What Industry Insiders Say Is Coming in 2026 (No Hype)

All right, good morning everybody. David

Shapiro here. Uh sorry it's been a little bit since my last update. I uh

I'm just not going to take too much time for this, but I just want to say like my burnout recovery continues and it turns out that doubling down on gut health was the way to go. So I felt a little crappy

there for a couple days. Um some of the secondary effects, it's called a die- off reaction. Um, but my uh numbers are

off reaction. Um, but my uh numbers are moving in the right direction and I feel better than I have in many many years.

So, with that being said, let's get right into today's episode. Uh, to

provide a little bit of context, a little bit of framing for uh where I'm going to go is that I have been talking to plenty of industry insiders uh as well as industry adjacent people. Uh so

I have been making use of this time as I'm working on uh wrapping up my book, The Great Decoupling. Uh so this is about where it's going in the long run and how we can pivot. Uh I'm not going

to bore you with the details about the book, but I just wanted to pro provide some context that I'm talking to all kinds of industry insiders and I'm going to more conferences and that sort of

thing. So, uh, what I'm sharing is not

thing. So, uh, what I'm sharing is not necessarily like predictions about the future like what I used to do because that was me as an outsider kind of guessing. So, what I'm here to tell you

guessing. So, what I'm here to tell you is what's actually coming down the pipeline in 2026. So, first and foremost on the technical front, the biggest

change that's coming in 2026 is multimodal is is going to become the default. So we already have where

default. So we already have where there's you know uh image native understanding and some uh voice native but video native is is really what's

coming. So this has been all but

coming. So this has been all but confirmed for Gro 5 and a handful of other models. Uh so real time video uh

other models. Uh so real time video uh understanding and generation is definitely going to be a big thing and that's going to make a lot of people sit up and take notice because now you're no

longer talking about just language.

We've bootstrapped artificial intelligence with language. Turns out

learning language first was the way to go just because text data is easily tokenized. But now we're learning to

tokenized. But now we're learning to just tokenize everything. Actually,

Dylan over at Dylan Curious, he pointed this out to me over a year ago. He's

like, there just seems to be some kind of magic in just tokenizing everything.

So, he was uh he was way ahead of the curve there just in that very incisive observation. So, that's going to

observation. So, that's going to continue. And the thing is when you have

continue. And the thing is when you have that kind of expansion of capability inside, you know, large multimodal models as as we probably should call them, although LLM has stuck. Um, and

the reason that I want to bring this up is because some people are still, you know, saying, you know, there's I had a friend reach out. He's like, you know, I'm seeing all these videos of people saying that, you know, LLMs can't scale

to AGI or ASI, so it's all dead in the water. I'm like, no, that there's

water. I'm like, no, that there's there's nothing. There's no there there.

there's nothing. There's no there there.

Uh and the reason is because we haven't even finished figuring out what all we can tokenize, let alone all the algorithmic improvements and other scale improvements. So on a technical front,

improvements. So on a technical front, the most exciting thing that I expect that we'll see next year is video native models, which that's really cool. Next

is going to be, you know, benchmarks. Of

course, benchmarks are continuing to saturate an at an incremental level. Uh

but what I will add is that uh benchmarks are a proxy for real world performance but as we've seen with Grock you can overindex and overfitit to benchmarks um and that that leaves you

with a very weak model. So there is some secret sauce that other uh companies like Google and Anthropic and OpenAI understand about model quality that Grock just hasn't figured out yet. And

that's not to say that Grock is useless.

It does have its uses. uh but it is very clearly overfitit to the benchmarks. So

we're going to see more benchmarks and the benchmarks that we do have are going to be incrementally more saturated. And

the point there is that we're going to get to a tipping point. So this is what happens in all technologies. This is not just an artificial intelligence thing, but we get to a tipping point where suddenly the new technology is not just

a nice shiny toy, something to play with. It's like, oh, this is actually

with. It's like, oh, this is actually good. Not only is this good, this is

good. Not only is this good, this is good enough to deploy and scale. So

that's kind of where we've been at where for certain tasks uh language models have been getting to the point where it's like ah this is no longer just a curiosity or a novelty. This is good enough for production. So we're going to

continue to see that but we're going to see it expanding in terms of the number of kinds of tasks that it is production ready for. So that's when you know

ready for. So that's when you know benchmarks are one thing because what's possible in the lab is is one thing but what's possible in business which is where you really move the needle for

most people that's a different thing.

Now the other technical front that we're going to see in 2026 is the world models and the interactive real-time models. So

this is basically the game engines.

