Most business owners I talk to use AI the same way: as a very fast typist, rather than using AI as a thinking partner to work through difficult decisions. Write this email. Tidy up that proposal. Summarise these notes. It works, it saves twenty minutes a day, and then it plateaus. That plateau is where almost everyone stops.
That second approach is a different job entirely, and it is the one that actually changes how you run the business. Not because the AI is smarter than you. Because it will sit there at 9 pm and interrogate a decision you have been circling for three weeks, without getting bored, defensive, or worried about what you think of it.
This post covers what that shift looks like in practice, the five jobs where thinking beats drafting, why your AI keeps giving you generic advice, and how to run a session that is worth the hour.
Key Takeaways
- AI as a thinking partner means using it to pressure-test decisions, not to produce copy. The output is a better decision, not a document.
- Drafting and reviewing save you minutes. Thinking work saves you weeks, because it stops you from committing to the wrong thing.
- The five highest-value thinking jobs: pressure-testing decisions, pre-mortems, deciding what not to do, untangling messy problems, and preparing for hard conversations.
- Generic AI advice is a context problem, not a model problem. An AI that knows nothing about your business gives you advice for a business that does not exist.
- Voice input changes the quality of the session. You speak at roughly 150 words a minute and type at 40, and you think differently when you talk.
- A thinking partner that never disagrees with you is worthless. Ask it directly for the strongest case against your position.
- The session is only valuable if the conclusion gets written down somewhere the business can act on. Otherwise it is an expensive chat.
The Ceiling You Hit When AI Is Just a Typist
Here is the pattern I see over and over. A founder starts using ChatGPT or Claude. First month, genuinely useful. Client emails get faster. Job descriptions write themselves. The follow-up sequence that had been sitting in the too-hard basket for a year gets done in an afternoon.
Then it stops mattering. The tool becomes another tab. It gets opened when there is something to write and ignored the rest of the time. Nothing has changed structurally. The business still runs entirely on one person’s judgment, and that person is still the bottleneck.
That ceiling is not the AI’s fault. It is a function of the job you gave it. Drafting and reviewing are output tasks. They come at the end of a process, after the hard part is already done. By the time you are writing the email, you have already decided what the email says. The AI is compressing the cheapest part of the work.
The expensive part happens earlier, and it is almost entirely invisible. It is the two weeks you spent unsure whether to hire, the three versions of the pricing model you built and binned, the client relationship you knew was going wrong but could not name why. That is where your time actually goes. Not in typing.
I ran a rough count on my own week once. Actual writing, maybe five hours. Circling decisions, second-guessing calls I had already made, re-litigating the same three questions in the shower, in the car, at 11 pm: far more than five hours. And none of it produced anything I could point at.
Why Founders Never Make the Jump
Three reasons, in my experience.
The first is that nobody told them it was an option. Every AI tutorial on the internet is about prompts that produce artefacts. Write a LinkedIn post. Generate a landing page. The genre is full of output.
The second is that thinking work feels like it needs a human. There is an assumption that a machine cannot help you decide whether to fire your operations manager. It can, and it is often better at it than the mate you would normally ask, because it has no stake in the answer and no relationship to protect.
The third is that the first attempt goes badly. They ask a broad strategic question, get four paragraphs of consultant-flavoured mush, and conclude the tool is not up to it. That failure is real, but the diagnosis is wrong. It is nearly always a context problem, which I get to below.
What AI as a Thinking Partner Actually Means
Strip the label back, and it is simple. You bring a problem you have not resolved. The AI’s job is not to answer it. Its job is to make your thinking better by asking things you have avoided, naming assumptions you did not know you were making, and arguing the side you have been quietly dismissing.
A good session looks less like a search engine and more like a session with a sharp advisor who has read every file in your business and has no interest in flattering you.
The difference from drafting is worth being precise about. When you ask AI to draft, you are the source of truth and it is executing. When you use AI as a thinking partner, neither of you is the source of truth. You are both circling something you cannot see clearly yet, and the value comes out of the friction.
