What if You Could Just Do Your Work?
Most people can instantly name the work they’d hand off if someone gave them the chance. What’s interesting is how the line is usually drawn in the same place — and what that tells you about how AI agents fit into the way your team works.
If you could get rid of something at work tomorrow — just like that, gone, no fine print, no consequences, — what would it be?
You work in a project-based business. Maybe an agency, maybe a consultancy. You sell expertise — code, design, strategy, communication, legal advice, or social media ads, solving someone else’s problem using your very specific set of skills. The actual work, in other words. The reason the business exists.
So what would you hand off? Most people know immediately. Not the work itself — the other work. The scheduling, the status updates, the resource juggling, the timesheets, the project admin that has to happen but doesn’t have to happen by you.
For most, the line is pretty clear, though: there’s work worth protecting, and work you’d delegate in a heartbeat if you trusted it would actually get done right.
What People Are Willing to Let Go of
This is the exact question we asked our customers. What would you delegate to an AI agent — and what would you keep?
The answers showed an interesting pattern.
Three in four agency professionals said they’d trust an AI agent to generate project estimates from historical data. Nearly two out of three were comfortable with delegating time tracking and automatic scheduling based on team availability.
All the tasks that ranked high on the “take it, please” list have one thing in common: they run on data that already exists inside your business. Your project history, your team’s calendar, your budgets. The work is real and it matters, but it doesn’t necessarily require your judgement. It requires accuracy and consistency, which is, incidentally, exactly what AI agents are good at.
And the tasks people wanted to keep? Anything touching a client relationship. Communication, relationship-building, the work that carries your name and your reputation. Fewer than one in five would hand that to an agent. And that instinct is right.
Why the Line Falls Where It Does
People aren’t anti-AI. They’re selectively pro-AI, and the selection makes sense.
Operational tasks — estimates, scheduling, time tracking — are analytical by nature. The output is either accurate or it isn’t. AI excels here not despite being a machine, but because of it. It doesn’t get tired, doesn’t forget, doesn’t estimate optimistically because it wants to win the project. It works from data and returns a result. That’s exactly what the task requires.
Client-facing work is a different category entirely. Not because AI can’t generate a coherent email or a professional update — it can — but because these tasks don’t just require a correct output. They require nuance, context, and the kind of reading between the lines that comes from knowing someone. The difference between what a client said and what they meant. The tone that works for this particular person on this particular day. That’s not something you retrieve from a dataset.
And then there’s the thing called effort heuristic. Even when AI-generated communication is good, sending it says something. It says this interaction wasn’t worth my actual time. Clients are also starting to recognize generated content, and when they do, the issue isn’t quality. It’s what the shortcut signals about how much you value the relationship.
So the real line is not capability, but an understanding, mostly intuitive, that some work is analytical and some work is human — and that confusing the two doesn’t just produce worse output. It costs you something harder to get back.
What Agents Are, and What They Are Not
A lot of the anxiety around AI agents comes from not quite knowing what they are. Some people imagine them as something dangerously powerful and out of control. Others picture a slightly smarter chatbot, something you have to prompt to get anything out of. Both miss the point, though.
An AI agent isn’t a decision-maker running loose in your business. The “agent” part just means it can take action — not just generate text, but actually do things. It’s closer to a very capable, completely tireless colleague who does exactly what you tell them, and (unlike many colleagues), you only need to tell them once.
You give the agent clear instructions and assign them to a specific area like time tracking, project management, or resourcing. It then runs off and takes care of the repetitive work in that area on its own while you do something else. You decide how much autonomy agents have, and every action they take is visible to you.
Agents can log your time, update the project, open and assign tasks, flag bottlenecks… They can do what you choose within the boundaries you set. What they can’t do is the work that requires knowing your client, reading a room, making a call that depends on experience and context no dataset contains. That work stays with you, and that’s the whole point.
What Gets Freed Up
When the right things get delegated, something shifts. The people who were spending Tuesday afternoons reconciling timesheets or chasing project updates are spending Tuesday afternoons on the work they were actually hired to do.
That’s not a small thing. In most agencies, the gap between the work people are capable of and the work they actually get to do on any given day is significant. And when other work is handled by something that never gets bored, never forgets, and doesn’t have better things to do — you are finally free to do your work.
The offer from the beginning of the story is no longer a hypothetical one. Some things you hand over. Some things you protect. The agencies that figure out which is which — and build their operations around that distinction — are the ones that will look back in a few years and wonder why it took everyone else so long.