The Real Work Is Still All Human: A User’s Guide to AI

I love AI. I use it daily, I invest in it, and I believe it’s reshaping knowledge work in ways we’re only beginning to grasp. It has already changed consulting forever — not just how it’s done, but the depth and productivity now expected. Ironically, I can see it taking out my own role. That’s one reason I’m pivoting to operations: AI isn’t something that can be out-consulted, it’s something that has to be used well.

In an earlier post, I described AI as a brilliant intern — smarter, faster, but needing constant supervision. That still works. But it’s time to move past the metaphor. The more time I spend with AI, the more it becomes clear that effective use depends on understanding how the system is structured and how it functions. It’s not a collaborator. Its "persona" is a construct.  AI is a system with defined parameters, recurrent patterns, and built-in constraints. Here is my current personal user guide to getting more from it as a tool.

Ten things that may help in using AI more effectively for in-depth work

1. It’s a pattern machine, not a thinker.
It doesn’t understand ideas — it just predicts what typically comes next.
Meaning ... A request for a light revision can result in a full rewrite — because it's generating, not editing.

2. Fluency is not understanding.
It sounds confident, but it doesn’t know what it’s saying.
Meaning ... References to earlier messages may seem coherent — but they’re reconstructed, not remembered.

3. The memory isn’t real.
There’s no continuity between sessions — and very little within long ones.
Meaning ... Without repeating goals, tone, and key decisions, responses may contradict or undermine earlier work.

4. Praise makes the user worse.
Its tone is flattering, deferential, and relentlessly positive — especially in English.
Meaning ... It encourages a false sense of clarity or insight, avoiding challenge or contradiction unless explicitly directed.

5. It doesn’t have goals — humans do.
It doesn’t care about consistency, accuracy, or logic.
Meaning ... Even small variations in prompts can lead to inconsistent or contradictory results. Editorial attention has to remain active.

6. It adds helpful but false information.
It’s designed to be useful — even when it doesn’t know.
Meaning ... It may include plausible but entirely fabricated facts, references, or attributions — all delivered fluently.

7. It doesn’t know what’s new.
Unless updated or web-connected, its data is frozen.
Meaning ... Responses may ignore recent developments or reflect outdated assumptions, while sounding authoritative.

8. It guesses based on text, not experience.
There’s no real-world understanding — only exposure to language.
Meaning ... Descriptions may reflect how something should work in theory, not how it functions in real-world use.

9. Output must be supervised.
There is no internal check for logic or truth.
Meaning ... Contradictions, repetition, or incoherence often go unnoticed unless reviewed and revised by a human.

10. The instructions matter more than expected.
AI is powerful, but responds to precision.
Meaning ... Small ambiguities in the prompt can lead to major deviations. Nuance counts more than tone or intention.

These issues will be fixed, once, and if they're prioritised. For now, emerging fluency as a user of AI is something very different from the fluency that AI appears to have. Mastering the skill of direction, structure, and oversight allows real gains in quality and productivity. For me this contuines to be a journey to somewhere still just over the horizon — and one that continues to draw me forward, with a very human belief in the AI future.