Magic TeamsHow to Spot the Right Tasks to Delegate to AI
To spot the right tasks to delegate to AI, score every task on three axes: how often it happens, how much time it eats, and how much human judgment it needs. AI takes the high-frequency, high-time, low-judgment work. Humans keep everything that needs judgment, relationships, or a neck on the line. Magic Teams AI installs an AI Operating System (AIOS) around your whole business in a one-week intensive, and that sorting rule is the exact first move we make: audit your week, score each task, and hand the machine only the work it does faster and never forgets. Get the sorting wrong and you either automate something that needed a human, or you keep grinding on work a bot should have swallowed months ago.
Here’s the scene. A founder shows me their calendar and says, “I want AI to handle sales.” Then we look at what “sales” actually is: 40 minutes a day chasing status, 20 minutes formatting the same recap, one genuinely hard pricing call a week. The chasing and formatting are perfect for AI. The pricing call is exactly where a human belongs.
Delegation to AI isn’t a vibe. It’s a scoring problem. This post gives you the rule, the quadrant, a worked audit, and the line between what a machine should own and what you should never hand it.
What tasks should you delegate to AI?
Delegate the tasks that repeat often, cost real time, and need almost no judgment: recurring reports, data entry, scheduling, status chasing, first-draft writing, meeting notes, and routine follow-ups. Keep the tasks that need judgment, taste, empathy, or accountability. That’s the whole rule, and it’s backed by how much of the average week is actually this kind of repeatable work.
The volume of delegatable work is larger than most founders guess. McKinsey’s November 2025 research found that today’s technology could automate about 57% of current U.S. work hours, with AI agents alone able to perform tasks that occupy 44% of those hours and robots the other 13%. That’s not a 2030 forecast. That’s demonstrated capability now.
But the same research draws the line you need. McKinsey found that fewer than 5% of occupations can be fully automated, while about 60% of jobs have at least 30% of their activities that could be. So delegation to AI is a task problem, not a job problem. You slice jobs into tasks and hand over the slices that qualify.
For a founder, the math is personal. Growing-business owners spend more than a third of their week, 36%, on administrative tasks like logging expenses, scheduling, invoicing, and data entry. Most of that 36% scores as delegatable on the rule above. It’s not only founders drowning in it. A Smartsheet survey found over 40% of workers spend at least a quarter of their week on manual, repetitive tasks, and roughly 60% say they could save six or more hours weekly if that work were automated.
Here’s the shape of a typical founder week once you sort it.
The big blocks on the left are your delegation candidates. The two on the right stay human. The trick is knowing exactly where the line falls, which is what the scoring rule does.
How do you decide which tasks to delegate to AI?
Score each task on frequency, time, and judgment. Frequency is how many times a week it happens. Time is minutes per instance, rounded up. Judgment is a 1-to-5 rating of how much human decision it needs. Multiply frequency by time to get weekly drain, then sort by judgment ascending. High-drain, low-judgment tasks go to AI first.
I call this the FTJ score, and it’s the same three-axis logic behind choosing what to automate first. Frequency times Time gives you the prize. Judgment tells you whether a machine can safely claim it.
Here’s the rule written out:
- Frequency: How many times per week? Daily standups, weekly reports, and every-new-client steps score highest.
- Time: Minutes per instance, including the context-switch to start. Round up. People chronically undercount this.
- Judgment (1-5): How much human decision? Sending the same reminder is a 1. Pricing a bespoke proposal is a 5.
The distinctive move is the cutoff. Any task scoring judgment 1 or 2 with a weekly drain over 30 minutes is an AI task. Judgment 4 or 5 stays human no matter how much time it costs. Judgment 3 is the review zone: AI drafts, a human approves.
The trap almost every founder falls into: they want to delegate the interesting task, not the expensive one. The flashy work feels worth automating. The boring recurring work is where the hours actually hide. Here’s the same logic as a map. Time drain across the bottom, judgment up the side. The bottom-right box is your delegation gold.
