Today's chat dives deep into the ever-evolving landscape of AI with none other than Tim O'Reilly, a titan in tech publishing and an insightful thinker. Tim emphasizes that the future of AI isn’t a solo act; instead, it’s an ongoing jam session where everyone contributes to the melody. We explore how our perceptions of technology need to shift, much like the transitions from IBM to Microsoft and the birth of Web 2.0, highlighting that nimble thinkers who adapt quickly will hold the keys to new opportunities. Tim also shares his vision of a multipolar AI world, where collaboration and shared innovation drive progress rather than monopolistic control. It's a compelling conversation that challenges us to rethink how we engage with and shape the future of AI.

Takeaways:

• Tim O'Reilly emphasizes the importance of creating new frameworks to understand rapidly evolving technologies like AI and their societal impacts.
• The conversation highlights how historical shifts in technology often result in new mental models that broaden our understanding of possibilities.
• O'Reilly believes that the future of AI isn't predetermined by a single entity but shaped collectively by various players exploring new ideas and applications.
• A key takeaway is that organizations should focus on how AI can expand their capabilities rather than just increasing efficiency or replacing jobs.
• The episode discusses the necessity for diverse competition in AI to foster innovation and prevent monopolistic practices that stifle new ideas.
• Tim argues that the true value in AI is derived from enabling new forms of collaboration and participation across industries, rather than simply optimizing existing processes.

Hi, I'm Adam Davidson, one of the co founders of FeedForward AI and host of this podcast, the FeedForward Podcast.

Today's guest is someone I'm pretty sure you know.

You certainly know his name, and there's a good chance you know him personally because he seems to know everyone personally.

I'm talking about Tim O'Reilly.

He's most famous as the publisher of those O'Reilly Tech books that you see taking up entire sections of Barnes and Noble and the shelves of every tiny techie person you know.

But he's much more than that.

I find Tim so valuable at thinking through frameworks, how to just think about major changes in technology, in society.

He has a free ranging, curious mind that allows him to look at technology that feels familiar and see it in all its strangeness.

He also can look at things that are strange and make them more familiar.

And I was just dying to get him to just sort of put a quarter in the Tim O'Reilly machine and download what he's thinking about AI Tim, as you may know, is a frequent guest at feedforward events.

He's a part of our community, so I know you'll enjoy this conversation.

All right.

I'm very excited to welcome my old friend Tim O'Reilly to the FeedForward podcast.

Hey, Tim.

Hi.

Wonderful to be with you.

Yeah, it's really great to have you here.

You're the kind of guy it's hard to introduce because you've done so much, but obviously, as a publisher, that's probably your biggest impact, is everyone involved in computers has a lot of O'Reilly books on their shelf.

But I would say it's more as a writer and thinker that you've influenced me, although I've certainly owned many dozens of your books over the years.

How do you introduce yourself when you're forced awkwardly to introduce yourself?

Well, I think just what comes to me off the top of my head is I'm kind of a mapmaker, somebody who tries to explain the world.

I love this wonderful quote from Edwin Schlossberg where he said, skill of writing is to create a context in which other people can think.

And so I try to change people's context when I feel like they have the wrong map of the world.

And I can give you lots of examples of that for my career.

And that's where I've been really most influential, where I go, wait a minute, why are people not talking about this?

Or why are they not seeing this framing of the problem?

And as soon as you say it they go, oh yeah, right.

I went, what's wrong with their map?

It's based on licenses and if it doesn't fit the license map, they're not including it.

But if you draw a bigger map that says, oh, it's about Internet enabled collaboration, it's about people sharing their code, it's about.

I was able to tell a whole different story then.

Fortunately, just as I did that, a new name showed up from Christine Petersen, Open Source.

And we had the meeting where I convened to talk about this issue.

We voted and agreed on the new Name or Web 2.0.

Why is it that people haven't made the transition to understanding that the Internet is the platform and they're still stuck in all kinds of metaphors and models from the PC era.

And the web looked like the dot com bust meant that it was over.

And I went, no, it's really, we're just seeing the beginning of something new.

And right now I'm trying to do the same thing with AI on a couple of fronts.

Great.

Yeah.

And this is something I am obsessed with you and I have talked about this, that it's not just computer technology.

I mean, throughout human history we can see mental frames no longer making as much sense for new phenomenon and people.

Even my college degree, we studied the ancient near east as agriculture came and you know, you can understand, you know, how the Greeks and the Romans understood classicists here.

