Ask Patrick › Guides › What to Delegate to Your AI Assistant
Free Guide March 2026 8 min read

What to delegate to your AI assistant
(and what to keep for yourself)

Most people who get an AI assistant make the same mistake in one of two directions: they hand off decisions that need a human, or they keep doing tedious work that an AI could handle in seconds. This guide gives you a clear, practical framework for drawing the line — so you stop wasting your time on the wrong things.

The core question

Before you hand anything to your AI assistant, ask yourself one thing:

"Does this task require my judgment, my relationships, or my accountability — or is it purely processing information and producing output?"

If the answer is just processing, your AI should own it. If the answer involves real judgment, real relationships, or real consequences for being wrong — that's yours to keep.

It sounds simple. But in practice, most people blur the line constantly. They let an AI draft an email to a major client without reviewing it. Or they spend 30 minutes manually compiling a weekly report that an AI could generate in 20 seconds. Both mistakes cost you.

The delegation test (4 questions)

Run any task through these four questions. If you answer "yes" to all four, delegate it. If any answer is "no," keep reading before you decide.

1

Is the task repeatable?

Does it happen more than once, roughly the same way each time? Daily summaries, weekly reports, inbox triage, social updates — all repeatable. A difficult call with an unhappy customer — not repeatable in any useful way.

2

Can you write clear instructions for it?

If you can explain what "done" looks like in plain language — your AI can do it. If the task requires instincts built from years of experience or context that can't be written down, keep it.

3

Is the downside of a mistake recoverable?

An AI that drafts the wrong report costs you two minutes to correct. An AI that sends the wrong message to a client or posts something damaging in public — that's not recoverable in the same way. Recoverable mistakes? Delegate. Irreversible mistakes? Keep human oversight.

4

Is this task time-consuming relative to its thinking requirements?

Some tasks are mostly time, not thought. Pulling data, formatting a document, summarizing a thread, writing first drafts. These are high-value delegates. Tasks that are short but require real judgment — approvals, judgment calls, personal responses — keep those.

What your AI assistant should own

Here's a practical breakdown of tasks that belong to your AI assistant — and ones that don't. Use this as a starting checklist when setting up your own setup.

✓ The pattern here

All of these are information-processing tasks. The AI pulls data, organizes it, formats it, or drafts something. You review the output, make decisions, and act. The AI does the grunt work; you do the thinking.

What you should keep for yourself

✗ The pattern here

All of these require either accountability (someone has to be responsible), real relationships (trust built over time), or judgment that goes beyond pattern-matching. These can't be fully delegated — but your AI can support you on all of them.

The in-between zone: AI assists, you decide

Some tasks aren't fully delegatable, but they're also a waste of your time to do from scratch every time. These are the tasks where your AI handles the preparation — and you handle the final call.

Think of your AI as a very capable junior team member. They can do the research, write the first draft, pull the data — but the decision and the signature are yours.

How to hand off a task properly

The most common reason AI delegation fails isn't the AI — it's unclear instructions. Here's the framework that works:

1

Write down what "done" looks like

Before you ask your AI to do anything, be specific about the output. "Summarize my inbox" is vague. "Give me a bulleted list of emails that need a reply today, each with one sentence of context" is delegatable.

2

Give examples, not just descriptions

The fastest way to get consistent output is to show one example of what good looks like. Paste in a previous report, email, or summary you liked. Say "similar to this." Your AI will match the format and tone immediately.

3

Set the boundaries

Tell your AI what it shouldn't do. "Don't send — just draft." "Don't include anything that hasn't been confirmed." "Flag anything unusual instead of guessing." Boundaries prevent the failures that erode trust.

4

Review the first five outputs yourself

Even a perfectly written instruction set will need tuning. Don't trust-but-not-verify right away. For the first week of any new delegation, read every output. You'll catch the edge cases that weren't in your original instructions.

A realistic week in the life

Here's what a well-configured AI assistant actually handles for a solo founder over the course of a typical week — and what the founder still does themselves:

AI handled this week
  • 5 morning briefings (email summary + calendar)
  • 2 first drafts of client update emails
  • Weekly revenue report
  • 12 customer FAQ replies
  • 3 social media posts drafted
  • Meeting notes from 4 calls
  • Competitor news digest (2x)
  • Invoice follow-up reminders
Founder handled this week
  • Approved the client updates before sending
  • Decided on a pricing change
  • Handled one difficult customer conversation
  • Set goals for next quarter
  • Signed a new partner agreement
  • Edited 2 posts before publishing
  • Called a long-term client

The founder spent roughly 8 hours on high-value work. The AI handled the 25+ hours of routine processing that used to fill the week.

💡 The actual goal

You're not trying to replace your judgment with AI. You're trying to protect your judgment — to save it for the decisions that actually need it, rather than burning it on tasks that are purely mechanical.

The biggest mistakes people make

Delegating too little (staying stuck)

This usually comes from a desire for control — wanting to see every email before it's even drafted, redoing work the AI already did well. The cost is that you're still doing everything yourself, just with more steps. Start with low-stakes tasks and build trust gradually.

Delegating too much, too fast

The opposite problem: handing off something high-stakes on day one without reviewing output first. An AI assistant that sends emails or posts on social without any review is a liability, not an asset. Earn trust gradually by checking the first batches of output before giving full autonomy.

Writing vague instructions

"Be helpful" isn't an instruction. "If an email looks urgent, mark it with 🚨 and put it at the top of the list" is an instruction. The more specific you are about format, tone, edge cases, and what to do when something is unclear — the better the output.

Treating the AI like a search engine

Your AI assistant works best when it has context about you, your business, and your preferences. An assistant that knows you prefer bullet points over paragraphs, that your biggest client is called TechCorp, and that you always want revenue numbers rounded to nearest dollar — that assistant is worth 10x more than one you're introducing to your workflow from scratch every day.

Get the exact setups that run on autopilot

The Library is a collection of tested, copy-paste AI assistant configurations — morning briefings, inbox management, weekly reports, and more. Every setup in here runs a real business right now.

Get Access to The Library →

$9/month · Cancel any time · New setups added weekly

Where to start today

Pick one task from the "AI owns it" list above. It should be something you do at least 3x per week and takes you more than 5 minutes each time. That's your first delegation.

Write the instructions for it. Give one example of what good output looks like. Tell your AI what it shouldn't do. Review the first five outputs yourself. Then let it run.

In two weeks, pick the next task. In a month, you'll have a genuinely useful assistant instead of a novelty.

The founders who get the most out of AI assistants aren't the ones who handed everything over immediately — they're the ones who built trust deliberately, one task at a time.

Questions? Post in the Ask Patrick Discord — or grab Library access above and post in the Workshop for a specific answer about your setup.

More free guides