When I run FTJ scoring live with a founder, they always lobby to hand AI the hardest, most strategic-feeling task first. Every single time, the real winner is something they were almost embarrassed to name out loud: copying numbers from a dashboard into a slide, or re-typing form entries into the CRM. That boring task is usually their single most expensive recurring hour.
Which tasks should AI do and which should humans keep?
AI should own high-frequency, low-judgment, pattern-based work: transcribing, summarizing, data entry, drafting, sorting, scheduling, and rules-based routing. Humans should keep judgment calls, relationship moments, creative direction, ethical decisions, and anything where being wrong is expensive and hard to reverse. The dividing line is judgment and consequence, not raw intelligence.
There’s research behind why some tasks resist delegation. The EPOCH framework, from MIT Sloan researchers Isabella Loaiza and Roberto Rigobón, names five human capability groups AI struggles to replicate: Empathy, Presence, Opinion and judgment, Creativity, and Hope. Tasks that lean on those stay human. Tasks outside them are fair game.
The same MIT Sloan work points where delegation should land. It finds AI is more likely to complement human workers than replace them, taking over routine components so people focus on higher-judgment work. Delegation done right splits the task; it doesn’t hand over the role. Speed and consistency belong to the machine. Judgment and relationships belong to the person.
Now the concrete list founders ask for most.
| Delegate to AI (judgment 1-2) | Keep with a human (judgment 4-5) |
|---|---|
| Weekly and monthly reports | Pricing a custom or high-stakes deal |
| Data entry and CRM updates | Firing or hard performance conversations |
| Meeting notes and summaries | Creative direction and brand voice calls |
| Scheduling and calendar coordination | Handling an upset key client |
| Status chasing and reminders | Strategy and resource-allocation decisions |
| First drafts of routine emails and docs | Legal, ethical, and compliance judgment |
| Routing and triaging inbound requests | Hiring decisions and final approvals |
| Invoice generation and reconciliation prep | Negotiations and partnership terms |
The right column isn’t “AI can’t touch these.” It’s “AI can assist, but a human owns the decision.” A machine can draft the pricing email. A human decides the price.
Why does the judgment score matter so much?
Judgment matters because delegating a high-judgment task to AI is where automation blows up. Machines are fast and consistent, but on tasks that need nuance or facts they can’t verify, they produce confident, wrong output at scale. The judgment score is your guardrail against handing the machine a decision it will get wrong quietly.
The failure data is blunt. On high-stakes legal queries, general AI models hallucinated between 58% and 82% of the time, and even specialized legal tools still produced errors in 17% to 34% of cases. Delegate an unsupervised legal judgment to AI and you’re rolling those dice on every answer.
Automation history says the same thing from the other direction. Industry practitioners have seen 30% to 50% of initial robotic process automation projects fail outright, and one leading cause is choosing the wrong process: one that’s too complex, too variable, or too dependent on human judgment. Bad task selection, not bad technology, kills most automation. In fact the same analysis pins 54% of automation disruptions on poor management versus only 3% on technical issues. The tool rarely fails. The task choice does.
That’s the instinct behind delegating without losing control: you hand over the doing, not the deciding. It’s also why judgment-3 tasks get a human approving the draft rather than a full handoff.
The day I stopped asking what AI could do and started asking what it should decide, my whole automation plan snapped into focus.
What does an AI delegation audit look like in practice?
A delegation audit is one afternoon: list every recurring task in your week, score each on frequency, time, and judgment, calculate weekly drain, then sort. The output is a ranked queue where the top rows are your first AI handoffs and the bottom rows stay human or get dropped. You can run it on your own calendar today.
