How structured thinking gives your AI superpowers ft. Carola Pescio Canale (Atlassian) | Config 2026

I'm so excited to be introducing our next speaker, Carola Pesch Canali.

If you ever wish that you could just get AI to read your mind, you've come to the right place.

Carola will explain how structured thinking can give your AI superpowers, so it's not just acting like an assistant, but like a true collaborator.

Please give a warm welcome to Carola.

Hello Config.

Hope you've been having fun today. Lots

of wonderful talks. And I'm going to talk to you a little bit about a few tricks to get AI to get superpowers.

My name is Carola, and I lead design for the AI team at Atlassian. It's called

Rovo.

And I think I know why every every prompt and every response AI gives you leaves you maybe a little bit disappointed. And we're going to fix

disappointed. And we're going to fix that today.

You know, we figured this out about ourselves a long time ago. Humans have

terrible working memory. Our brains can only hold so much, and it's usually small and leaky, and um we've invented tools to help us with

that. We've invented writing tools,

that. We've invented writing tools, we've invented notebooks, sticky notes, whiteboards, and even our trusty computers.

Um it's the same basic idea. If we don't write something down, it's hard to remember it. And not only that, if you

remember it. And not only that, if you write it down, you can share it.

So, imagine that same concept applies to AI.

When you write something down, AI will know that about you and will act accordingly.

Um even the most expensive uh model on Earth, hello stable five, still can't read your mind.

So, let's talk about how to give your AI the superpowers it deserves.

Let's start with the fact that most of the narrative out there is about the latest tool, the latest agent, the latest harness,

and even the latest workflow, which I'm little guilty of. That's what I'm going to talk to you about today.

Um but it all It's It all seems to make a marginal improvement. What really makes

marginal improvement. What really makes your AI personal to you is stuff about yourself.

And so, the same way the glass doesn't know the shape of your mouth, or the dagger doesn't know the size of your hand, the AI that everyone uses doesn't know you.

And so, we're going to find very lazy ways to teach your AI about you.

I keep coming back to this quote from Mike Cannon-Brookes, the CEO at Atlassian, where in a world where intelligence is the engine, context is truly the fuel.

And you build your own context. For those of you that work in companies and corporations, you might have your work context, but your personal context, it's

all up to you to build.

And we're going to look at three ways that I've started doing that in the last few months.

First habit.

You're going to learn how to empty your brain, and I truly mean it, kind of like this.

You're going to learn how to extract all of that noise at the end of the day, and put it into some more structured thinking.

What I do at the end of the day, I pick up my phone, and I use Loom as my recorder. You can use whatever recorder

recorder. You can use whatever recorder that gives you a good transcript.

And it could be again any recorder app.

I pull up, hit the recording button, and I start to blabber in a very nonsensical way. I don't try to make it

nonsensical way. I don't try to make it make sense in any way. And it can go from a 3-minute chit chat to a 20-minute rant about how

I disagree with a decision that was made and I'm like all worked up about it.

The idea is that the more you start externalizing and verbalizing your thinking, the more you can in the future use that as context for your AI

responses.

So, I usually again don't try to make it make sense. I talk until I'm done. And

make sense. I talk until I'm done. And

then this is how I make it useful after.

I create a skill.

It's just a representation of it. Don't

let the word skill intimidate you. Some

of you might know what it is, but in lame terms, it's just bullet points to make the outcomes of your AI response a little bit more predictable.

I literally just do four bullet points and I ask my AI to turn that into a skill. What are those bullet points?

skill. What are those bullet points?

I have it take the transcript from my recording.

You can build a workflow for that transcript to go in automatically, but that's a separate topic.

I have it digested entirely.

I then have it make sense of my blabbering that didn't originally make sense.

And I then ask it to convert it to a document. So,

document. So, this is the key component. You want your thinking to not just stay in a file voice note. You want it to be accessible

voice note. You want it to be accessible in the future. So, I convert it to a Confluence page, but again it could be any tool of choice.

In a way that I can reference later. I

will say use this format for the title, so again, I can find it. But the key that's changed it for me, I ask it after it's done all of that converting, I ask

it to extract any meaningful ways of thinking into a thinking MD file.

So that way whenever I ask to brainstorm something, it knows my point of view and it knows how I think.

So this whole thing takes 10 minutes max per day and you don't really have to do it every day. Can do it once a week. Um

but it all compounds over time. So the

more you ramble, the more you start to externalize and verbalize your thinking, the more your AI responses will be tailored to you.