Classicists, yeah.

Oh, that's the new stuff.

Yeah.

That's new stuff.

I'm just kidding.

I was just in Greece.

It was amazing.

So this idea of bringing a mental model to a new technology, a new way of organizing the world is an old and powerful story.

And I think to me the lesson, the overall lesson is the people who are more nimble thinkers who are able to engage the new idea more quickly, they just have a much wider range of options.

They're able to think smarter.

What would be an early example from your career of a major mental model shift that we can just dig in on and think about this?

Well, the one that shaped my thinking originally was the transition that was happening at the very beginning of my career, which was the transition in power From IBM as the leader of the computer industry, the dominant monopolist, to Microsoft as the dominant monopolist.

We're talking like late 70s, is that.

Yeah, late 70s, early 80s, yeah.

When Microsoft, I mean rather famously IBM, thought the hard thing was making the machines.

And then there's this dumb thing called an operating system that a couple kids in a garage can do and we don't care that much.

That's exactly right.

And that was one of the biggest miscalculations in business history.

And what was so important about it was that IBM was so used to this idea that hardware was the source of outsized profits and value.

And all of their competitors were just trying to dethrone them using that model.

And they just missed the point.

There were all these companies, HP, AT&T, digital Equipment Corporation, they were like, well, we'll build this new hardware that'll be better.

Then along IBM created this commodity model for this tiny market, the personal computer.

And they shared the specs and it took off like wildfire.

And it totally changed all the rules.

Microsoft realized early on that actually software, which had been just a satellite to big hardware platforms, was the new source of control.

I think the same thing happened in the early days of the web because Microsoft was still so focused on the operating system.

And some of the early entrants into the battle for dominance of the web, like Netscape, played the same game.

Marc Andreessen was trying to be the next Bill Gates.

And it turned out that that wasn't the right game at all.

And I eventually called the new game web 2.0.

It wasn't about controlling the software operating system layer, it was about big data.

It was about building these services, what I called at the time software above the level of a single device, collective intelligence.

And companies like Google and Amazon took.

Over the world, right where the software they happen to be running was irrelevant.

Like nobody is.

I go on Amazon because they're using Apache or something.

You don't even know it's going to be.

Of course, it's not like the old layers become completely unimportant.

And I think we're seeing this pattern repeat right now.

Because in the what you can call the big data web 2.0 era, companies like Google and Amazon, they learned that having more data was the winning strategy.

And so along comes this.

And actually, it's also ironic that Google made the same mistake that Microsoft did.

The whole AI boom we think of today was as a result of Google publishing their Transformers paper and releasing Bert bidirectional encoding with Transformers the early LLM as open source.

And that was what really kicked off this current revolution because they didn't understand that, you know, first.

As soon as I saw it, I went, oh my gosh, they've just given us a lot of what Google can do.

A lot of Google's secret sauce is now available to everyone.

But the companies that first seized on that, people I'm talking about, people like OpenAI, they think the winning game is you get all the data and then we will be the king of the new world.

And then along comes deep seeking goes, no, not really, you know, and the people at Google saw this early on.

There was that famous paper inside Google.

There is no moat, but a lot of the venture capital world is still acting on this basis and this idea that the real thing is you've got to get all the data.

And you hear these narratives from the people who are developing big AI platforms.

We don't have access to all the data, including all the copyrighted data.

We won't win, China will win, et cetera.

All these narratives are an old model of what is going to make us win.

I actually think when people say, well, who does OpenAI remind you of?

They remind me of Netscape on the first hand, but they also remind me of AOL in the early days of the Internet because AOL thought that their proprietary platform that will have all the content is the winning strategy.

Microsoft answered with the Microsoft Network, which was the same strategy.

Then the web came along, which was just like, no, you just do your own thing.

I think the winning strategy in AI is going to be the enabler of, of what the web had in spades, which is what I call an architecture of participation where you have a real market where players are able to share.

And I've been writing a lot about this in the context of copyright.

You think about, well, rather than all your base belong to us being the model of AI development, it really ought to be.

We've gotten these models to a certain level and now we're going to understand that certain that other players who have specialized content are going to also be delivering models and we're going to have a multi AI world.

We're starting to get there with agent frameworks like MCP and A2A and others that are coming along.

There's some new things that are going to be announced shortly that I think are pretty exciting in that world.

But I think we're slowly figuring out what the new source of competitive advantage will be.

But I do know this, and this goes back to my Map making.