Here’s a slice of a founder’s week, scored with FTJ. Weekly drain is frequency times time.
| Task | Freq/wk | Min each | Weekly drain | Judgment | Verdict |
|---|---|---|---|---|---|
| Monday status report | 1 | 50 | 50 min | 1 | Delegate to AI |
| CRM updates after calls | 12 | 6 | 72 min | 2 | Delegate to AI |
| Meeting notes and recaps | 8 | 12 | 96 min | 2 | Delegate to AI |
| Chasing team for updates | 15 | 4 | 60 min | 1 | Delegate to AI |
| First-draft proposals | 3 | 25 | 75 min | 3 | AI drafts, you approve |
| Pricing the hard deals | 2 | 30 | 60 min | 5 | Keep with a human |
| Firing a vendor | 0.2 | 40 | 8 min | 5 | Keep with a human |
Total delegatable drain in that slice: 278 minutes a week. Nearly five hours, gone to work no human needed to do. That tracks with the field data, where 20.5% of weekly generative-AI users report saving four or more hours a week, rising to 33.5% among daily users.
Notice the two 60-minute rows: chasing team updates and pricing hard deals cost the same time. The first is a judgment-1 delegation. The second is a judgment-5 human task. Same drain, opposite verdict. That’s the whole point of scoring judgment separately from time.
The most common mistake I see in a first audit is founders scoring their own judgment too high on tasks that are really just habit. “Only I can write the client update” almost always means “I’ve never written down how I write the client update.” Once we document the pattern, the judgment score drops from a 4 to a 2, and the task becomes delegatable overnight.
Run the audit and you get a ranked queue. Here’s the order the work should leave your plate in.
How does this tie to your Task Automation percentage?
Your Task Automation percentage is the share of your recurring, delegatable tasks that AI now owns end to end. The FTJ audit gives you the denominator, all the low-judgment recurring work, and every task you hand off raises the numerator. It’s the single number that tells you how much of your week you’ve actually clawed back.
Track it like a scoreboard. If your audit surfaces 40 recurring low-judgment tasks and AI owns 12, you’re at 30% Task Automation. The goal isn’t 100%. The goal is every judgment-1 and judgment-2 task delegated, which usually lands a founder between 60% and 80%, with the balance being review-zone and human-only work.
Why this metric and not “hours saved”? Because hours saved is easy to fudge and hard to sustain. Task Automation percentage is a durable count of what runs without you. It maps directly to whether you can step out of the day-to-day.
Here’s the delegation loop that keeps that number climbing.
Each turn of the loop, tasks that scored a 3 last quarter often drop to a 2 once documented, becoming newly delegatable. The percentage keeps climbing without you re-running strategy from scratch. This is why deciding whether to automate or hire gets clearer over time: recurring rules-based work keeps flowing to the system, and only genuine judgment work justifies a new salary.
Key takeaways
- Score, don’t guess. Rate every task on frequency, time, and judgment. Frequency times time is the prize; judgment is the guardrail.
- The cutoff is simple. Judgment 1-2 with over 30 minutes weekly drain goes to AI. Judgment 3 gets AI-drafts-human-approves. Judgment 4-5 stays human.
- It’s a task problem, not a job problem. Fewer than 5% of jobs fully automate, but 60% have at least 30% of activities that can. Slice jobs into tasks.
- Judgment protects you. AI hallucinates on 58-82% of high-stakes legal queries; never delegate high-consequence decisions unsupervised.
- Wrong task selection kills automation, not weak tech. 30-50% of RPA projects fail outright, mostly from picking the wrong process.
- Track Task Automation percentage, the share of recurring low-judgment tasks AI owns. Aim for 60-80%, not 100%.
Frequently asked questions
What tasks are best to delegate to AI first?
Start with recurring, low-judgment, time-heavy tasks: weekly reports, data entry, meeting notes, scheduling, status chasing, and routine follow-ups. These score highest on frequency times time and lowest on judgment, so they pay back fastest. The Monday report and CRM updates are almost always in a founder’s top three. For the full ordering, see our guide to what to automate first.
What tasks should never be delegated to AI?
