the handoff, not during it. Not to write the brief for you. To ask the questions you’d skip on your own: what’s the actual goal, not the task in front of you. Why does this rank above the other three things on your list today. What would you tell someone if they got this wrong. What decisions are you assuming they’ll make the way you would, without ever having said how you’d make them. Run through that once and you’ve done the vacation test on the specific thing in front of you right now, in the time it takes to get coffee.
Why don’t new hires just ask when they’re confused?
Because asking feels like admitting they can’t handle it, so they default to a plausible guess instead. This is a documented pattern, not a personality flaw, and it’s the mechanism that turns a vague brief into an expensive mistake instead of a quick question.
Amy Edmondson’s research on psychological safety describes this as a form of self-protection: people constantly manage the impression they’re giving off, and flagging confusion reads as looking incompetent, so they suppress the instinct that would have prevented the rework (Edmondson, 1999). The person who could have caught the problem early is actively sitting on the question that would have caught it.
An AI agent hits the same wall from a different direction. It doesn’t have anything to suppress, so it never even gets to the “should I ask” moment. It just drifts on whatever it inferred your intent to be, and it doesn’t throw an error when it drifts. The output looks fine on step one and wrong by step seven, and nothing along the way told you to check.
Same root failure, unstated intent, two different silent delivery mechanisms. One is social. One is mechanical. Both end with you finding out at the worst possible time.
How much does bad delegation actually cost a small business?
There’s no reliable industry number for this, and you should be skeptical of anyone who gives you one. I looked. The figures that circulate, a productivity loss percentage here, a dollar figure there, don’t trace back to an actual study when you follow the citation. Treat any specific claim like that as marketing, not evidence.
What’s real is what you can measure in your own business:
| Cost | What it looks like | Who pays it |
|---|---|---|
| Rework hours | The deliverable ships wrong and gets rebuilt from scratch | You, or whoever inherits the fix |
| Clarification round-trips | Three back-and-forths to extract what should’ve been in the brief | Both people, and the calendar |
| Early churn | A hire quits in month two because every question felt like bothering you | You, in a rehire and a restart |
| Silent AI drift | An agent executes confidently on the wrong assumption for a week before anyone notices | You, discovering it late |
| No real vacation | Nothing runs without you answering Slack from the beach | You, permanently |
None of those show up on a P&L labeled “handoff tax.” They show up as a slower quarter, a hire who didn’t work out, a launch that took three extra weeks. Measure your own rework hours and your own redo rate for a month. That number will tell you more than any statistic I could hand you.
Does writing detailed instructions guarantee the AI agent gets it right?
No, and this is the trap that catches people once they’ve learned the first lesson. A written brief can create false confidence that lets you check out of the loop entirely. An AI agent has no tacit judgment to catch a step failing in real time. It will execute a stale or wrong plan with total confidence, and it won’t flag that the situation has moved past what you wrote.
The brief is the floor, not the whole building. It gets you out of guessing what someone needs. It doesn’t get you out of checking in while the work is happening, especially with an agent that can’t tell you it’s confused. If you write the brief and then disappear for two weeks, you’ve just moved the guessing game later and made it more expensive.
Same gap, two recipients
Hand a vague task to a new hire and they fill the gap with a guess, usually the wrong one, usually an expensive one. Hand a vague prompt to an AI agent and it does the same thing faster, with more confidence in the wrong answer. That’s not a failure on their part. It’s a gap you left for them to fill.
Hand either one a brief with the goal, the steps, the why, and the named decision points, and both execute correctly. Not because either one got smarter. Because you gave them what they needed instead of making them guess.
If your team includes AI agents alongside people, and most growing companies do now whether they’ve said so out loud or not, this is the one discipline that makes both halves work. It’s the same skill for managing AI and managing people, run every time instead of only when there happens to be time for it. That’s the infrastructure problem underneath what looks like a hiring problem or a prompting problem: no repeatable system for forcing the plan out of your head and onto the page before you hand it to anyone.
What to do instead
Run this before your next handoff, human or AI:
- State the goal in one sentence. Not the task. What does done actually look like, and how would someone else know they got there.
- State the why. Why does this matter more than the other three things on your list today. What breaks if it’s wrong.