That friction is the whole point. If you finish a session and the AI agreed with everything you said, you wasted an hour. You did not get a thinking partner. You got a mirror with a vocabulary.
The Bar It Has to Clear
For this to be worth doing, the AI has to beat the alternatives you already have. Those alternatives are: thinking about it alone in the car, asking your partner who is sick of hearing about the business, asking a staff member who cannot be honest with you because you sign their pay, or paying a consultant who needs three sessions before they understand what you do.
Against alone-in-the-car, AI wins on structure. It forces you to articulate the problem in full sentences, which is half the work.
Against your partner or your team, it wins on honesty. There is no relationship cost to it telling you that your plan depends on an assumption you have never tested.
Against a consultant, it wins on availability and iteration. You can run the same problem eight different ways at 6 am on a Sunday. Where the consultant wins is accountability and real-world pattern matching, which I will come back to at the end because it matters.
Five Jobs Where AI as a Thinking Partner Earns Its Keep
These are the five I use most. They are ordered by how much time they save, not by how clever they sound.
1. Pressure-Testing a Decision You Have Already Made
You know what you are going to do. You are looking for reasons not to. This is the single highest-value use and almost nobody does it deliberately.
The prompt is not “should I do this?” It is: here is what I have decided, here is my reasoning, give me the three strongest arguments against it and tell me which assumption is doing the most load-bearing work.
Two outcomes, both good. Either the decision survives, and you commit properly instead of dithering for another fortnight, or it does not survive, and you just saved yourself a very expensive mistake. I have killed two product ideas this way. Both would have cost me a quarter.
2. The Pre-Mortem
Different from pressure-testing. Here you assume the decision is made and it has already failed, twelve months from now. The AI’s job is to write the story of how it went wrong.
This is much more productive than a risk list, because a risk list is abstract and a story is specific. “Cash flow risk” tells you nothing. “You hired in March, the two big clients you were counting on both delayed their projects to Q3, and by June you were paying a salary out of your own account” tells you exactly what to watch and exactly what to line up first.
3. Deciding What Not To Do
Most founders I meet do not have an ideas problem. They have a plates problem. Fourteen things on the go, four of them mattering, and no honest way to work out which four.
Give the AI the full list, the actual numbers behind each one, and the constraint you are working under. Then ask it to cut the list to three and defend the cuts. Then argue with it. The argument is where the value sits, because the moment you feel yourself defending something, you find out whether the reason is real or whether it is sunk cost and ego.
4. Untangling a Problem You Cannot Name
Some problems are not questions yet. Something in the business is wrong. Delivery feels harder than it should. The team is quieter than usual. Margins are drifting, and you cannot point to why.
For this one, do not try to be organised. Talk for ten minutes, badly, out loud, and ask the AI to reflect back what it heard: the patterns, the things you said twice, the thing you skirted around. It is very good at spotting the thing you mentioned once in passing and then never returned to. That thing is usually the actual problem.
5. Preparing for a Conversation You Are Dreading
The pay conversation. The client who is about to be let go. The business partner discussion you have postponed twice.
Have the AI play the other person. Give it their position, their incentives, and what they are likely to be feeling. Then run the conversation. It is uncomfortable, and it works, because you get to make your worst arguments in a room where nobody is harmed by them.
Why Your AI Keeps Giving You Generic Advice
This is the objection I hear more than any other. “I tried asking it strategy questions, and it gave me rubbish.” Fair. It probably did. Here is why.
An AI with no context about your business is answering for a business that does not exist. It does not know that your revenue is lumpy because two clients are 40% of it. It does not know your operations manager is six months in and still finding their feet. It does not know you tried outbound in 2024 and it went nowhere for reasons that had nothing to do with the offer.
Without that, it does the only thing it can, which is give you the average answer for the average business. The average answer is useless to a specific business.
