> ## Content Index
> Fetch the complete content index at: https://madewithlove.com/blog/llms.txt
> Use this file to discover other available public pages before exploring further.

# Hacking the human psyche with AI
- URL: https://madewithlove.com/blog/hacking-the-human-psyche-with-ai/
- Published: 2026-11-19T14:50:00.000Z
- Updated: 2026-11-19T14:50:00.000Z
- Description: Meeting transcripts can show what stakeholders actually asked for and how they prefer to hear things. Used well, that turns refinement arguments into fifteen-minute decisions. Used carelessly, it becomes a psychological dossier nobody consented to.
- Author: Brenden Cambier
- Tags: AI, Claude, Communication, Leadership, Processes, CTO

Ahmed recently told me something that made me feel rather good about myself.

> "Your coaching sessions changed how I present to stakeholders."

For a brief, glorious moment, I thought I might actually be good at this coaching thing.

We'd discussed how to present to stakeholders. Completely separately, we'd explored using AI to analyse meeting transcripts, extract requirements and challenge assumptions.

Ahmed combined the two. Why stop at *what* people said? Why not analyse *how* they communicated, and use that to approach them differently?

The apprentice had become the master.

(I'm fairly sure this is how the Sith started.)

Right. I've apparently sparked corporate psychological warfare. Entirely by accident.

Nobody starts their villain origin story with "So we were aligning on feature requirements."

And yet…

## We need a dashboard. Apparently.

We've all been in [those meetings](https://madewithlove.com/blog/why-involving-engineers-in-product-meetings-saves-time-and-money/). A stakeholder wants "more visibility". Someone writes down "dashboard". Engineering starts discussing charts and filters.

Three sprints later: six filters, PDF export, AI-generated insights. Because apparently a bar chart can no longer be trusted to exist on its own.

We proudly demonstrate it. Turns out they just wanted an email when an order gets stuck.

An email.

We built a small observability platform around an email.

The problem isn't that people are stupid. Well, not always. **The problem is that we interpret conversations through our own understanding of the world.**

Meeting transcripts give AI a chance to challenge that. What did they actually ask for? Which assumptions did we quietly turn into requirements? What never got answered?

Remember that requirement we discussed for eleven minutes before someone said "we'll figure that out later"?

AI remembers. (Unfortunately.)

Three weeks later, someone says "I thought we agreed on this."

We didn't.

We agreed to stop talking about it.

## Humans don't ship API documentation

But we took it one step further. Some stakeholders want the conclusion immediately. Others need the reasoning, alternatives and risks before making a decision.

Present a 35-slide architecture deck to the first person and you've essentially taken them hostage. Give the second person a three-line summary and they'll spend 45 minutes asking questions you could have answered on slide four.

Wouldn't it be wonderful if people came with an OpenAPI specification?

`GET /stakeholders/{id}/communication-preferences`

Returns preferred level of detail, decision-making criteria, tolerance for bullshit.

Unfortunately, humans are considerably worse at documenting their interfaces than software engineers. And that's saying something.

Meeting transcripts contain clues. The questions people repeatedly ask. The details they challenge. How they respond when something is unclear. AI can connect those dots and help you prepare accordingly.

Same facts. Same recommendation. Different approach.

It's not about telling people what they want to hear. It's about explaining things in a way they can actually use.

A revolutionary concept in stakeholder management, apparently.

Of course, coming up with ideas during coaching is one thing. Actually making them work is another. And I kept asking annoying questions. Are we observing behaviour or inventing a personality? Just because AI can tell us something about someone, should we know it?

Fortunately, Ahmed did the difficult part.

## The part where it actually worked

Ahmed here. The one who survived the meetings.

For the record, I didn't cleverly combine two coaching conversations. I was lazy. I asked my [second brain](https://madewithlove.com/blog/your-second-brain-knows-more-than-you-do/) a question I should have asked myself: how do I talk to this person?

It had an answer. With sources.

Every conversation comes with a hidden tax. Part of everyone's brain is busy working out how to talk to you: how much detail, which words, whether this is a stupid question. **Talk the way someone already thinks, with their analogies and their words, and that tax goes away.** They stop editing themselves and start throwing half-baked ideas at you. The half-baked ideas are the good part.

### How my second brain works

My calls get recorded and transcribed. The recorder sits in the call where everyone can see it.

