Article
AI Is Your Team Player. The Skill You Get To Master Is Knowing When To Catch The Ball.
There is a feeling business owners are starting to describe to me, and it goes something like this. "I asked the AI to do a thing, and it went and took ten steps, and I cannot tell if it actually did the right thing, or what it did along the way."
We are moving through a real shift right now. AI is no longer just answering questions in a chat window. It is booking, drafting, calculating, chasing, updating, and moving work forward across whole processes. The temptation, especially for busy business owners who have wanted more capacity for years, is to hand the whole thing over and hope for the best.
But hoping is not a strategy, and the more powerful AI gets at doing the work, the more important the human question underneath becomes. Not "how much can AI do on its own", but "where do I step back in, and why there instead of somewhere else?"
That is the skill I'd like to unpack. Because on the internet you get bombarded about AI and autonomy, but not many people really break down the play between the human and the AI. And how well you define that play, and how well you play it, decides whether AI actually gives you back a whole lot of time, or costs you time, blood, sweat, tears, and a lot of uncertainty when something has to be unwound.
What are the things that could go wrong when AI runs a business task totally on its own?
Usually AI does not really fail on the maths, or on following the steps. It is very good at following the exact steps it has been taught. But when AI is left to its own devices to run ten, fifteen, twenty steps by itself, without a human in the play watching, assessing, and okaying what the AI is doing, the human on the other end loses the ability to see the working. And without seeing the working, it gets really hard to trust the result, and to understand how it got from A to B.
Jeff from JT Homes builds houses on the Sunshine Coast.
Every project he takes on starts with a number he has to bet on, and getting that number right is not a small thing. He gets a baseline price from a costing company, and then his real job begins. He goes line by line, calls every trade he would hire on the build, and comes back with a control estimate he is willing to stake his profit on. Get it wrong by ten percent, and a two hundred thousand dollar profit becomes a two hundred thousand dollar loss. That is the size of the bet on every quote.
So when Jeff started using AI to help, he did what most of us do. He set it running. He let it work. And for several hours, he kept coming back to the same feeling. How do I know if this is correct? I don't know what it is doing. And if I mess up the costing, we could lose all the profit on this build.
That feeling is the whole problem.
It is not that the AI got the numbers wrong. It probably did not. AI is very good at maths. The problem is that Jeff was not in the play. He could not see the working. He could not watch each number get built. And in Jeff's world, you do not hand in a quote you did not watch get built. Not when there is real money on the line.
So we slowed it right down.
Now, AI is not replacing Jeff doing the work. AI is Jeff's estimate team player. It knows the end to end path of a control estimate. It tells Jeff what to do next. It tells him who to call, who to follow up with, drafts the emails to the trades, keeps the spreadsheet moving, tracks the tasks, holds the progress. Jeff talks to the suppliers. Jeff validates. Jeff checks. Jeff approves every step.
That is the play.
The AI executes. Jeff brings the senses. The intuition, the read on a trade he has known for ten years, the smell test on a number that came in too low. Neither one could do this alone. The AI cannot call Mick about the joinery and hear that he is slammed till Christmas. Jeff cannot run the project management, draft the follow-ups, and keep the sheet current while he is on site. But together, they can build a quote Jeff can hand in with a straight back.
Not "the AI did it". Not "Jeff did it". Jeff and the AI played it together.
Why does autonomous AI feel unsafe even when the numbers are right?
AI does not live in this world. So it cannot gather the context of the situation you, the human, and the business are actually in. That context is something only humans can experience. It lives inside us, gathered through our six senses: sight, sound, smell, taste, touch, and intuition. And it cannot be part of the play unless the workflow between you and AI is designed to invite it in.
The instinct most business owners have when they first try AI at this scale is to measure success by how much of a task the AI can do on its own. But full autonomy is the wrong scorecard. It puts the focus on removing humans from a workflow, rather than placing them where they actually matter.
AI is absolutely brilliant at executing and following your instructions. It follows them to the T. But oftentimes we are not able to 100% articulate the process, the steps, and every rule that sits underneath the work. And because AI does not live here in the way we do, it also misses out on a lot of the situational context, the intuition, the gut feel, and the overarching wisdom that any project, any job needs. The kind of thing only we humans have.
That context lives in you. In your senses. In the ten years of experience your team carries. In the instincts you have earned by being in this work every day. AI does not replace those senses. It plays alongside them.