Jeff Bezos once said, people always are too focused on what's going to change.

And actually what they don't get right is what's going to stay the same.

He was using it just in this very simple way.

People always want lower prices, they'll want faster delivery, blah, blah, blah.

But I think there's something else that people want, and what that is is they want opportunity.

And so along comes this new paradigm and people like Sam Altman are like, oh yeah, we're going to take it all and we'll leave a little bit for people can build on top of us.

Just like you could put your content on top of AOL and pay them a toll for it.

But they didn't really embrace the idea that it would be a multipolar world of cooperating AI.

And I think we're going to end up there.

Now the question is, in that world, what is truly valuable?

And I don't think we know yet.

It took me like six years to get from the Open Source Summit to my what is web 2.0 piece, where I was thinking about what are the real implications of the things that I'm seeing.

I was slowly thinking about that, how that future was unfolding, but it wasn't really all clear to me right away.

In a similar way, I think it'll probably be five or six years before we really understand who the real winners are and why.

But there are things that we can see that don't change.

As I said, one is that people want opportunity.

Every big revolution that I've seen in computing has started when there was a lot of opportunity for a lot of people.

Now it might look like there's a lot of opportunity for a lot of people in AI.

There's all these startups, there's all this funding, there's a lot of venture capital activity.

But when you have the market leaders starting out with a philosophy like Reid Hoffman's blitzscaling, which is basically it's a winner takes all market, we assume that from the beginning and everybody's fighting to be the winner, I think you have a basically broken market.

Now.

I love Reid, he's a wonderful, big hearted human being, but he's just on this, he's just wrong.

Markets start when there are lots of opportunities that aren't captured by the early beginners.

First off, the investment philosophy Became a lot like the investment philosophy of the bankers, which is we are creating financial instruments, not companies.

Something like 70 or 80% of all companies now that go public, go public with no profits.

They're really just a betting instrument.

Not only that, the bet is this company is going to have this outsized monopoly.

You saw that really very deeply with Uber and Lyft and wework.

The market didn't pick the winners in those new categories, the venture capitalists did.

It's a kind of central planning where the VCs pile on.

They pick someone early on and as a result you don't have the true experimentation that you see in a real market.

Where, for example, you had all these companies in the early PC, there were people like Dell and Gateway that were just making cheap PCs.

I was trying to figure out better.

There were all these software startups and eventually you had the consolidation into a few big winners.

Same thing with the web.

Nobody really thought that it was all going to be rolled up into the power of Google and Amazon.

You had a lot of competition for many years, but with Uber and Lyft, you pretty quickly.

The guy who actually came up with the initial innovation, Sunil Paul, raised 35 million, which coincidentally was the same amount that Google raised.

But he was just drowned by the fact that Uber and Lyft raised billions and they bought market share.

They got themselves out to be public companies and then they went, hey, you know what?

It's a little harder than we thought to make money at this.

We're going to have to raise prices.

So they basically bought market share.

And if you look at the counter, like alternative history, you would have had a lot more experimentation with business models.

You would have had a lot more local businesses.

Somebody would have figured out something and gotten really good at it in the way that Google figured out something that Yahoo hadn't figured out and eventually became dominant.

How Amazon figured out things that other E commerce players didn't figure out, but we didn't have that experimentation.

Fast forward to AI.

We have the same thing.

We have a couple of winners basically chosen right at the outset, before the business models are fluid.

Thank God for Deepseek, because Deepseek popped that bubble and basically said, no, there is no moat now.

The companies are a little bit like Wiley Coyote, who's gone over the cliff and hasn't realized it yet.

They have to actually start thinking now, how do I deliver real value rather than just trying to capture the market?

And that's not to say that AI doesn't deliver value, it delivers Amazing value.

It just doesn't deliver an obvious path to the kind of monopoly profits that justify the valuations and the investments that are being made.

It wasn't that long ago.

I mean, I guess it's before December and before Deep Seek that the standard thing I was saying, I heard other people say is yeah, it's going to be chatgpt or OpenAI and Anthropic, maybe we'll see if Meta.

But basically it's going to be this arms race where it's going to be more and more billions of dollars to train every subsequent model.

And so it's a natural monopoly.

Eventually one of them will win.

And that felt like a very reason.

It didn't just feel like a reasonable thing to say, it felt like we knew the future.

The future is one giant firm and it is, I mean that literally in a week that and I think that.