The fix is not a better prompt. It is a permanent context layer: a set of files the AI reads before every conversation, holding who you are, what you sell, how you make money, who is on the team, what you have already tried, and what you are currently trying to achieve. It is the same briefing you would give a new executive on day one, except this one never forgets it and never leaves.
The change in output quality is not subtle. The difference between “consider improving your lead qualification process” and “your last 30 enquiries show two thirds came from referral, your paid leads converted at less than half that rate, so the qualification issue you are worried about is really a source issue” is the difference between advice and analysis.
Context Compounds, Prompts Do Not
A clever prompt helps once. A context layer helps every time, forever, and gets better as you add to it. This is the same reason a good employee gets more valuable in year three than they were in month one. They have accumulated context.
Once the context is in place, the thinking partner stops being a novelty and starts being the first place you go. That is the actual threshold. It is not “can AI think.” It is “does this AI know enough about my business for its thinking to be worth anything?”
I wrote more about how the underlying setup works in agentic workflows, and about the wider pattern of AI sticking or not sticking in small businesses in AI adoption for SMEs.
How to Run a Thinking Session That Is Worth the Hour
Mechanics matter here more than people expect. Four things make the difference between a session that changes something and a session that produces a nice-sounding summary you never read again.
Talk, Do Not Type
You speak at around 150 words per minute and type at 40. That is the obvious part. The less obvious part is that you think differently when you talk. Typing makes you edit as you go, which means you unconsciously tidy the problem before you have described it. Talking gets the messy version out, and the messy version is the true one.
Use voice input. Ramble. Do not organise it first. Half of what is useful in a session comes from something you said sideways while trying to get to the actual point.
Ask It To Interrogate You First
Do not open with your question. Open with: Here is the situation; ask me ten questions before you say anything.
The questions are the value. A good set will surface at least two things you had not considered and one thing you knew but had not admitted. This mirrors what a good consultant does in their first hour, which is not to advise, but to ask.
Force Disagreement
Models are agreeable by default. Left alone, they will polish your reasoning rather than test it. So instruct against it, explicitly and every time: argue the opposite case, tell me where I am wrong, assume I am too close to this to see it.
My standard line is some version of “do not be diplomatic, tell me what you actually think is weak here.” It works. The output changes character immediately.
Land It Somewhere Real
The session ends, and you feel clear. Three days later, gone. This is the most common failure, and it is entirely avoidable.
Finish every session by having the AI write the decision, the reasoning, and the next action into a file, a task, or a message to the person who has to execute it. If the thinking does not turn into an artefact the business can act on, it did not happen.
This is where thinking work and doing work meet. A conclusion is only worth as much as the action that follows it. One of my clients, a finance broker, sat on a dormant database for months knowing there was money in it. The thinking was done long before. What changed things was a system that actually worked the list: 319 contacts everyone had written off, and $49,000 recovered.
Where a Thinking Partner Still Loses to a Human
I would rather be straight about the limits than oversell this.
It has no skin in the game. It will not follow up in a fortnight to ask whether you actually did the thing. Accountability is the single biggest advantage a human advisor, a coach, or a peer group holds, and it is not a small one. Most founders do not fail from bad analysis. They fail from not doing what they already knew they should do.
It has no local pattern matching from lived experience. It has read a great deal, but it has not sat across the table from forty Auckland business owners in the last two years and watched which of them made it. There is a kind of judgment that only comes from having been in the room, and it is not available here.
It cannot read the person in front of you. When the real question is whether your operations manager is struggling or checked out, someone who has met them will beat any model.
And it will confidently be wrong. Less often than it used to be, but often enough that you check anything that matters. Research from McKinsey’s ongoing State of AI work keeps landing on the same point: the organisations getting real value are the ones that redesigned how work happens around AI rather than bolting it onto existing habits. The failure mode is almost never the model. It is the process wrapped around it.