So the question is not "how much can I let the AI do on its own?". The question is "where in this process am I bringing something the AI cannot have, and how do I make sure I am there for that moment?"
That is the whole game.
How should I design human check-ins when AI is doing the work?
Not every check-in is the same. There are two kinds, and they do very different jobs. A notification tells you where the AI is up to, so you can see the play without stopping it. A decision point asks for your judgement before the workflow moves forward. Most teams collapse these into one, and it costs them either time or trust.
The first kind is a notification. The AI is running along, doing its work, and it lets you know where it is. You do not stop it. You do not check it line by line. You just want to see the play as it unfolds. That is enough certainty for that stage of the work, and it costs almost nothing in time.
The second kind is a decision point. The AI has done its part, and now something needs your judgement before it moves on. This is where you step in properly. You look, you weigh it against what you know, you approve or you redirect. You are in the play.
Most teams collapse these two into one, and it costs them. They either stop the AI at every step because they are anxious, which means the AI barely saves them any time. Or they let it run through every step without any pause, which means when something is off, it has travelled three or four moves downstream before anyone spots it, and unwinding it feels like moving backwards.
Both of those extremes are draining. And neither of them is the play.
How do I find the right moment to step back in on my own AI workflow?
Pick one workflow that has more than three steps. Read through it and mark the single moment where a wrong output would be most expensive to unwind if nobody caught it. Put yourself there on purpose. Everything before and after can run untouched. That one moment is where your senses belong.
The mistake is to check every step. It burns time and teaches nothing. The other mistake is to check no steps. It saves time until something goes wrong, and then it costs everything you saved and more.
The move sits between the two. One workflow. One moment. Give that moment a proper decision point, something concrete you actually look at, not a vague intention to keep an eye on things. Then, for two weeks, keep a note of how often that check actually catches something. Not to punish yourself if the number is low. To learn where the moment really belongs.
Sometimes the riskiest moment moves as the workflow itself grows up, and the check needs to move with it. So this is not a one-off design decision. It is a practice.
One well-placed moment where you show up. Not a check on every step. Not zero checks. One moment where your senses are in the play.
Frequently asked questions
Is this the same as human-in-the-loop?
It is a version of it. Human-in-the-loop is a broader technical concept. What we are describing is the practical, business-owner version of it. Which is knowing which moment of the play is actually yours to be in, and being there deliberately rather than accidentally.
Does this mean I should not use fully autonomous agents at all?
No. There are workflows where full autonomy is genuinely fine, especially for low-stakes, high-volume tasks. The point is that the decision to make something autonomous should be a decision, not a default. Autonomy that was chosen is very different from autonomy that just happened because nobody named where the human belonged.
How do I know where the riskiest moment in my workflow is?
Ask yourself one question. If this step goes wrong and nobody catches it for a day, how expensive is it to unwind, and how many other steps has the mistake already touched by then? The step with the biggest answer is your moment.
What if my team does not have time to be in the play at every workflow?
That is exactly why you pick one moment per workflow, not five. Placing your attention well is the point. The whole reason to do this is so you spend less total time managing AI, not more.
Is this something you help businesses design?
Yes. This is a big part of what happens inside a strategic engagement with Set to Growth. We do not just teach your team how to build AI assistants. We also teach them where the human belongs in each assistant they build. Because a trained team knows both how to build, and how to stay on top of what they have built.
AI cannot replace you. AI does not live in this world. It has not spent ten years knowing Mick. It has not felt a supplier's tone shift on the phone. It has not walked a job site and known instantly that the number in the spreadsheet is wrong for this build.
But AI is genuinely brilliant at being an assistant, and a team player. With the right instructions, it can play the ball too. It can take some really awesome steps for you and make your life so much easier.
The most important thing though. You have to be able to see and understand, at all times, why it is taking the steps it is taking, and what the outcome of each step actually is. Because if you cannot fully trust it, and you do not have 100% visibility, things can go pear-shaped really quickly.
That is the whole play. AI is your team player. Your job is to know when to catch the ball.
Nina Huchthausen
Nina Huchthausen is a process engineer by trade. For 16 years she has helped businesses streamline the processes and systems, and now also handing simply repetitive tasks to AI. The result her clients feel is the manual, draining work lifting off the team's plate, and either the room to actually grow and scale, or the same revenue made in a 7 hour work day instead of 12.