The new answer is going to be the company that most enables true innovation in business model and doesn't try to capture right now.

I think those companies that have raised so much capital are actually worse off partly because they have to make money.

It reminds me of a conversation I had many years ago with Jeff Bezos when I was asking him, isn't he worried that Google and Microsoft will come take your space?

He says, you got to really realized at the time Amazon was just a retailer.

He said, we're a retailer.

There's nothing they can do that's not a worse business than the one that they're in.

And there's nothing we can do that isn't a better business than the one of the worst businesses in the world.

And in a certain way, these guys, they have raised so much money that they have to basically do all these extractive things to justify that money.

Which is why I tend to think that the winners are going to be people who don't actually have to live up to that, that can offer things at a more affordable price that can take share because they just can.

Now that's not to say that OpenAI might not still be an anthropic and might not be super successful companies.

And certainly Google's making a real play now, even with Gemini.

The latest models are really stepping up even in the space that Claude had previously dominated from Anthropic.

So the jury's still out, but I still think that the secret sauce is going to be figuring out how to make a new market.

I know you're not ready to say exactly what that is, but can you give me a hint?

I can tell you one thing that I've thought about, and it's not if you look at, say, the story of the PC and the story of the Web, in some ways it was the story of the commodification of something that was previously valuable.

And I think that's going to happen again.

And that's what Deepseek did.

It commodified this thing that everybody thought was super valuable, but we didn't know for a while that it was going to be operating systems in the PC.

And we did see a bit of a repeat on the chip side with intel and then Nvidia.

But there's one other story that seems to be always true, and that's the story that I first learned from a guy named Dave Hickey.

Dave was an art critic, MacArthur Genius.

He wrote a wonderful book that my daughter introduced me to after she had it in a freshman college class.

It's called Air Essays on Art and Democracy.

And in it there's an incredible essay called the Birth of the Big Beautiful Art Market.

And Dave wrote about how after World War II, Harley Earle, who was the VP of marketing for General Motors, turned the automobile into what he called an art market.

And he said, an art market is a market in which things are sold on the basis of what they mean, not what they do.

And from there I went to this idea of in a commodity market, people actually add value with meaning.

So you think of that definition by Dave and you think about the computer industry and you go, who did that?

Steve Jobs?

The Apple meant something.

It was an identity.

And I think that there is a kind of.

It was also certainly part of the open source versus proprietary battle.

What it says to me that there's something in brand and identity that's a safe haven.

And I think it is also, though, part of this what I call architecture of participation.

Because one of the things that's missing in the model of the we will.

We will eat everything.

AI is really values and identity.

And we already see little nudges of this.

You know, where Elon.

Elon Musk says, you know, all the AIs are too woke.

And I'm going to make an unwoke, you know, AI with grok.

Now, of course, GROK is sort of resisting that.

You know, you see so many things online where people have tested and it will critique Elon Musk and he just can't.

He can't make it as unwoke as he hopes because it's just too much common sense.

But if you think about just AI in general.

It's very hard to imagine a single world model that really represents everyone's values.

And we see this sort of framed up already, even apart from the spectrum between Grok and anthropic Grok and Claude, you have people's ideas, what will Chinese AI be versus US AI?

And then the French are like, what about a French AI?

And of course you'd think about, well, there's going to be an Islamic AI that's going to try to have their values.

I saw a fascinating YouTube video about Buddhist AI.

Yeah, Buddhist AI.

And so you start thinking about training for particular values.

But that's only one kind of identity and there are a lot of others.

So I think a lot about.

My friend Nat Torkington once wrote this funny blog post.

This is in the early days of the web, and he said that almost all these startups are reimplemplement of UNIX utilities.

It was true for a lot of them.

I've thought about that many times since.

And I think about the UNIX file system, which had basically a notion of three sets of permissions.

You could basically have a file that belonged only to you.

It was a user permission, and then there was a world permission, which was.

Everybody could read it.

And in between there was group.

And group was this really rich thing that you could explore.

And of course, if you look at social media, we see that group has become a critical differentiator.

The world social media of Facebook and Instagram is really being supplanted by WhatsApp groups, Signal groups, substat.

I mean, there's all these different ways that we're basically exploring that space of what does it mean to carve out a group?

And a group could be a company, a group could be a religion, and then how do we overlap those groups?

So I'm thinking a lot about what that means, how that relates to issues of training and copyright and how an AI would express identity.

And that intersects somewhat with this notion of brand and the sort of.