After a call I run `/sync-meetings`, a Claude Code skill. It extracts decisions, open questions and action items. Then it goes after the people. Transcripts spell names however they like, so the skill matches names by sound first. Otherwise I'd get a new colleague called Tommy every week.

Every person gets a note that grows with each meeting. Three parts: **Personality** is how they decide and what they react to. **Communication style** is how they think and express themselves. **Preferences and tone** is what lands with them and what loses them.

Below that are tags: personality, communication, preferences, hobbies, likes, dislikes and unnoticed habits. Each needs evidence from a meeting. A habit seen once gets marked (1x). Without evidence, nothing gets tagged.

For example:

`Unnoticed habits: self-deprecating-opener`

`self-deprecating-opener: "I probably think it's a stupid thing"`

Then there's MBTI. I know.

Before trusting it, I sent a few subagents off to research all sixteen types. They read the Myers-Briggs Company's material, practitioner sites and what people say about themselves on forums.

The first section of their findings explains why MBTI is scientifically shaky. People retake the test and get a different type. Most of us score somewhere in the middle.

So my research note doesn't fully trust my research note. And neither do I. MBTI is one input. Most of the weight sits on what people actually did in our past interactions.

Everything ends up in two sections: **How to Talk to Them** and **How to Message Them**. Ten rules each. Five from research, five from actual interactions, clearly labelled.

And the skill doesn't get to be confident early. Until I've had enough calls with someone, including at least one 1-1, it only collects observations. No type, no rules. Every new meeting challenges its conclusions.

Like any responsible engineer, the first thing I do in the morning is open a terminal, type `claude` and say hi.

My second brain pulls today's calls and prepares a note for each one: agenda, open items, who's attending and how to approach them.

Enough setup. Here's what it looked like with real people.

### The domain expert who called their questions stupid

On a previous project I worked closely with a domain expert on our data, who found my bugs for me and then apologised for it. They were sharp and funny, and they opened almost every question with "this is probably a stupid thing, but".

Their concerns were mostly genuine. To me, though, "stupid question" meant one thing: start from zero. So I'd explain the whole system from the ground up, and only then ask whether their question still stood. It usually did. They'd just sat politely through my lecture first.

My second brain first looked at eight meetings with them. In most conversations they were chatty and open. On anything product or technical they got shy. They only shared a problem once they could reproduce it, and even then they started by apologising, worried they might say something wrong. Anything they couldn't reproduce, they kept to themselves.

That's how a real search bug stayed with them for over a year. And their intuition was mostly right.

So the apology was only a warm-up. **The bigger problem was the bugs they never mentioned.**

It suggested skipping the lecture and asking what they think is going on. Asking them directly what's felt off lately, even if they can't reproduce it. And with anything new, adding one line on how they can check it in the tool they already use.

The lecture disappeared, so our calls went straight to their actual problem. And every time a feature touched their domain, I sent it over with a way to test it in their own tool. They started testing harder and with more confidence, and their hunches reached me while they were still small, long before they could turn into a year-old bug.

Their note also keeps a small habit: they say "y'all" where most of us say "guys". So I use inclusive language with them. They explain everything through their own tools and what they show on screen, so I explain things that way too. When I talk in their terms, their energy goes into the problem.

Now when they say "stupid question", I hear "I've found something".

### The founder who hated waiting and loved cutting

Picture a founder who thinks AI should have finished the whole roadmap by Tuesday.

On an earlier project, a translation step for the data team had grown fuzzy. Half of it was translating values. The other half was a vague plan to standardise those values across datasets, and nobody had specified it.

I wanted that second half out. So I did what engineers do: I pulled the ticket back into refinement and said it needed more work.

He disagreed. To him, "back to refinement" meant one thing: later.

My second brain looked at months of his meetings. He pushed back whenever something sounded like waiting, and he was regularly amazed at how long work took. But he was also the fastest scope-cutter in any room. He questioned new features, trimmed ideas down to their simplest version and called out overthinking.

So doing less was an easy sell with him. Waiting was the hard one, and I'd led with waiting.

It suggested leading with the simplification instead of the delay. The next time I was in a call with him, I tried it.

He agreed. The ticket shipped on his timeline, with only the part I trusted in it.

**Same cut, two different pitches. The first one started an argument and the second one ended it.**

Now when I need less from him, I lead with the simpler scope instead of framing it as a wait.

### The founder who wanted to weigh it himself

This founder would rather read sixty examples than hear one good pitch.

