AI is getting remarkably good at doing work. But that doesn’t mean you should hand everything over to it.

The more useful question is:

What should you actually delegate to AI?

Some tasks are almost perfect candidates for automation. Others are better suited to AI assistance, where the AI does much of the preparation but a human remains in control. And some tasks should probably stay firmly in human hands.

This tool gives you a quick way to figure out where a particular task sits.

Think about a real task from your work, answer the questions below, and you’ll receive an AI Delegation Score along with suggestions for what AI should potentially automate, assist with, or leave to you.

How the score works

The tool looks at seven characteristics of your task:

Frequency — Tasks that happen repeatedly tend to offer more opportunities for automation.

Repetition — A predictable process is much easier to delegate than something completely different every time.

Data access — AI needs access to the information required to perform the work. A theoretically automatable task may not be practical if its necessary context is locked away.

Human judgment — The more nuanced judgment a task requires, the more valuable human involvement becomes.

Consequences of failure — An AI making a mistake while formatting a spreadsheet is very different from an AI making a mistake in a high-stakes decision.

Human communication — Negotiation, relationship-building and situations where social context matters are generally harder to delegate completely.

Definition of success — AI works best when you can clearly explain what a good result looks like.

The resulting score isn’t a prediction of whether an AI model can technically perform your task. Instead, it estimates how suitable the task is for delegation.

Three ways to delegate

Your result falls into three broad categories.

Automate

The task contains characteristics that make substantial automation realistic.

That doesn’t necessarily mean you should immediately build a completely autonomous system. Start with the repetitive and predictable parts, then gradually increase the amount of work AI handles as you gain confidence in the results.

AI Assist

The task is a good candidate for collaboration rather than complete automation.

AI might research information, summarize documents, prepare drafts, organize data, generate alternatives or perform the first pass — while you retain control over the important decisions.

This is likely to be the most useful category for a huge amount of modern knowledge work.

Keep Human

Some tasks contain too much judgment, responsibility, uncertainty or human interaction to comfortably delegate.

That doesn’t mean AI is useless.

Even when the task itself should remain human, AI may still be useful for preparing information, brainstorming possibilities, checking work or handling small supporting steps.

Don’t think in terms of jobs. Think in terms of tasks.

One of the biggest mistakes people make when thinking about AI automation is asking whether an entire job can be replaced.

That’s usually the wrong question.

Most jobs are collections of dozens or hundreds of different tasks.

A marketing manager might spend part of the day:

• Researching competitors
• Writing reports
• Analyzing campaign data
• Answering emails
• Brainstorming campaigns
• Attending meetings
• Negotiating with partners
• Making strategic decisions

Those tasks have radically different levels of AI suitability.

The goal isn’t necessarily to replace the marketing manager.

It might be to give that person back several hours every week by delegating the parts of the job that don’t actually require a human.

The same principle applies to programmers, designers, lawyers, teachers, analysts, managers, entrepreneurs and almost anyone else doing knowledge work.

The best first candidates are often surprisingly boring

When people imagine AI automation, they tend to think about spectacular things: autonomous agents, AI employees and systems running entire businesses.

In practice, some of the best opportunities are much less exciting.

A recurring report.

A weekly summary.

Sorting incoming requests.

Turning meeting notes into action items.

Extracting information from documents.

Preparing a first draft.

Comparing large amounts of information.

Converting information from one format into another.

These tasks may not sound revolutionary, but if something takes an employee 30 minutes every day, automating it can recover more than 120 hours of work over a year.

And unlike replacing an entire role, delegating a small task is usually much easier to test safely.

Start with delegation, not automation

There is an important distinction between the two.

Automation implies that the system performs a process with little or no human intervention.

Delegation means giving AI responsibility for a piece of work while deciding how much human oversight remains.

That distinction matters.

You don’t have to go from:

Human does everything

straight to:

AI does everything.

There is a much more useful spectrum:

Human does it → AI prepares it → AI does most of it → AI does it automatically → AI manages the workflow

Different tasks can sit at completely different points on that spectrum.

And you can move them gradually.

Your score is only the beginning

A high AI Delegation Score doesn’t mean you should immediately automate the task.

Before handing over real work, consider:

• What information will the AI need?
• Can it access that information reliably?
• How will you detect mistakes?
• Who is responsible if something goes wrong?
• How expensive is a failure?
• Can the process be tested on low-risk examples first?
• Does the AI actually save time after human review is included?

That last question is particularly important.

If an AI produces a draft that takes you longer to correct than it would have taken to create yourself, you haven’t automated anything. You’ve created another task.

The real objective isn’t using AI as much as possible.

It’s removing unnecessary human effort while keeping humans in control of the things that genuinely require them.

Try it with a real task

Don’t answer the questions hypothetically.

Think of something you actually did this week that consumed your time.

It could be a report, an email workflow, research, customer support, data entry, content creation, scheduling or almost anything else.

Then see how much of that work might realistically belong in AI’s hands.

Your score isn’t a verdict. It’s a starting point for finding work you don’t necessarily need to keep doing yourself.