Arc Of AI Video Podcast (Transcript)

Recorded 2026-04 for AI Native Dev, Austin 2026

HTML

NA

NA

MP4

MP3

HTML

Welcome, Mike.

Good to see you.

Thank you very much for being [inaudible 00:00:01] second day we do that, like, ever.

So, yeah, yeah.

That’s fantastic. This has been great. Yeah.

So, we started yesterday. It’s been…like, we had some technical difficulties, but we powered through.

Oh, yeah. It’s live.

Exactly. And today, hopefully, it will be even better.

Yeah.

And yeah, it’s live. I mean, whatever happens, what happens?

Yeah. We’re here. We’re here for that. That’s why I’m here.

Exactly.

Yes.

Exactly.

Exactly.

Right. Speaking about all this technology and machines and stuff…

Exactly, exactly.

And we have stuff to discuss.

We have lots to discuss.

Right. So, what’s going on?

So, the talk that I’m giving here today is called "Thinking With Machines." And the emphasis is on the thinking part, rather than the machine part. So, we talk a lot, we spend a lot of time trying to figure out how to create skills, agents, Claude files, how to make the machines better at what they do. What I wanna talk about is how to make humans better at what we do. This idea of humans and computers working together has been through the history of computing, 50, 60, 70, 80 years, and we’re at a point now where we can do so many more things that were only imagined half a century ago, and now I think we are at a fantastic opportunity point to start thinking with, rather than for, machines.

Oh, I love it. I love it. And the reason why we can do it is that because this time, the leap is to non-deterministic reasoning.

Yes, yes. I think that’s a huge part of it. We talked about fuzzy logic and all these other things 10, 20, 30 years ago, but now we actually live it. These devices are using the same statistical patterns that so many other parts of nature use in order to adjust and adapt and so on and so forth, so I think we’ve got great opportunities now to start taking advantage. We don’t have to do line-by-line coding. We don’t even have to do declarative coding. We just simply have this kind of deterministic statistical interaction, that allows us to start exploring things in ways that we never could have thought of even 25 years ago.

So, how it differs for us? What changes for the humans?

Here’s what I think has been going on. I think for the last 10 years or so, we’ve been exploring this notion of how we, AI can start doing things for us, or start doing things that we used to do, the drudgery and so on and so forth. And I think that’s fine, but one of the things I think we’re missing is one of the things AI can do for us and with us is help us get better, help us get smarter. One of the people I’ll talk about in my keynote today is Douglas Engelbart, the guy who invented the mouse for us. But he had this idea of the whole thing that he should be working on is helping people improve on how they improve, bootstrapping, getting us smarter all the time. And that’s what AI can help do for us, not replace us, or have us no longer have to do a task. In fact, in some cases, we see some examples of people de-skilling, not learning as much as they used to, having a difficulty of transferring knowledge from senior to junior, because AI has taken over so many things as black boxes. So I want us to start spending more time using the tools to teach us to learn, as well as solve problems.

How AI can help? What this feedback looks like? What inputs we can get, to get better from AI?

Yes, I think this really gets down to what’s called creative process. There are typically three stages to creative thinking, creative problem-solving, and that’s brainstorming, refinement, and execution. AI is fantastic at brainstorming new ideas, new possibilities, new suggestions, reaching out to papers and information that we never would have seen, we never would have thought of before. It’s okay at helping us practice, but it isn’t really being designed to help us as a practice tool. It’s more of an idea-generating tool. It’s really not very good at execution. We’re having to program an awful lot of single-shot in order to get it to execute. So, focusing on this creative loop, and creating interactive AI, that knows when we’re in brainstorming, knows when we’re in practice and refining, knows when we’re in execution, I think is gonna give us an extra leap forward in being able to accomplish things we hadn’t done before.

This is fascinating. Give me an example, how this loop looks like in practice.

So, an example I’ll give in my talk today. Say I wanna build a simple web application. My organization has its own rules about how those things are built. Rather than me feeding a bunch of data into a system and pressing a button, and having a black box generate an app, the experience is gonna say, okay, we’re building an app today. We’re gonna use the rules of the company style guide. One of the things is you need to decide which formats you’re gonna be interacting in. We have five that we support. Which ones do you wanna support? So, I’m making the decisions on how to do this, and it says, "Fine." We also have variables. We have these other steps along the way. We have to support more than one language. Making decisions, step by step, because AI is not good at deciding, it’s good at presenting. So, now I’ve made decisions. Now it says, okay, we’re at a critical juncture. Have you made all the decisions you wanna make? Because I’ll go ahead and build a prototype for you if you like, at this point. So, we’re at a cliff, or we’re at a point where we can decide to go back, do some more refinement, do some more experiment going ahead. So, actually creating these coaches, these coaching documents, are what I’ve been spending my last year doing, and working with educators and other people to now have this notion where, where you would normally have a skills file or an AI function, you have a coach, and that coach is aimed at me as a human. We decide what to do, we do it, and then we get results. This is what you learned, this is what you did. It turns out you can teach at a rate that’s faster than typically we do today in other kind of learning modes, and still get the job done in about the same amount of time as if I was already a senior dev.