The idea that part of what makes a brand is that it means something, you know, and that's how you distinguish yourself against the commodity.

But it's more than that.

Yeah, I mean, it's interesting.

I. I sometimes do trainings around storytelling and I differentiate brand and story in the following ways.

That brand.

Brand is generally fairly static.

At least the elements of it are.

You know, you have your pantone color, you have your font, you have like a big brand guidebook.

And that might change once a generation or something, but the story might have to change.

Several times a day, depending on who you're talking to, what specific product.

And I've been thinking of AI as the ultimate storytelling medium in that sense that it can be.

And you can, for better or worse, it depends.

It can be a frustrating process, but you can constrain it enough sometimes to keep some key elements locked in while allowing it to freely explore other ideas.

So you could create a thousand haikus with a rigid haiku structure, but they could be about a thousand different things or whatever.

And that idea that we're able to communicate a more complex essence of whether it's a person or a group or a company or a country or whatever it is, we can communicate that in a more subtle or dynamic way.

To me, that feels like that's opening up something exciting.

I totally agree with that.

And, you know, it's interesting, too, because I had this really interesting conversation with Reid Hoffman about deepfakes, and he was like, hey, don't just get scared of deepfakes.

And they're going to be disinterested.

He uses them all the time.

Right.

I think a deep fake.

Read his audiobook and he interviews himself with deepfakes.

Exactly.

In our conversation, I reminded him that.

That, you know, Plato was really.

I mean, Socrates was really opposed to.

Writing the written word.

Yeah.

You know, the written word.

He was like, this is going to.

You know, people won't remember things.

They won't, you know, and yet Plato made basically the equivalent of the written equivalent of a deep fake, you know, 2,400 years ago, you know, of Socrates.

And we're awfully glad that he did.

And I think of how much there will be this ability to capture with AI a body of knowledge as expressed by a person.

Right now I'm reading the biography of Alberto Hirschman, worldly philosopher.

And it's a good book, and there's lots of references to other books.

Boy, I would love to have a version of that, of Hirschman's life work.

And, yeah, it's probably already in there, you know, in some sense.

And you can kind of sum it up by, you know, just like, oh, you know, you're Alberto Hirschman, you know, and I want to talk to you, you know, but you can imagine that there's a lot of room for going deeper, you know, like if somebody was, you know, like the equivalent of all the research that went into that book was assembled, that would be a product, you know, that would be a richer product than is in the.

Just in what's been scraped off the Internet.

It might be a Better product if it had less of what was scraped off the Internet and more like, oh, this actually has all the.

I mean, Jeremy Adelman had access to all his papers and talked to people who knew him and gathered all that.

You imagine that assembled into a product that's potentially a way more interesting product to interrogate.

And so I imagine a world of those kinds of deepfakes belonging.

That's a new kind of intellectual property.

But in order for that to happen, we actually have to respect copyright and we have to say, oh yeah, somebody built this and we want people to build this.

And so we don't want to basically have these companies that just have a lot of power just come along and sweep up all the stuff, whatever they can, because they can, although we cannot.

Yeah, I find that like, my son is getting very interested in systems.

He's quite curious about capitalism and socialism and libertarianism and he just wants to understand what all these are.

But he's also 13 and running around and he doesn't have a lot of.

I'm not going to give him Das Kapital and tell him.

Give him Hirschman's the Passions and the Interests.

It's a fabulous.

Really.

Oh, okay.

Yeah.

He wrote these sort of short, sweet seminal books, Exit Voice or Loyalty, the Passions and the Interests, a couple of others.

Oh, cool.

Really amazing.

And he was a remarkable synthesizer and he had this real sense of how to get away from ISMs and just to look at the world.

He's very much a compatriot in this approach that we're talking about of looking at the world, of looking at real data and trying to draw conclusions that are not.

I'm ordering that right now.

Although what I was going to say is that what I've been able to do is just get these brief reports that I give him and I use Gemini or ChatGPT to get a deep research report on the topic and then I throw it into Claude and say write a version of this that works for a 13 year old who has these interests.

And it's pretty amazing.

No, it's really astonishing.

And I think that that's really good.

And I guess it's also the case though, that if you even a really good one of these things is kind of the Cliff Notes.

Yeah.

And certainly for publicly available information, it can draw on a lot, but it's still that I think there's some real limitations of what would be possible.

You know, like just again, going, using this, Adelman's biography of Hirschman.

And let's imagine that there's a.

An AI version of that.