Never fully delegate tasks that score judgment 4 or 5: pricing hard deals, firing decisions, handling upset key clients, legal and ethical calls, creative direction, and final approvals. AI can assist by drafting or summarizing, but a human must own the decision. On high-stakes queries, general AI models still hallucinate the majority of the time, so consequence-heavy judgment stays human.
How do I know if a task needs human judgment?
Ask three questions: Does being wrong cost real money or trust? Does it need context that isn’t written down anywhere? Does it involve a relationship or an emotion? If you answer yes to any, score judgment 4 or 5 and keep it human. If it’s the same steps every time with a clear right answer, it’s a 1 or 2 and delegatable.
What percentage of my work can AI actually handle?
For most founders, roughly a third of the week is delegatable admin and coordination. McKinsey found today’s technology could automate about 57% of U.S. work hours, with AI agents covering 44%. In practice, aim to delegate every judgment-1 and judgment-2 task, which usually gets a founder to 60-80% Task Automation on their recurring work.
How is delegating to AI different from delegating to a person?
You delegate to a person and hope they build judgment over time. You delegate to AI and it never builds judgment, but it never gets tired, distracted, or inconsistent either. So you hand AI the parts that need consistency and speed, and hand people the parts that need judgment and growth. The scoring uses the same three axes; the cutoff is stricter for AI on judgment.
What if a task is somewhere in the middle on judgment?
Judgment-3 tasks go to the review zone: AI produces the draft, a human approves before it ships. First-draft proposals, client updates, and routine analysis usually live here. Over time, as you document the pattern, many judgment-3 tasks drop to a 2 and become fully delegatable. That documentation step is often the fastest win, covered in documenting processes without spending weeks.
Can I delegate a whole role to AI, or just tasks?
Just tasks, at least for now. Fewer than 5% of jobs can be fully automated, because nearly every role mixes low-judgment and high-judgment work. The right move is to slice the role into tasks, delegate the qualifying ones, and let the human focus on the judgment work that’s left. That’s augmentation, and MIT Sloan finds it’s the more likely outcome than replacement.
How long does it take to see time back after delegating tasks?
Fast, if you start at the top of the audit. Weekly generative-AI users report saving four or more hours a week, and daily users save even more. Because you delegate the highest-drain tasks first, the biggest chunk of that comes back in the first week or two, not after months of setup.
What’s the biggest mistake founders make when delegating to AI?
Delegating the interesting task instead of the expensive one, and scoring their own judgment too high on tasks that are really just habit. The flashy strategic task feels worth automating; the boring recurring one is where the hours actually hide. And “only I can do this” usually means “I’ve never written down how I do this.” Document it and the judgment score often drops a point or two.
How do I measure whether AI delegation is working?
Track your Task Automation percentage: the share of recurring low-judgment tasks AI now owns end to end. Set a target of 60-80%, re-audit quarterly, and watch review-zone tasks graduate to full delegation as you document them. Pair it with hours-back-per-week for a sanity check, but the percentage is the durable number.
Should I automate these tasks myself or have someone install it?
You can absolutely start solo: run the FTJ audit and delegate a few obvious tasks with off-the-shelf tools. The limit shows up when tasks span multiple tools that don’t talk to each other, which is where most DIY automation stalls and where 30-50% of automation projects fail. That’s the gap an AI Operating System closes, connecting context and action so delegated tasks run as one system instead of a pile of disconnected bots.
Does delegating tasks to AI mean I’ll need fewer people?
Usually it means your existing people do more of the work that actually needs a human, and you delay or avoid a hire you’d otherwise make for admin. Since recurring rules-based work flows to the system, you hire for judgment, relationships, and growth instead. Our automate-or-hire framework walks the cost math.
If you run the FTJ audit and find five hours of judgment-1 work hiding in your week, that’s not a spreadsheet exercise. That’s the shape of your next quarter, and it’s usually the moment a founder wants to talk about what installing the whole layer would actually look like.