In partnership with
I’d love to know how you’re thinking about leveraging AI as an engineering leader! I’m considering creating a new course to dive into this specific topic. Fill out this brief survey if interested.
I'm good at delegating to people. Or at least, I've gotten better at it over the years.
I know how to hand off a project with clear context. I know how to check in without micromanaging. I know when to step back and let someone figure it out on their own.
Delegating to AI is a different skill entirely.
Why it feels different
When I delegate to an engineer, I know what they can and can't do. I know when they need help. I know when to push back if something seems off.
With AI, I'm still building that instinct.
It's confident when it's wrong. It doesn't ask clarifying questions when it should. It gives me an answer even when the right move would be to tell me it doesn't have enough context. I'm learning to spot the difference between "good enough" and "totally off," but I'm not there yet.
Last week I wrote about the risk of AI sucking the joy out of work. I still believe that's a real danger. But I've also realized I was treating this like an all-or-nothing decision. Either I hand everything over and trust it completely, or I avoid it entirely.
That's not how delegation works with people. Why would it work that way with AI?
A thank you from this week’s sponsor:
There’s more to AI than ChatGPT.
If you’re only using AI to rewrite emails, you’re doing it wrong.
The AI for Business & Finance Certificate from Columbia Business School Exec Ed breaks down how to use AI to make faster, more strategic decisions at work.
Save $300 with code SAVE300 + $200 with early enrollment by Feb. 17.
What I'm learning to delegate
I started with the stuff I'd happily hand off to a junior teammate if I had one.
Jira ticket review. Before I assign a story, I'll paste it in and ask if it has enough context. It'll call out missing acceptance criteria, unclear dependencies, edge cases I forgot. Not perfect, but it catches stuff I used to miss until someone pinged me three days later saying they were blocked.
Draft cleanup. I write fast and messy. I used to spend way too much time polishing before I hit send. Now I'll dump a draft email or doc into a tool and ask where I'm being unclear. It flags the parts where I'm hedging, repeating myself, or burying the point.
Organizing information. I keep an archive of past newsletters, strategy docs, and team goals. When I'm planning for next quarter, I can ask "what are we prioritizing?" instead of digging through 10 different Google Docs. It's not making decisions for me. It's consolidating what I already decided.
Meeting prep and accountability. Before a tricky 1:1, I'll load in context—past feedback, project notes, Slack threads—and ask it to surface patterns. I also use my meeting notes from Granola to make sure I'm following through on the things I said I'd do. I'm a pen and paper to-do list girl, but having AI gut-check my follow-through is honestly not a bad thing.
I’m not letting it do everything for me, of course.
I'm not letting it write my feedback. (It’ll help me edit for clarity.) I'm not using it to decide who gets promoted. I'm not trusting it to navigate conflict. Basically, I'm not handing it anything where being wrong would hurt someone.
The stuff I'm delegating is repetitive, time-consuming, and low-stakes if it's off. The same criteria I'd use before handing work to a junior engineer. If I wouldn't trust a new hire to do it solo, I'm not trusting AI to do it either.
The J-curve of learning to use AI
There's this thing that happens when you adopt any new tool. You slow down before you speed up.
At first, you're spending more time figuring out how to prompt it than you would've spent just doing the work yourself. You're iterating on the same request three times to get a decent output. You're second-guessing whether the tool is even helping or just adding overhead.
This is expected. You're not bad at this. You're just learning.
Building a good Claude or Cursor prompt is a skill. You're going to iterate. You're going to refine. You're going to figure out what works for you and what doesn't. Some things will stick. Some won't.
I've deleted a bunch of workflows that sounded great but turned out to be more trouble than they were worth. I've also found a few that save me genuine time. The only way to know the difference is to experiment.
Delegation rules that still apply
Turns out, most of what I know about delegating to people also applies here.
Be specific about what you need. "Clean up this doc" gets messy results. "Flag anywhere I'm being vague or repetitive" is clearer.
Give it context. Just like I wouldn't hand an engineer a project with zero background, I don't expect AI to read my mind. The more context I give, the better the output.
Check the work. I wouldn't assume a junior engineer nailed something on the first try. Same here.
Know when to take it back. If I'm spending more time correcting AI output than I would've spent doing it myself, I stop delegating that task.
Don't delegate the judgment calls. If the work requires empathy, nuance, or understanding someone's motivations, I'm keeping it.
I’m still experimenting
I'm treating AI the way I treat delegation. Figure out what's worth handing off, set clear expectations, and check the work.
Some days it saves me an hour. Some days it gives me garbage and I wish I'd just done it myself.
But the skill I'm building isn't just "how to use AI." It's "how to delegate to something that doesn't think like I do."
And that feels like a useful skill no matter where this all goes.