I watched a senior engineer spend forty-five minutes arguing with ChatGPT about the “right” way to structure a database migration. Back and forth, refining the prompt, getting frustrated that the AI “wasn’t understanding” what he wanted.
He finally gave up and wrote it himself in twenty minutes.
This is what bad delegation looks like. And we’re all doing it with AI.
Delegation Isn’t Abdication
When you delegate to a person, you don’t just throw a task over the wall and hope for the best. You consider:
- What does this person already know?
- What context do they need?
- What decisions can they make vs. what needs to come back to me?
- How will I know if they’re stuck?
- What does good look like?
But with AI, we’re skipping all of this. We’re typing “write me a function that does X” and then getting mad when it doesn’t read our minds.
The problem isn’t the AI. It’s that we haven’t figured out what we’re actually delegating.
Treat AI Like A Junior Engineer
Here’s where the metaphor breaks down: junior engineers learn. They build context over time. They can ask clarifying questions when they’re confused. They internalize your preferences and get better at predicting what you want.
AI either doesn’t do those things or needs to be explicitly told to do them; the ability to reason along with prompting, like CoVe and CoK, and other techniques to avoid hallucinations1 are fine-tuning an LLM’s ability to answer (or not). It inherently has no memory of what frustrated you yesterday. It won’t learn that you prefer explicit error handling or that you hate comments stating the obvious.
This is actually freeing once you accept it. You’re not building a relationship. You’re using a tool. A very sophisticated, very capable tool, but a tool nonetheless.
What AI Is Actually Good At Being Delegated
Here’s what I’ve learned AI handles well:
First drafts of anything
Not final versions. Not “the answer.” First drafts. The blank page is the hard part for most of us. AI is fearless about blank pages. Let it generate the boilerplate, the structure, the obvious stuff. Then you do what you’re actually good at: editing, refining, adding the nuance.
Tedious transformations
Converting formats. Restructuring data. Generating test cases. Writing the fortieth variation of the same API endpoint. This isn’t creative work—it’s pattern matching. AI is better at this than you are, and it never gets bored.
Research synthesis
Not research itself (AI hallucinates sources like nobody’s business). But if you’ve done the research and need to synthesize it? AI can help structure your thinking. It’s like talking to a very well-read person who has no opinions and infinite patience for your half-formed ideas.
Rubber ducking with code
“Here’s my code. What am I missing?” Sometimes AI spots the thing. Sometimes it doesn’t. But the act of explaining your code to something forces you to think differently about it.
What You Should Never Fully Delegate
Decisions that affect people
Performance reviews. Team feedback. Difficult conversations. If you’re using AI to write these, you’ve fundamentally misunderstood your job. Your judgment, your relationship, your understanding of context—that’s the value. AI can help you organize your thoughts. It cannot replace your judgment.
Anything where “good enough” isn’t good enough
Customer-facing content. Architecture decisions. Security implementations. These need your expertise, your judgment, your willingness to be accountable when things go wrong. AI can assist. It cannot own.
Learning you actually need to do
If you’re a junior engineer and you let AI write all your code, you’re not learning to code—you’re learning to manage AI. Maybe that’s the future. But if you skip the fundamentals, you won’t know when AI is leading you astray. And it will.
The Actual Skill: Knowing What to Keep
The best delegators I know are the ones who are vicious editors; they are the ones that are really effective using AI. They let AI generate volume, then they cut 80% of it. They use AI to explore ten different approaches, then they synthesize the best parts themselves.
They’re not trying to avoid work. They’re using AI to get to the interesting work faster.
Here’s what this looks like in practice:
Instead of: “Write me a complete function that handles user authentication with all edge cases.”
Try: “Give me three different approaches to handling user authentication, with pros and cons of each.” Then you pick the approach and implement it, or have AI draft it and you refine it.
Instead of: “Write my team update email.”
Try: “Here are the three things I want to communicate. Help me structure this so the most important thing is clear.” Then you write the email in your voice.
Instead of: “Debug this code.”
Try: “What are five possible reasons this code might be failing?” Then you investigate the most likely ones.
You’re not delegating the work. You’re delegating the grunt work so you can focus on the judgment work.
What This Means for Your Team
If you’re a leader, you need to help your team figure this out. Because right now, they’re all flailing.
Some people are refusing to use AI at all, which means they’re spending hours on work that could take minutes. Some people are over-relying on AI, which means they’re producing high volumes of mediocre work. Neither is sustainable.
Here’s what actually helps:
Stop pretending this isn’t happening. Make it explicit. “We use AI. Here’s how we think about it.”
Share examples of good delegation vs. bad delegation. Show your team what you mean by “use AI for the first draft, not the final draft.”
Create norms around what needs human review. If someone’s using AI to generate code, what’s your standard for review? If someone’s using AI to draft documentation, what’s the bar?
Talk about what not to delegate. Be explicit about where judgment matters, where relationships matter, where your team’s expertise is the actual value.
Most importantly: Stop measuring productivity by volume. If your team can generate 10x more code with AI, that doesn’t mean they should ship 10x more features. It means they should ship better features. More thoughtful architecture. Fewer bugs. Better documentation.
The leverage isn’t in doing more. It’s in doing better.
The Uncomfortable Truth
AI is revealing something uncomfortable about how we work: a lot of what we do is grunt work. Boilerplate. Pattern matching. Stuff that feels like work but isn’t actually where the value is. The actual value—your judgment, your taste, your ability to see what matters and what doesn’t—that’s still all you.
AI is just making it really obvious where the line is. So stop trying to delegate everything to AI. And stop refusing to delegate anything to AI. Figure out what’s actually yours to do. Then delegate the rest without guilt.