The moment an AI starts acting in my name, its mistakes become my reputation.
I’ve been thinking about that as AI moves beyond drafts and starts making decisions for us. Sending an email is different when nobody checks it first. So is recommending something to a customer or deciding what another person gets to see.
A year ago, I was more interested in whether AI sounded like AI. There were obvious giveaways everywhere. Certain words appeared too often and sentences had the same polished rhythm. Entire paragraphs could say surprisingly little while sounding as though something important just happened.
That became the first version of AI slop for me. You could spot it from the language alone. People kept banned-word lists and told models to stop saying things like ‘delve’ and ‘tapestry’.
The models learnt quickly. With decent instructions, many of those giveaways disappear. You can remove every obvious AI phrase and still end up with something bland or detached from the decision it’s supposed to help you make.
I started asking if I agreed with the thinking behind the words, even when they sounded human. Once the language became harder to catch, the judgement underneath stood out more. 👇🏻
New here? Millennial Masters shares lessons from founders who’ve built, grown and led companies. Subscribe for ideas on AI, growth, leadership and the decisions that rarely come with a playbook.
Slop starts before the draft
You can ask AI to draft a proposal and get something that looks ready to send. AI can summarise a customer problem or tell you what to do next in seconds, and you stop noticing how many decisions are hiding behind the request.
You still need to understand what the customer cares about and what you don’t know yet. What’s missing may be exactly what makes the recommendation wrong.
You used to have to stop and make more of those decisions yourself. When the result arrives in seconds, they’re easier to skip. Getting the task off your list tells you very little about whether you made a better decision.
The manager never disappeared
If somebody joined your company tomorrow, you wouldn’t throw a vague task at them and assume they understood how you think. You’d explain the job and give them context, then correct the work until they knew the standard.
A meeting summary is fairly forgiving and a follow-up email is easy to review. Choosing who gets a discount or how to handle an unhappy client is much harder to leave to AI.
With AI, you still have to decide what it’s allowed to do and when it needs you. You’re still the manager, except now AI can make decisions without waiting for you.
Mistakes get more expensive
For a long time, bad AI mostly meant clean-up. You’d generate something weak, spot the problem and fix it. Maybe you wasted twenty minutes, or an employee spent an afternoon rewriting work that was supposed to save time.
With agents, a mistake can reach someone before you see it. A sales agent can follow up with dozens of leads while you’re doing something else. Poor judgement can leave dozens of people with the same bad impression of your company.
In customer support, an agent can answer the question correctly and handle the person badly. The customer may have needed an apology. A reply can follow policy perfectly and make a £20 problem feel like a reason to leave.
I now notice another kind of AI slop when weak thinking gets repeated automatically. A perfectly normal sentence can hide a decision that’s wrong for the company. At that point, whether the AI sounds convincing matters far less to me. I want to know if I would have made the same call.
Exceptions are harder to hand over
You may know instinctively that a customer needs a phone call rather than another email. Experience also teaches you when a refund makes sense beyond the policy, or when an attractive sales opportunity is likely to become painful later. Much of that judgement comes from years of doing the work without writing the reasoning down.
A long-standing customer paying you £20,000 a year may need a different response from somebody who signed up yesterday and ignored three attempts to help. A missed payment from a normally reliable client means something different from a new customer who has broken every term since day one. Those distinctions rarely fit neatly into one rule.
You can see the same thing in how you already manage people. The best employees learn the exceptions by watching you approve one refund and refuse another, or turn down a large client because the work would wreck the team. Over time, they understand more than the written process.
An AI agent doesn’t get that history by sitting in the office for six months. Trying to teach that judgement to an agent exposes all the places where your business relies on decisions nobody has properly explained.
Your judgement can’t stay in your head
You have to spell out which decisions the agent can make on its own and when it needs to stop and ask you.
Customers often give you half the information, and an apparently perfect sales lead may turn out unable to afford you. Something that saves one team two hours can create six hours of clean-up somewhere else. That’s where you find out whether the rule actually works outside the clean demo.
When the judgement is sound, AI can apply it far more widely than you could on your own, whether that means answering a customer at midnight or helping your team when the usual person isn’t available.
The customer only sees what happened
If an automated recommendation pushes somebody towards something they don’t need, that decision comes from your company and becomes part of what they think about you.
AI still stands out because it feels new. I expect it to disappear into the background like the CRM your salesperson uses or the software that calculates an invoice.
Every decision you hand over to AI becomes another way somebody experiences your company.
I care more about the call now
I now ask if I’d have made the recommendation myself and whether I could explain it if a customer challenged me.
If the agent makes the same call a hundred times this month, I want to be comfortable with the pattern it creates.
I’ll happily automate decisions where little is at stake and the standard is clear. Anything with too much history behind it or too much at stake for the customer still needs someone involved.
The moment an AI starts acting in my name, its mistakes become my reputation.
Such an important article. Did you read about the Korean Starbucks campaign that used AI for a marketing campaign and it got approved without executive oversight and the outcome was embarrassing and damaging. They used a slogan that no one realized had direct meaning to a horrific event in Korean history. 🤷♀️a nightmare for that team. Cautionary tale!
It doesn't matter how much thought you put into the decision. If the AI makes a bad call, the customer just sees your company making a bad call.