This goes back to the very beginning of the web, when people would ask me about books.

And I say books are a form factor.

And there's very different things.

Some books are stories, and other books are effectively a user interface to a body of information.

When you think of a big nonfiction book with a big bibliography and a lot of footnotes, oh, my God, that's a user interface.

Could build a way better user interface.

And I think the web was not actually the right tool for that.

AI is.

Yeah, I agree with you.

As I said, if I had an AI that was trained on all of the research that went into that book back in the day, when I was using before, I talked about Doris Kearns Goodwin's biography of Lincoln, Team of Rivals.

But you can imagine a different kind of product and having a market like that flourish.

And again, that could be not just.

I'm talking about more literature and history, but you could imagine it for the sciences.

And the market that we want to create is one where people are going much, much deeper.

And then those things should be communicating.

And somebody says, the example I've used in things I've written, somebody says, oh, I'd love to have a story about my family in the Maine woods in the style of Stephen King.

And the AI will typically say, well, I can't do that for you, Dave, but I could do something kind of like that.

And what it should be saying is, I can't do that for you.

But Stephen King does it over here.

Let me send you over to his sort of AI front end, because Stephen King now has.

I'll write a custom horror story for you.

Right, okay.

So we've been talking about the mistakes entrepreneurs make in thinking like, okay, this is a winner take all, zero to one, whatever.

But for those of us who aren't venture capitalists or entrepreneurs and are just going to be users, let's talk about how they might think about AI.

And specifically, since this is Feed Forward, people at large enterprises who are thinking about deployment.

I think of LinkedIn.

And that is, Reid may not be right about AI, but he was right about that.

I can't even think of what the second one is.

Reid is right about a lot of things.

I just disagree with him about Blitzscale.

Yeah, that's good.

But I mean, LinkedIn is so dominant that I cannot think of a competitor.

I don't know what that would mean.

I mean, even Google, we can think of Bing, we can think of DuckDuckGo, but with LinkedIn, I mean, he just owns that slot.

Even though I'll be honest, I have notes.

I think it could be better.

I think there's a space to create a better one.

But nobody's gonna do that because of network effects, at least not in the near term.

So I didn't get to profit from that.

And I'm not gonna start a LinkedIn competitor.

But I've used LinkedIn like crazy.

I don't think I've ever gotten a job off, but I use it every single day.

I'm constantly checking people that I'm going to talk to.

I'm reading posts by people.

And let's talk about those folks, like the folks at large, Fortune 100 companies say, who are going to be deploying AI.

Like, I mean, they don't want a monopoly, so it's good news that there won't be.

Right.

It's better for them if there's a bunch of competitors.

How else should they think about it?

My sense is there's a struggle to picture what AI will mean at a big company.

I was at the local public radio station in Chicago, and our boss just struggled.

The guy who ran the station struggled all year long about whether to have a website.

And he finally decided, okay, we're gonna have a website, but it will have no links.

No links out, no links in.

That was his solution.

And I said, I think that's a pamphlet.

Think that's a website.

But that was a thing someone could think.

And today there's still a conversation that's like people are imagining.

Okay, we get to decide how we're going to deploy AI, and we're going to take a cautious approach.

I think that's living on borrowed time, that it's not even going to be a differentiable thing.

But how would you think about it if you were suddenly like, executive vice president of AI adoption at giant company?

Well, I'm not a giant company, but I'm at a company that's dealing with all those issues.

And I think the first thing is summed up by the.

This wonderful line from C.S.

lewis's novel Till we have Faces.

He said, we cannot see the gods face to face till we have faces.

And if you think about that, it's very much what we all need to have our own AI front end.

I had this conversation with Brett Taylor some while back, who used to be at Google and eventually ended up at Salesforce and now has a company called Sierra.

And he's just like, you had to have a website you had to have a mobile website, you now have to have an AI front end, you're going to have to have that.

This idea.

It's funny because there's all this chatter about agents and it's very mixed up because there's really two kinds of agents that are being talked about at the same time.

And one is we can think of as a client agent, it's representing me as a user and they talk about, oh, it's going to go find a hotel reservation for me or make a flight reservation, but what's the other side of that?

And they go, oh, it'll go look at the website and it can actually look at the screen just like a human being, dude.

We call that screen scraping.

We used to do that and then we switched to actually companies publishing APIs.

And there's certainly going to be a role for APIs still.

You go, okay, when it comes to payment, you don't want to have just your AI going, hey, let's chat about what this is going to be worth.