So the honest version is this. A thinking partner is not a replacement for an advisor, a board, or a peer group. It is the thing that fills the ninety per cent of the time when none of those people is available, and it makes the ten per cent when they are far more productive, because you turn up to those conversations with your thinking already done.
The Bigger Shift
Once you have used AI properly as a thinking partner, the drafting use case starts to look like what it always was: a small convenience at the very end of a long process.
But there is a further step, and it is the one that actually changes the shape of a business. AI as a thinking partner is one half of the picture. The other half is a workforce that acts on what you decide, without you having to be the one who does it. An AI brain that thinks with you, and an AI workforce that handles the follow-up, the responses, the reporting, the chasing, and the hundred small recurring things that currently sit in your head.
That combination is what takes a founder from being the system to running on one. The thinking partner is where most people start, because it is the piece you can feel the day you switch it on. It is not the destination.
If you are already using AI to write things and you have hit the plateau I described at the top, the next move is not a better prompt library. It is giving it enough context about your business that its thinking is worth listening to.
Try This Next
If any of this landed, the useful next step is a conversation rather than another article.
I run a free 30-minute Discovery Call where we look at what is actually eating your week and where a thinking partner plus an AI workforce would change the shape of it. No pitch deck, no obligation. If the answer is that you do not need this yet, I will tell you that.
Book a 30-minute Discovery Call
Worth reading alongside this: the 90-day AI automation checkpoint for what to measure once you start, and admin overload in small business if the drafting plateau is really an admin problem in disguise.
Frequently Asked Questions
What does it mean to use AI as a thinking partner?
It means using AI to improve a decision rather than to produce a document. You bring an unresolved problem, the AI asks questions, names your assumptions, and argues the case against your position. The output is a clearer decision and a written conclusion, not a draft email or a blog post. The value sits in the friction, not the fluency.
Is AI actually any good at strategic thinking?
It is good at structuring, questioning, and finding holes in reasoning, which is most of what strategic thinking involves in practice. It is weak at judgment that depends on lived experience or on reading a specific person. Given real context about your business, it consistently beats thinking alone. Given no context, it produces generic advice for a business that does not exist.
How is this different from using ChatGPT to write emails?
Writing an email is an output task that happens after the thinking is done, so it saves you minutes. Thinking work happens before, and it saves you weeks by stopping you from committing to the wrong thing. The prompts are different too. Drafting asks for a result. Thinking asks the AI to interrogate you, disagree with you, and argue the opposite case.
Why does AI give me such generic business advice?
Because it does not know your business. Without context on your revenue mix, your team, your constraints and what you have already tried, the only thing it can do is give the average answer for an average business. The fix is a permanent context layer the AI reads before every conversation, rather than a cleverer prompt typed each time.
How do I stop AI from just agreeing with me?
Instruct against it directly and in every session. Ask for the three strongest arguments against your position, tell it to assume you are too close to see the problem, and ask what a sceptic would say. A line like “do not be diplomatic, tell me what is weak here” changes the output immediately. If it agreed with everything, the session was wasted.
Can AI replace a business coach or advisor?
No, and it is worth being clear about that. A coach holds you accountable, follows up, and brings pattern matching from years of sitting across from real operators. AI has none of that. What it does is cover the time when no advisor is available, and it makes paid advice go further, because you arrive with your thinking already worked through.
How long should a thinking session take?
Twenty to forty minutes for most decisions. Talk rather than type; spend the first ten minutes describing the situation badly and in full; let it ask questions before it offers anything; then argue. Finish by having it write the decision, the reasoning and the next action somewhere the business can act on. Without that last step, it will not survive the week.
About Octavius
Titus Mulquiney is the founder of Octavius AI, where he builds AI brains and AI workforces for founder-led businesses stuck running everything out of their own head. Twenty years in marketing, ex-Sony product manager, ex-GM Zeal NZ. Based in Auckland, working with operators across NZ, Australia, and the US. Connect on LinkedIn.