Fascinating.

That’s AI.

So, what I’m hearing is, and there is a particular implication for us how we write those skills…

Yes.

…is that those skills should gather information, recognize patterns, group them, and then present options.

Yes, and present options at the pace that humans are interested in. We have a big thing of AI brain fry, or cognitive load. There are too many options, too many possibilities. We need to construct our human skills, we need to construct our coaches, to dole this out at a pace that humans understand, and that helps us to retain the information. High engagement means I’ll remember this later. Low engagement, if I’m just pushing buttons, means I won’t be able to solve this problem later. I don’t have embodied knowledge in myself anymore. So that’s what we wanna create.

I wonder if we can even capture signals of how thoughtful humans were when they answered those questions. For example, I was presented with options. And the thing that we usually do is we ask our skill to present their opinions.

That’s right.

Maybe we shouldn’t do that…

Yeah.

…because humans will be inclined to choose…

Yeah.

…whatever AI gives them…

Yeah.

…instead saying all the options are great. Now think and choose one, and also measure the time that passes between the presentation of the question and the answer. And if it was too fast, actually ask, did you think about it?

Did you really… This is brilliant.

Yes, yes.

So, I haven’t really gotten to this stage, but now what you’re really talking about is engineering this, more than just an engagement machine, but also one that helps us. "You didn’t… I’m not sure you’re paying attention. I’m not sure you thought this through." That is totally fine as well, because that’s what a senior would do, focus on individuals and their abilities, and change the pace, change the advice, change the format, to help them learn the most. That’s a brilliant idea.

Yep, yep. And then we can also combine the answers, to learn whether the user actually thought about it.

Yeah.

Because, instead of just accepting in our skill, instead of just accepting what the user told us, the skill can apply reasoning…

Yes.

…and say, "They chose two options that don’t make sense."

Yes.

"I should flag that…"

Yes.

"…instead of just continuing with it."

Yeah. Some of the coaches I’ve been building, I’m working on a book of these. Some of the coaches I’ve been building are very conceptual. They’re very much like you say. I built a coach for trade-offs. We are presented with a scenario. What choices do you make? And it says, "Now, you realize, by choosing this, these other options are no longer available to you. Are you sure you want to do that?" So it’s literally, it’s not me trying to build an app or solve a particular bug problem. It’s me in a learning experience, the same kind of experience I would have with my seniors in an organization. I think you’re hitting right on a really important element. This is more than just computer-aided instruction or anything like this. This is real stochastic interaction, with someone who has a body of knowledge I might not have right now…

Yeah.

…and I’m gonna get something more out of this by the time it’s done. I think that’s a brilliant way to think about it.

Thank you. And that’s an interesting, again, an interesting aspect of skill-building, that I don’t think a lot of people think about. How do you make skill force you to think about what you are doing?

Yes.

And it’s doable, right?

It’s totally doable.

Because we have reasoning on the other side.

Yes, and we have so many more options and tools we have today. One of the ways we build skills is making mistakes.

Yeah.

Right? Trial and what? Trial and error.

Yeah.

And that’s an essential part of the creative loop. And we can build tools that help us make mistakes without hurting ourselves. Without, as we say, drilling a hole in the bottom of the boat of the organization. You get to try something out in an environment that’s safe and effective. So I think we’ve got so many more opportunities before us now. We can write our own future as to what our relationship with AI is, and how it makes us feel, and how it empowers us. This is a fantastic opportunity.

Absolutely.

Yeah.

Love it. There is such a fascinating and important topic that I want to finish with couple of call to actions.

Okay.

So, what people should do in order to do this better? What people can start doing tomorrow, when they use their skills, when they build their skills, etc.?

Yep. I think what we need to start thinking about is increasing the engagement level of a skill. Rather than treating them as black boxes, make sure there’s more interaction involved. Figure out how you can embed into the skill the knowledge of seniors in your organization, people who have a lot of experience, people have done this before, and make that a kind of a conversational aspect. These AI tools are fantastic at conversation, not just doing, and we should take advantage of that.

And it’s a completely new way to think about it. I think we’ll probably [inaudible 00:11:26] more interviews.