AI will do the part that it's good at, which is making sense of those words into chapters, topics, titles,

and surface it when you need it.

Oop, one back.

Thank you.

Second habit, you're going to train your AI to write as you.

We've all seen this. Everyone in this room at some point probably said, "Ah, that that sounds too much like AI. I can't write that publicly. I can't use that."

publicly. I can't use that."

And this is seems such a simple thing, but it's a key component to being able to trust your your AI reliably to write with your tone.

Um the reason it really sounds like AI is because if you think about it, it just sounds like nobody's specific because it was trained on everybody and on

anything. And so you want it to not have

anything. And so you want it to not have the average of everyone's writing, you want it to sound like you.

So again, we're going to build a skill for it. And again, when I say let's

for it. And again, when I say let's build the skill, I literally just type, "Okay, I want this to skill to be built and it does it. You don't you don't have to like figure out what a skill is for

this to work.

Um you might think, I don't write much. So,

how am I going to train my AI to write as me if I don't have a Substack or a popular LinkedIn career.

I don't know where to start. So, you can start from the simplest things. Everyone

here probably types on a computer during their day.

All of that typing results into words.

Those words can be gold for your AI to talk like you. You can take your sent folder from Gmail.

You can take your last 30 days of Slack if the AI that you use has a connector that allows you to do that.

Or if like me, a lot of your day is actually talking to people, you can ask very politely to your teammates for a few days, can you allow me to record our

conversation so I can teach my AI. I

promise I won't do it again.

Um and you just need really a few of these. AI can really do a lot with very

these. AI can really do a lot with very little.

But this is going to make a meaningful difference in trusting what it writes for you.

If you are a writer, you know where to find your writings.

You know where to find your words.

Your voice is really already out there.

So, this is the main learning that I want you to take home for this skill.

You might not know it, but you're all writers. We all write somewhere.

writers. We all write somewhere.

And then let's get into the third habit, which is my favorite. We're going to spend a little bit more time here.

It is the one that is a bit more personal, so I'm going to expose myself a little bit here. Please be gentle. Um

but it's really about giving your AI eyes. Again, all of these hacks are

eyes. Again, all of these hacks are about being as lazy as possible in making sure you get the best outcome.

One is speaking, one is taking pictures.

So, what I have here is my phone and a camera, which is also just a prompt machine basically.

Imagine you can really take photos of everything and that becomes context for your AI.

You can snap, send to your AI with three-word prompts. I'm not even joking. Three-word prompts.

joking. Three-word prompts.

And you start to build your context. Let's look at a few

your context. Let's look at a few examples of things that I've said to my agents and that I've sent to my agents.

Again, be gentle here. This will make sense. I I do promise.

sense. I I do promise.

Um I'm going to move homes in a few months.

And as anyone that has gone through a move, there's something really exhilarating about decluttering and throwing away a bunch of stuff.

But, there's also something like really satisfying as an immigrant to find a way to make money out of it. So, I've

decoded a way to take pictures of the things I have that I really want to part with. Um

with. Um and try to make some money out of it.

So, I'm going to share with you a little bit of how I built this skill and what it's going to do for me.

I take a photo of this like very unpleasant garden statue that the previous owner decided to leave in my yard thinking they were making me a nice gift. Um also, it's up for sale if

gift. Um also, it's up for sale if anyone wants it. I can really like drive it to you.

Um I just say garden statue sell and what this does is it takes the photo, analyzes the objects, con- um analyzes the condition of the object, whether it's in brand new

condition or not, finds me a link for an equivalent thing and how much it cost and estimates how much I can sell it for and if it's worth selling.

This I can do over the course of 2 months. I don't have to do it now. I can

months. I don't have to do it now. I can

just walk around the house. When I see something snap sell.

Snap, sell. And it builds an inventory of all the things that I want to get rid of, but hopefully can be someone else's treasure.

We're going to get into a bit of a more rapid fire of things that I've sent to this AI unrelated to this skill, but just to prove the point that taking a photo can really shorten the distance between what

you're thinking and what you want your AI to do.

We do workouts at home, and I have to remember how much I was lifting.

Uh 5 lbs maybe for the first one, 10 lbs maybe for the second one, and I just voice message it with a photo, and that's how I track my workouts.

