TL;DR: Adopting AI at work without breaking your business comes down to four moves. Use AI to amplify the boring, repetitive work that drains your people, not to replace the judgment they get paid for. Identify the edge cases before you deploy. Keep a human in the loop wherever a mistake costs money or trust. Retrain your existing people on the new tools instead of firing them. More on how I use AI without the crap. The companies that follow this pattern win the next decade.
The ones that confuse “AI can do this task” with “AI can replace this job” learn the hard way, publicly and expensively. Klarna and the chatbot at your doctor’s office are the same disease at different scales.
I called my doctor’s office the other day to leave a message. The new AI chatbot picked up and launched into a two-minute uninterruptible speech about how it was going to help me. When that finally ended, it started a one-minute one.
Then more after that. Five minutes every call, every time, before I could get to a human or even leave a message.
The old phone tree was clunky. This was worse. And somebody at that medical practice spent real money to make it worse. That clinic is the pattern. Companies are rolling out AI everywhere, and a lot of those rollouts are hurting the businesses paying for them.
The fintech giant Klarna said its OpenAI-powered chatbot was doing the work of 700 customer service agents, then started rehiring humans when the CEO publicly admitted the AI produced “lower quality” service. Same disease, different scale.
Neither failure started with the software. Nobody asked the boring question first: where does this belong in our work, and where does it ruin things if we put it there? I blame the people who signed off on those rollouts without asking it, and I’d blame them long before I’d blame the tool.
I’ve been around long enough to watch this happen before. I spent 33 years in enterprise IT, and the pattern never changed: buy the tool, skip the dull work of cleaning up what it runs on, and blame the vendor when it fails. AI is getting the same treatment, and it makes me furious, because the people who pay for it are the staff and customers stuck with the result.
Does AI fix a bad process?
The single most useful thing I can tell you about AI at work is that it doesn’t fix anything. It amplifies. Whatever you feed it, good or bad, comes back louder. Give it clean data and a clear question and it gives you something useful fast. Give it a mess and it gives you a bigger mess, delivered with total confidence.
This is why so many AI projects fail and why the AI takes the blame for the failure. The model followed its instructions exactly. It took the garbage it received and produced more garbage, faster than any human could. The trouble sat upstream the whole time, in the data, the process, or the thinking, and the AI made it visible at scale. I don’t have much sympathy for the executive who blames the model after that. The mess was theirs before the AI ever touched it.
Why Does AI Amplify Bad Data?
Most companies have worse data than they think.
The single most useful thing I can tell you about AI at work is that it doesn’t fix anything.
Records that contradict each other. Fields filled in three different ways by three different departments. Spreadsheets that a person can read with context but a machine reads literally. Point an AI at that and it doesn’t flag the mess. It averages it, guesses at it, and hands you an answer that looks authoritative and is wrong.
A human doing the same task would stop and say this doesn’t add up. The AI doesn’t stop. It fills the gap with the most likely-sounding text and moves on. That confidence is the dangerous part. People check a wrong answer that sounds unsure. They ship a wrong answer that sounds certain.
Think about who’s on the receiving end. The system tells a customer the wrong balance. It flags an employee for something they didn’t do. A patient gets the wrong callback. None of them can see the bad record behind the answer, and none of them will accept that the AI was only repeating what it was given.
Vague Process In, Automated Confusion Out
The second source of crap is process nobody can explain. Ask most teams to write down exactly how they make a decision and they cannot. They do it by feel, by experience, by a hundred small judgments they’ve never put into words.
That works when a person does it. It falls apart the moment you try to hand it to a machine.
If you can’t describe the process clearly, the AI can’t follow it. It’ll pick a version, usually the simplest one, and automate that. Now you have a fast, consistent, wrong process running at volume.
The workflow was never really defined, and the AI just proved it in public. I’d much prefer a company find that out on a whiteboard, before its customers find it out for them.
Vague Questions In, Vague Answers Out
From the podcast
Diana Lee teaches a four-part structure for exactly this, and describes the missing piece as context you’d give any advisor, on Leaders and Their Stories:
The variables piece is giving it the context that it needs to know about you, as you would if you were to work with an actual consultant or an advisor.Nobody expects a consultant to produce good work from a one-line brief. The machine gets held to a lower standard, and returns work to match.
The third source is the question itself. People ask AI broad, lazy questions and get broad, lazy answers, then conclude the AI isn’t smart enough. Ask it to make our marketing better and you get generic filler. Ask it a sharp, specific question with the real constraints included and the quality jumps.
This is a skill, and most people haven’t built it yet. The quality of what you put in caps the quality of what you get out, the same rule that governs every tool ever made. AI is a tool, and I tell anyone starting out to treat it like one and learn to use it well. Hand it your thinking and expect it to do your work, and you’ll have problems.
My take
Garbage out of an AI tool starts as garbage you put in. Dirty records, a process nobody can write down, and a lazy one line question will sink any tool. The machine won’t stop and say something doesn’t add up, so it fills the gap and sounds sure of itself. Fix the data, put the process on a whiteboard, and ask sharp questions before you automate anything. If the output stinks, look at what you fed it before you blame the machine.
What Should You Fix Before Adopting AI?
Before you blame the model or shop for a different one, look at the three inputs. Is the data clean and consistent, or are you feeding it contradictions? Can you write down the process in plain steps, or does it live in someone’s head? Are you asking precise questions, or vague ones and hoping? Fix those and most of what people call an AI problem disappears.
| Input | What goes wrong | What to check |
|---|---|---|
| Data | AI averages contradictory records and hands back a confident wrong answer | Are records clean and consistent, with fields filled in one way? |
| Process | AI picks the simplest version and automates a fast, consistent, wrong process | Can you write the process down in plain steps? |
| Question | Broad, lazy questions get broad, lazy answers | Are you asking a sharp question with the real constraints included? |
I’ve watched companies spend months and real money trying to make an AI project work when the actual fix was a week of cleaning up their data and defining what they wanted. That waste is inexcusable, because it lands on the people stuck using the broken system while leadership shops for a better model. Clean up the inputs before you spend another dime on the tool, and hold whoever skipped that step accountable for the mess.
Bad inputs are one way to waste AI. The Birth of the Augmented Human covers the other side, how the tools amplify thinking instead of replacing it when people use them properly.
