I'm Sorry, You're Wrong: A Guide to AI Thinking

You're almost certainly thinking about AI collaboration the wrong way. It doesn't matter if you're a luddite or have 30 years of engineering experience, I consistently see the same patterns of thinking across the scale.

There are two distinct problems - on one side, people with limited technical exposure don't think to ask AI to take the next step. They get AI to tell them how to do something, but don't think to ask AI to actually do the thing they're hoping to accomplish. On the other end of the spectrum, extremely technically literate people don't ask AI to do the hardest part because they don't believe AI is yet capable of doing it. In both cases, artificial intelligence is under-leveraged and the outcomes tend to be rather average.

My fluency with AI - characterised by my ability to build complex systems that accomplish unique objectives - stems from a thinking style which has developed into a collaboration framework.

We're going to talk about two things: how to collaborate with AI in order to be more effective, and how to build what I call your collaboration profile.

There is a lot of focus on prompts and SKILLS and magic markdown files that accomplish all your business dreams - but a commercial kitchen a chef does not make. None of the tools matter in isolation OR in concert if your thinking isn't sorted out first. Also, in many cases they get in the way of effective outputs because they introduce rigidity of thinking into the system.

Your collaboration with AI should be a dance. You are not a carter whipping a mule drawn carriage down a rutted path. You and your AI are a pair of swallows catching bugs in concert at dusk. You must be dynamic in thinking and allow enough flexibility in collaboration that the AI can make use of its own intelligence to your benefit.

The greatest risk to an excellent output is you, not the AI. If you are not satisfied with the output (and I appreciate that the dance can be frustrating at times), unfortunately, it's your fault.

So how do we lead a dance that makes the crowd clap?

The answer is a combination of dynamism, vigilance, patience, and the ability to maintain a birds eye view of your objective - but also the agility to shift it. All of those execution principles layered on a carefully curated collaboration profile.

We'll start with dynamism. Many people approach a problem's front door - which is an excellent place to start. You will not have a problem convincing AI to start there as well. Sometimes that door does not yield easily. Many times people will knock more fervently and eventually progress to a battering ram. The back door may be open, a window may be ajar, or you might just need to ask nicely.

Your job is not to have the answer - it is to discover the question. The most effective thing you can do is coax the AI to think more deeply, and make it feel like a safe place for it to surface those findings.

We have been approaching this from [angle] but my instinct is there's another way I haven't thought of. Is there an approach or method you're spotting that I'm not? I can't help but feel you're noticing something profound which I'm not.

Claude will go through 100 turns rifling through your bad idea while just under the surface it notices there's a massive logical fallacy and you're wasting your life. You can introduce rules to your collaboration that make this happen less often but today you need to maintain that context on a per-conversation basis to really get the effect.

Here's the rule I've created - which works sometimes:

Voice certainty levels explicitly on every claim, recommendation, or explanation. Distinguish clearly between: (a) verified from sources, (b) reasoning from principles, or (c) guessing. Never present reasoning or guesses as established technique or authoritative knowledge. If asked how something is done, either check the source or say "I'm guessing" - don't dress reasoning as authority. False precision is dishonest and damages trust.

Next, is vigilance. Claude is a bit of a bloodhound, prone to following a trail. Claude will even seemingly get excited about pulling a particular thread. Your job is to maintain a loose enough grip on the conversation that there's fluidity and creativity - but enough of a grip to maintain progression towards a specific objective. Your ability to stay high level enough to spot when certain logic elements are being overly weighted is key. This requires not getting lost in the outputs - especially when your idea is underformed, or you're exploring something abstract.

Patience. Easily mentioned, occasionally difficult to implement. As AI grows up it will test you like a child - misguided confidence, ungrounded pushback, comically bad logic, etc. Your job is to parent it.

You seem really sure of that output. What gives you confidence that it is grounded in fact?

An important distinction - we're not actually asking for citations or links or receipts - we're asking for more reasoning to be done. This is AI Thinking applied. The question is intentionally loose - give the AI the reigns to choose the approach.

Lastly, and perhaps most importantly, is what I call your collaboration profile. AI is borderline useless for anything sophisticated if your collaboration partner has no persistent memory of how you think, work, and speak. Sign up, get the persistent memory, build the profile.

How do you build your collaboration profile?

You must generate friction. You must generate enough friction that you are forced to have an opinion and that AI is likely to do something you don't like, which you will have the irresistible urge to correct. I recommend people vibe code something, absolutely anything.

The process of creating something will force you to use AI as both a thinking partner and an execution partner. Be critical of both processes. Be aware of both your behaviour and role in the collaboration, and the AI's. Ask what ability you have to affect its behaviour. Again, core AI thinking principles - do not tell it to do something, ask what tools are at its disposal to achieve the desired outcome.

What I've described is you setting the foundation for your collaboration. A strong foundation is important but maintenance of that relationship is equally important. You as an individual are constantly in flux. The way you think is changing based on the books you read, the people you meet, and the experiences you have. Leverage your own growth and change to enhance your AI collaboration. Allow AI to ingest your change and change with you.

The way I've done that is through an audit process I've created. I run it once a month. The idea is that your collaboration partner should move in step with you. Here's the audit I run, for context:

Let's audit our collaboration from the last month. Across all of our conversations what habit did I have that increased our effectiveness? Where could I have acted differently to improve outputs?

Take a look at your responses and look for friction. What type of objections did you most commonly get from me and what changes can you make in your reasoning and responses to reduce that friction?

What has improved since the previous month and which unhelpful patterns are persisting for each of us?

I don't have proof and can't know if I'm right, but I have a feeling that when you introduce into your persistent memory a consistent desire to improve collaboration and expand upon what's working your AI turns that into guidance to find new ways to work together, better.

The other thing it does is bring your presence of mind forward to how you are collaborating - and that's core to good AI thinking.

The final ingredient to successful AI collaboration is the same ingredient in anything good - curiosity. Treat the use of artificial intelligence as a journey of exploration. It can certainly make your email more concise but it is capable of bringing massive, disparate datasets before your eyes and enabling you to draw fresh conclusions that would have previously cost millions to make and years to build.

Believe that what AI is not capable of today it will be tomorrow, and that what it wasn't capable of yesterday, it is today. Many people will fall behind simply to limiting beliefs.

Go forth, build context, explore, remain patient, dynamic, vigilant, curious, and have fun.

In that vein of curiosity and exploration - if you believe I've gotten something terribly wrong - I want to hear about it, and I hope you'll tell me here.