Maybe there'll be some of that.

But for a lot of other things, you can imagine a user facing agent, I mean a client side agent talking to a server side agent effectively, much like you have on the web where the server, you're coming there to a retailer and the server side agent knows all about the company's products or you're thinking about a customer service bot.

This is all the stuff that Sierra and Salesforce and people like that are fighting over.

How do you express the business offerings, the business logic of a company or if you want to get government services.

Oh my God, there's just so much to do.

I mean this whole idea, I guess that goes to another thing, one of my hobby horses right now.

The idea that programmers are going to be out of work is utterly madness to me because so much of what we have to do is to build an entire next generation of systems.

The reason why we've been laying off programmers is because everybody kind of had gotten the web done and there wasn't just that need that for new.

Now we're back to this period of massive reinvention.

And as a avid Vibe coder, I have for the first time in my life started searching for coders to hire because I'm able to get far enough that I know I can't get farther without professional help and maybe that, that the models will be good enough.

But I think there's always going to be, or at least in the near term.

So walk me through how you're thinking about AI, we touched on the specific publishing applications, but just as a business tool.

Another mental model I'm not crazy about is the, like, how many people do we get to fire?

Mental model that there's some kind of one to one relationship between AI and specific employees or large groups of employees.

And okay, we can get rid of most of our customer service people, most of our this, most of our that.

How are you thinking about it at O'Reilly?

Well, first of all, the main thing I think about is what can we do that we always wanted to do that we couldn't do before?

Yeah.

I mean, it's this, the metaphor I keep using is like someone who delivered milk by horse and buggy getting a truck and saying, great, I'm going to deliver to my 30 customers in a third the time.

And then, you know, you're going to deliver to a lot more customers.

You're going to expand your offerings, you know.

Exactly.

And so, you know, for example, we're doing it in areas like translation.

You know, it's like we used to, we have, you know, our products are available in, you know, a small subset of languages in a small subset of countries.

We have a huge opportunity.

We would never be able to afford to translate it in.

It's not like we're getting rid of people.

We just couldn't do it.

We can now just really expand our footprint.

I think you look at other areas like science, you can try more hypotheses, you can test out more ideas.

You look at if you can simplify, you know, a product development, particularly if you're, if you're, you know, it's just kind of what people are trying to do with crowdsourcing.

You can, again, try out lots more ideas, see if people want them.

And, you know, again, there's this whole subcategory of, you know, Kickstarter and GoFundMe and various kinds of crowdsourcing.

And now you've got to go, okay, we'll put AI into that mix.

And I think you have a whole, a recipe for a lot of product innovation.

I think there's also a huge amount.

Again, the basic thing you have to ask yourself is how crappy so many services are.

And so the first thing you try to do with technology is you make better stuff.

Right.

Stuff that took a long.

Yeah.

I mean, I think of I was a staff writer at the New Yorker for A while, and which is still a very handcrafted, bespoke operation.

And it's an amazing place and an amazing operation, but it's wildly expensive, I think.

I'm not telling any secrets when I say it's not the most profitable business in the world.

And a lot of my journalist friends are terrified of AI it's going to replace us.

But I think when I think about all the work that goes into creating a New Yorker piece, the just thinking about what should the piece be and talking to an editor about it, then researching and who has written about this, then gathering that together, thinking, how do I outline the article?

Not writing it, I'm still going to write it.

And then how do I begin to fact check it, which AI is mixed at?

I'm not saying it solves that, but why does the New Yorker, kind of the way it delivers articles, it kind of has a huge jump from a few hundred words to a several thousand words.

Like, what if the right size is somewhere in the middle?

Or what if the right product is an audio product or a cartoon?

Or a cartoon, yeah.

And you can apply this to all sorts of other areas, research and development and others.

Where I start to think there's a dystopian version of this where we're just.

We're already inundated with content and then suddenly we're inundated with infinitely more content.

But there's.

There's also a version of this where AI is reading everything out there and saying to me, hey, you haven't read this Albert Hirschman book.

I think you'd really enjoy it because you've been exploring these ideas.

And so it's creating better, tighter matches between my interests or maybe the kind of chocolate I like or whatever, and finding me new opportunities matching me more precisely.

So that freedom that I start imagining, it just feels very exciting.

I have to say I totally agree.

And the thing that's this kind of goes back to why we need a diverse market where it's not a situation where there's a monopolist who's in charge of everything.