A small ticket of mine could either fix one product or quietly change how sixty product names get built. The sixty needed his blessing. The engineer in me wanted to pitch the generic rule. It's cleaner, it's the right fix, and I could explain why in five minutes.

My second brain looked at our meetings and the page he wrote about himself on the project wiki. He came from academia. He asks for specific cases before accepting a general rule, wants the gains and losses of each option visible, and takes time on complex decisions. When he agrees, he adds a caveat and names who else should weigh in.

So a pitch was the worst thing I could bring him. He likes to reach the conclusion himself, from the evidence, and his first yes is only a working hypothesis.

It suggested letting the cases do the talking and keeping my ask small enough for him to own on the call.

So I shared a sheet with all sixty products, before and after, and asked for one thing that day: fix the one product. He went through the cases like a reviewer and thanked me for being transparent.

I came in asking for the small fix. **He argued for the big one himself, as a one-time investment worth making.** Fifteen minutes, and nobody built a dashboard.

Now with him, I bring the evidence, ask for the smallest decision, and let him get to the big one on his own.

That's the skill. I don't guess how to talk to people anymore. I read their note first.

Brenden, here's the one-sentence version: it works.

Brenden again.

Great. It works. That's exactly what worries me.

## Okay, but isn't this a little creepy?

Yes. Potentially very.

There's a rather thin line between understanding someone better and figuring out how to manipulate them. And AI is exceptionally good at making that line look like a formatting detail.

The unsettling part is just how eerily accurate these analyses can be.

I've seen AI analyse my own writing and identify habits I didn't even know I had. How I structure arguments. Where I place punchlines. Apparently, I also use parentheses to sneak in observations too uncomfortable for the main sentence.

I found that deeply offensive.

(Mostly because it was accurate.)

In one of our early experiments, AI went further than intended. It identified someone's self-deprecating humour, their tendency to downplay questions, and even inferred feminist beliefs from the way they referred to software and people.

The frightening part was how plausible it sounded.

Personal insights nobody asked for. Now available in Markdown.

But wait. We were trying to communicate feature requirements more effectively. Not build psychological dossiers.

What exactly was I supposed to do with that information? Adjust my architecture presentation based on someone's alleged feminism?

Brilliant. Apparently, PostgreSQL now needs to pass a personality compatibility test.

So we changed the prompts. Inferred personal beliefs? Out. Speculating about motivations? Out. Presenting psychological labels as facts? Absolutely not.

Now, Ahmed still experiments with personality types. And yes, I gave him a hard time about it. But there's a difference between reducing someone's personality to four letters and using them as a starting point for navigating professional communication.

Four letters aren't the source code to a human being.

(Unfortunately. Imagine how much easier stakeholder management would be.)

And this is where Ahmed deserves credit. He backs classifications with actual observations, demands evidence and revisits assumptions as more conversations happen.

The model gives him somewhere to start. The actual interactions tell him whether he's right.

And if you can't explain how an insight helps you communicate better, why the hell are you collecting it?

Even when an analysis is accurate, there's a difference between recognising a pattern and understanding why it exists. Someone might consistently ask about risks. That doesn't make them risk-averse. Maybe the production database went down yesterday.

A meeting transcript is not a psychological evaluation. We should be analysing observable communication patterns, not constructing secret psychological dossiers.

And obviously, we need permission, transparency and responsible handling of the recordings and profiles. Don't secretly record colleagues and upload their conversations to whichever AI model offers free tokens this week.

Please.

But the biggest question isn't whether we *can* analyse people this way. It's what we intend to do with that information.

Are we helping someone understand a decision, or figuring out which psychological buttons to press until they agree with us?

One is good communication.

The other is manipulation with a subscription fee.

## The biggest shift wasn't understanding them

When Ahmed told me those coaching sessions had changed how he presented to stakeholders, I thought about the AI techniques we'd discussed. But that's not really what changed.

Instead of asking, "How do I explain my solution?", he started asking, "What does this person need to understand, and what am I still missing?"

His second brain didn't just reveal patterns in others. It exposed his own. When he lectured instead of listening. When he pitched solutions instead of presenting evidence. That's a much better way to use AI.

We spend ridiculous amounts of money on tools that help us communicate. Meeting software. Transcripts. AI assistants. And we're still astonishingly bad at listening.

AI didn't suddenly give us the ability to read minds. It just made it a little harder to ignore what people have been telling us all along.

Apparently, listening was the hack.