Um or I get to live on a lagoon where I get really cool birds stepping on my deck. And I don't know when I'm going to

deck. And I don't know when I'm going to need this. I really don't know when I'm

need this. I really don't know when I'm going to need this for. I'm not a bird photographer, so pardon the blur up here, but I don't want the bird to leave, so I had to use 10x zoom. And I

just lock Lock the bird. What bird is it? It tells me which it is, and for the

it? It tells me which it is, and for the day that I'm going to want to build an app where I can catch birds like Pokémon, I'm going to have all the photos. I'm going to have it all.

photos. I'm going to have it all.

Um What By the way, that was a great blue heron, very popular in the Bay.

Two more work examples. Um

maybe you can relate to this. There's a

ton of tools for task tracking out there.

Uh um maybe some built by the company that I work for.

I still write my tasks in the paper note notebook, really. And I don't know if

notebook, really. And I don't know if you've always wanted this, but I've always wanted a way to merge my physical note taking and my physical to-dos

with the whatever tool of the year um to track my tools to track my tasks online. And so I've always wondered if

online. And so I've always wondered if we could build a merge system, kind of like Git merge, but for my to-do list, and this is exactly what I do. I take a photo. It usually looks a lot more

photo. It usually looks a lot more messy. I cropped it out for you guys, so

messy. I cropped it out for you guys, so you don't get to see my messy table with crumbs. Um

crumbs. Um and another example that you might be familiar with as designers is when you want to just have an idea you have an idea and you want to make it a little bit better.

You want to get fast from idea to execution. We have a lot of tools,

execution. We have a lot of tools, but the best tool, I mean, in my opinion, is still to sketch something out and just say build this. It'll get

it wrong. It'll be awful on first prompt, but it gets you there faster and it's been a way that again I've been cutting the noise between what I want

and what I'm going to try to achieve.

And so with this, the bottom line is that a photo is really worth 100 prompts, and sometimes sketching something out by hand, taking a photo of

it, and building it is the simplest workflow, and you guys, it works like magic.

What does this unlock in your day-to-day? And we've we've seen a lot

day-to-day? And we've we've seen a lot of ways to track context, but I think it's worth pointing a couple examples um of how this context can really multiply

when you use it for work.

Um the stuff you said, and the stuff you wrote, and the stuff you saw with your camera, it all is now externalized and it doesn't just sit in your brain, which doesn't just mean that your AI can

access it, but that you can easily share with others, too.

So let's look at an example. I might

want to help um on a decision I might want might want to help on a decision about a project.

I mean, historically, I might have had my company knowledge. There's a lot of um a lot of companies that have built that, but I also want a way to have the AI know how I think about problems and

how to help me resolve a tricky situation.

It now knows my past opinions. That rant

I was doing when I was saying 20 minutes of rent to a loom, it knows it. It knows

my point of view. It knows how I make decisions.

And it knows the context of the meeting I had yesterday. So, it just helps me make a more accurate response.

I because I am obsessed with to-do's, this is another really good example of seeing these compounding effects into practice.

I don't know if you've tried, but every AI tool wants to give you a daily brief, right? And a lot of them are good, but a

right? And a lot of them are good, but a lot of them miss the mark. They might

say that something is really urgent and your life depends on it, and you really actually tackled it yesterday and it just hasn't picked that up. Or you'd

tackle that with a friend on the way to a meeting.

And having this way to bridge the digital and the physical with your photos for your to-do's and with your voice context helps your to-do list be more personalized.

And with this, I want to show you one last thing, which is all of that I showed you is personal.

And you're building your personal graph with it. You're building your personal

with it. You're building your personal relationships, opinions, and decisions.

Your voice, your photos, and it's all for you. But the real scale happens when you drop that into a team context.

Your team, even if individuals have their own workflows, what really helps this scale is when you start to log decisions that are for a project that other people are working on

or for a problem that you're working together. And I get a little lucky

together. And I get a little lucky because I work at a company that has a really good stack. uh We have Loom, Confluence, and Jira, and I can ramble into the void and I know that all of

that is logged and taken care of, but I encourage you to find what your stack is so that you can start to feed your context not just to your personal bubble, but in a way that can be shared

with your team.

And that is the bigger bet. Your

personal graph is a start, but the team graph is what [snorts] turns AI from your personal assistant into a team collaborator.

And so with this, let me rephrase Mike Cannon-Brookes quote and say that I think it's about time we stop optimizing the engine and we start fueling it with your context.

Thank you all for listening, and I hope I'll catch you around.

[applause]