- List the steps as a starting point, not a script. They’re a map for someone unfamiliar with the terrain, not a rulebook that has to survive contact with reality.
- Name the decision points, specifically. Not “let me know if you have questions.” Try “if the client pushes back on price, stop and ask me” or “if the API call fails twice, don’t retry a third time, flag it.”
- Set a check-in, not a check-out. The brief doesn’t replace oversight. It buys you the right kind of oversight, checking whether the direction is still correct instead of re-explaining the task from scratch.
If you can’t fill in all five in one sitting, that’s the signal. Don’t send the handoff yet. You don’t have a plan to hand off. You have a feeling about a plan, and someone else is about to pay for the difference.
Frequently Asked Questions
Why does delegation fail even when I explain the task clearly?
Because “clearly” usually covers the steps, not the goal or the reasoning behind them. The person executing has no way to make a good call once the situation diverges from the plan, which happens almost immediately. Fix the goal-and-why gap, not the step list, and most handoff failures stop repeating.
What should I include when delegating a task to a new hire or an AI agent?
Four things, in order: the goal (what “done” looks like), the why (the reasoning behind the task), the steps (a starting point, not a script), and named decision points specific enough to actually trigger action, like “if the client pushes back on price, stop and ask me,” not a generic “let me know if you have questions.”
Is delegating to an AI agent different from delegating to an employee?
Structurally, no. Both fail from the same cause: unstated intent that gets filled in with a guess. The failure mode is different. A person self-censors and doesn’t ask the clarifying question. An AI agent doesn’t have anything to suppress, so it just drifts on inferred intent and never throws an error while doing it.
How much does bad delegation actually cost a small business?
There’s no reliable universal figure for this in the research, and any specific percentage or dollar claim you see should be treated skeptically. Measure your own rework hours, clarification round-trips, and redo rate over a month instead. That’s the real number, and it’s specific to your business, not an industry average.
Why don’t new hires ask questions when they’re confused about a task?
Because asking reads as admitting incompetence, so people default to a plausible guess instead of flagging the gap. This is a documented psychological pattern, not laziness. Build a scheduled check-in cadence instead of relying on an open-door policy nobody wants to walk through.
Next time you’re about to hand something off, run the vacation test on it before you send it. If you had to hire someone to cover this specific task for two months starting today, what would you actually need to tell them? If you can’t answer that in one sitting, the brief isn’t ready, and neither is the handoff.
Take the five steps above and run them on the next task you’re about to delegate, to a person or to an agent. Ten minutes, before you hit send.
Want a second set of eyes on the ones that keep going wrong? Book a 15-minute call and walk me through one handoff that came back sideways. I’ll tell you, specifically, what was missing from the brief.
References
- Tubre, T. C., & Collins, J. M. (2000). “Jackson and Schuler (1985) Revisited: A Meta-Analysis of the Relationships Between Role Ambiguity, Role Conflict, and Job Performance.” Journal of Management, 26(1). https://doi.org/10.1177/014920630002600104
- Rasul, I., Rogger, D., & Williams, M. J. (2021). “Management, Organizational Performance, and Task Clarity: Evidence from Ghana’s Civil Service.” Journal of Public Administration Research and Theory, 31(2). https://academic.oup.com/jpart/article/31/2/259/5974047
- Edmondson, A. (1999). “Psychological Safety and Learning Behavior in Work Teams.” Administrative Science Quarterly, 44(2). https://web.mit.edu/curhan/www/docs/Articles/15341_Readings/Group_Performance/Edmondson%20Psychological%20safety.pdf
- Gray, M. (2024). “The Art of Delegation.” YouTube. https://www.youtube.com/watch?v=OMXlGtLr2M8
- Martell, D. (2017). “How To Create a Business Playbook: How To Make SOPs.” YouTube. https://www.youtube.com/watch?v=cfsH5rjcaW8
David Chung is an Operating Partner and the founder of KyberFive, where he builds the AI and people infrastructure that lets SMB founders hand off real work without losing the plan in translation. He hosts the “Coach as Entrepreneur” podcast and works directly with founders to install the systems, human and AI, that make delegation actually stick.