There’s an idea that has been gaining traction ever since ChatGPT arrived on the scene: that AI is going to do the work for us. That all you need to do is write a prompt, hit enter, and wait for magic. But anyone who has used these tools seriously for more than six months knows it doesn’t work like that.
AI is a multiplier. And multipliers work both ways: they amplify good judgment just as they amplify bad judgment. If you don’t know what you’re asking for, or how to evaluate what you get back, AI won’t save you — it will sink you faster.
This week, a Hacker News article has been making the rounds (with 72 points and plenty of debate) that hits the nail on the head: your AI tools are only as good as your judgment, and that is exactly the point. It’s worth unpacking why.
The illusion of autonomous AI
The commercial promise of many AI products is seductive: “let AI write your emails,” “let AI close your sales,” “let AI run your business.” It sounds great until you try it in earnest.
What you quickly discover is that today’s generative AI:
- Produces with confidence even when it’s wrong. It doesn’t warn you when something is off — it presents it with the same certainty as when it’s right.
- Optimizes for what you ask, not for what you need. If you ask for a “persuasive” email, it will give you something persuasive. Whether it’s appropriate for that specific client is not its concern.
- Has no context about your business unless you provide it. It doesn’t know that this client has been frustrated for three months, or that this proposal needs legal approval before it goes out.
The result: without someone who knows what to ask, what to review, and what to discard, AI produces convincing noise. And convincing noise is worse than silence, because it pushes you to make bad decisions quickly.
What “judgment” means in practice
When we talk about judgment applied to AI, we’re not referring to something abstract. These are concrete skills:
1. Knowing which problems are worth automating
Not every process should be automated. There are tasks where manual time is the least of the problems — where the real cost lies in the decision, not the execution. Automating the execution of a bad decision only multiplies the damage.
The right question is not “can AI do this?” but “what happens if AI gets this wrong 5% of the time?” If the answer is “nothing serious,” go ahead. If the answer is “we lose an important client,” you need a different approach.
2. Designing the prompt and the context
A well-designed prompt includes:
| Element | Why it matters |
|---|---|
| Clear role | AI adjusts its tone and depth based on who it “is” |
| Business context | Without this, it gives generic internet answers |
| Examples of the desired output | It’s the fastest way to calibrate style |
| Explicit constraints | What it should NOT do is just as important as what it should |
| Success criteria | How AI (and you) will know the output is good |
This isn’t magical “prompt engineering.” It’s the same thing you’d do when delegating to a new employee: give them enough context to make good decisions.
3. Evaluating the output honestly
This is where most people fail. The temptation is to accept the first output because “it sounds good.” But sounding good and being good are two very different things.
A judgment-driven reviewer asks:
- Is this true, or does it just sound true?
- Is it appropriate for this specific context, or is it generic?
- Is there anything important missing?
- Would I send this as is with my name on it?
If the answer to the last question is “no,” the system isn’t ready.
Judgment isn’t delegated, it’s built in
Here’s the part many people don’t understand: human judgment doesn’t disappear when you implement AI. It shifts. It moves from doing the task to designing the system that does it.
Before: you spend 3 hours a day answering emails with judgment. After: you spend 30 minutes a week refining the system that answers emails with your judgment built in.
It’s a huge shift. But it requires someone — you, your team, or whoever helps you implement it — to have the judgment to begin with. AI doesn’t create it. It scales it.
That’s why so many AI projects fail in companies: software is purchased expecting it to replace thinking, instead of being implemented to amplify the thinking that already exists.
Where this fits in your business
If you run an SME or a department, this has practical implications:
Don’t look for tools, look for processes. A loose AI tool doesn’t solve anything. A well-designed process with AI built in — where it’s clear what the machine decides, what the person decides, and when things get escalated — does.
Don’t automate what you don’t understand. If you can’t describe the process step by step, including its exceptions, you’re not ready to automate it. Document first, automate second.
Keep humans at critical points. In any serious AI workflow, there are moments when a person reviews, approves, or decides. Those points aren’t friction — they’re quality control.
Measure what matters. “We saved time” is not a metric. “We reduced response time from 3 hours to 8 minutes while maintaining or improving conversion rate” is.
How we approach this at Studio SmartWork
In our day-to-day work, this idea that AI amplifies judgment is the foundation of how we operate. When a client says, “I want to automate X,” the first conversation isn’t about technology — it’s about how X is done today, who makes which decisions, where the critical points are, and what happens if something goes wrong.
Only then do we design the workflow. And always with the client’s logic embedded: their criteria, their exceptions, their priorities. We don’t install “an AI” — we build a scaled version of how their business operates.
That’s why we use tools like n8n and AI APIs instead of closed products: because every business has its own judgment, and an off-the-shelf product can’t capture it. A tailored solution can — one where every system decision reflects a decision the business owner would make.
The uncomfortable conclusion (and the opportunity)
The uncomfortable conclusion: if you were hoping AI would think for you, bad news. That’s still your job.
The opportunity: if you have judgment — about your business, your clients, your processes — AI is the biggest lever you’ve ever had to multiply it. One person with good judgment and good AI tools can now do work that once required an entire team.
The goal is not to delegate thinking to the machine. It’s to delegate execution so thinking has more room. Your strategic brain is scarce; your time spent answering repetitive emails shouldn’t be.
The question you should be asking yourself this week is not “which AI tool should I try?” It is: where is my judgment trapped doing machine work, and how do I free it?
That question is worth your time. The answer could change how your business works.