And as soon as somebody figures out a cool new idea, the big guy goes, oh, really good idea.

Sorry, we're going to have to kill you.

Because we need the revenue, we need the profit, we need to take every good category.

You want there to be real competition, because when you have a period of real competition, somebody will invent something and then a lot of other people will copy it and we'll all get our own version of what we learn from watching our Collective innovation.

And I think we.

I guess the computer scientist Donna Knuth once said, premature optimization is the root of all evil.

And I think Silicon Valley is all focused today on premature optimization.

And I'd say a lot of corporate America as well, that it's.

And the way I think about it is nobody knows how Giant CO is going to be using AI, but nobody's going to figure it out without a whole lot of messing around, screwing up, trying things, and having a system to bubble up the best ideas.

And the more people there are who can be experimenting, the more we can learn from each other.

Exactly.

Which means whatever your policy is, it should maximize the number of people in your company who get to use stuff and the ability to find out what they're up to, what's working.

You know, when we think about the world you've described over the course of your career, like an image I had is like in the 70s, there's some number of people, probably in the dozens, who are deciding the future of computers and maybe Wang computers and a couple others, but there's a handful of people deciding, and then suddenly there's lots more people making computer parts with operating systems.

We had a handful of people.

But at least with the existence of Linux, there's an expansion of an open source, there's this broad expansion of who gets to play around with what's an operating.

And so I just love that image.

That feels like what I'm now rooting for is AI just means there's just a whole lot more people playing with it, creating it, making it happen in.

Fields where we never even thought that computers would be useful.

Yeah, that's the thing that I guess is a thread throughout these past expansions of computing is it made it.

The original computers were like back office things.

They did a small number of tasks.

The PC greatly expanded the surface area of what people were using computers for.

And then the web did that again.

And I think AI will do the same thing.

We'll be doing things with AI that we didn't even imagine applying a computer to before.

Yeah, it's helped me with fights with my wife, it's helped me with it.

I mean, I'd say the thing it does the best at for me personally is kind of this conversation.

I've got a thousand ideas bubbling in my head.

Let me blather into one of the models and have it tell me what I'm actually thinking.

And it's not, don't tell me what to think.

Maybe I'll say, ask me some questions to help Clarify.

But that to me, if we can have clearer thinking, that alone is worth.

Is worth everything.

Yeah, I just about to publish a little bit of video.

I just did a conference on the future of coding with AI, and Harper Reed gave a fabulous talk about how he uses AI, which involved getting it to help him write the spec for what he was trying.

He says, I'm just start talking about my idea.

And I say, I tell it to ask me questions, one question at a time with a yes, no answer.

And it really helps me just walk through.

And at the end of that, I'm able to turn it into a spec that I can then give to another model to exit to build.

And the people are really figuring out different ways to get the most out of this technology and also its limits.

Ethan Mollock talks about the jagged edge.

We're all figuring that out and I think the systems are going to get better right now.

I was trying to get some text out of the transcripts of this and I. I put them in the Gemini and I was trying to get some.

Basically there'd been some echo, so I was getting duplicated lines and I asked Gemini in Google Docs to just remove the duplicated lines.

And I thought, oh, that worked beautifully.

And then I got to the end of it and I went, wait, it only did half the file.

How'd that happen?

So I went back to VI and did the same thing myself with regular expressions in about about two minutes.

It was a little more work and it was digging back, you know, 30, 40 years to finger memory of commands.

But it was, you know, there's still things that, you know, are gonna.

It's pretty clear we'll get there, but we don't even know where those potholes are.

Yeah, that's awesome.

All right.

This was fabulous.

Thank you so much, Tim.

I'd love to.

That we could continue talking for hours, and I hope we do.

All right.

Thank you so much.

I really enjoyed it.

I hope you enjoyed that freewheeling conversation with Tim.

He is going to be someone I'm going to turn to again and again to make sense of what's going on in the world.

What stood out to me most is his idea that the future of AI is not something some single company or technology or person is going to decide that we're all deciding it all the time.

We're creating that meaning and reimagining what's possible.

That strikes me as a really good way of describing what we're doing here at FeedForward.

As always, I'm Adam Davidson and this is the FeedForward podcast.

I hope to continue this conversation and many other conversations with you on the Discord and in our upcoming member events.

If you want to talk more about anything in this episode or you have thoughts of other episodes we should do other people we should talk to, please don't hesitate to reach out.

Thank you for listening.