We've already seen the first clues of this uh mostly coming out of Google and Deep Mind. Uh, but that's going to scale

Deep Mind. Uh, but that's going to scale very rapidly because there's a lot of people that are saying, "Hey, you know what? If you can have a real-time video

what? If you can have a real-time video game engine and just wholesale replace your your current, you know, Unreal 5, there's a whole lot of money there. So,

there's a lot of potential for creative destruction. Now, I'm not saying that

destruction. Now, I'm not saying that Google Deep Mind is going to get into the video game engine uh field. And I'm

not saying that Unreal is going to, you know, or or uh or the Frostbite engine or whatever that they're going to throw out their codebase and switch to AI models next year. But what we probably

will see is the toolkits being built, the the baseline SDKs and those sorts of things. Uh, so I, you know, I want to

things. Uh, so I, you know, I want to temper a little bit of the optimism and say that 2026 we're probably not seeing a fully deep neural network AI native game engine. Um, but we're going to see

game engine. Um, but we're going to see the rudiments and we're certainly probably going to have the first playable examples, the first MVPs or proofs of concept. Uh, but we're not going to see it scale up. You know,

you're probably not going to see something on Steam for sale that's like this is 100% AI. Uh, that's just that's we're just not there yet. But we're

going to see that that's what's coming and that's going to be really kind of on people's radar. Now, so that's that

people's radar. Now, so that's that wraps it up for technically what's coming down the pipeline uh in in 2026.

And that's not of course that's not 100% of everything. That's just what's top of

of everything. That's just what's top of mind for me based on the conversations I'm having. The next thing is business.

I'm having. The next thing is business.

So I think that there that the primary thing that's going to happen in 2026 for business is that a lot of businesses at all scales are going to sit up and say wait agents are actually good now and

that's going to be a combination of one just exposure but number two agents are going to continue to improve. So if

you're in the AI agent building space make sure you stay at the cutting edge the bleeding edge because I I believe 2026 is going to be your year to shine.

uh and and what the reason that I'm confident saying this is because plenty of businesses are already deploying agents. Uh but the thing is uh there

agents. Uh but the thing is uh there what a business calls an agent is not necessarily something that's fully autonomous. It might just be automating

autonomous. It might just be automating one task rather than an entire value stream or workflow. Furthermore, uh

because enterprises in particular are very very riskaverse, they want something that is very highly controlled and you know where you've got a transparency pane um and where you've

got logs and you know reversible decisions and all that kind of fun stuff. So many businesses, and this is

stuff. So many businesses, and this is not most by the way, but many, more than one, and and certainly uh pushing closer

to 50% um than than you might think, are tinkering around with at least one use case for fully autonomous or semi-autonomous AI agents. And the the

areas that they're targeting are number one HR and number two legal, but then also every business also has some of their own bespoke kind of offerings. Uh

so whatever is particular to their industry. Um they're looking for AI

industry. Um they're looking for AI agents to be a differentiating factor.

So like let's say for instance uh your your business model has to do with pharmaceuticals. you might be trying to

pharmaceuticals. you might be trying to use AI agents to help, you know, with research or regulatory paperwork or that sort of thing because that's something that is particular to your industry and

can be a differentiating factor. And so

there's so that's there's kind of two ways that agents are being deployed which is one that industry specific thing and then the industry agnostic thing which is HR and legal. Every

business has HR and legal. So that's

already happening. But when you look at the technical improvements that are coming down the pipeline, combined with the ongoing uh improvements in terms of the software architectures, the design

patterns, the uh API best practices and all that other fun stuff that's constantly maturing, we're going to get to the tipping point sometime during 2026 where uh many industries and

enterprises sit up and take notice and say agents are actually good now. Let's

let's explore where what else we can do with this. This has happened plenty of

with this. This has happened plenty of times throughout the last two decades in technology uh during my career where there's a technology that is like oh that's that's cool that's nice to have.

So, you know, the the example that I that I'm most familiar with is uh VMware. When I first started, it was

VMware. When I first started, it was GSX, which was a a supervisor that ran on top of Windows. And then they switched to ESX, and that was when uh it was a hypervisor, so it was a

minimalized operating system that didn't take much resources and just ran, you know, the containerized operating systems. And now containerization is, you know, standard. So, I was there at

the very beginning, you know, I'm I'm I'm an old guy now. I think I can qualify. Um, and what I'm telling you

qualify. Um, and what I'm telling you right now is that AI agents are in the GSX uh kind of um, you know, era and that 2026 is going to be the equivalent of

the ESX era. And if you don't understand what GSX and ESX are, that's perfectly fine. I'm a VMware nerd. Um, so, you

fine. I'm a VMware nerd. Um, so, you know, but the point is is it's going to get to a tipping point. Think of it as the difference between a BlackBerry and an iPhone. you know, the Blackberry was

an iPhone. you know, the Blackberry was super popular for people that could afford it, but it wasn't anywhere near as useful as the iPhone. But then when the iPhone came out, smartphones were for everyone. Um, so that's kind of I

for everyone. Um, so that's kind of I think we're going to have the iPhone moment of agents during 2026. And I want to point out that at the end of 2024, I predicted that 2025 was going to be the

year of the agent because what I saw coming down the pipeline is that agents were not really even at MVP yet. uh this

time last year, 12 months ago, they were still very much hypothetical, but now we're learning the rules of the road and the best practices. So, keep your eyes and ears open for agents, not just the

technical advancements, but also the business use cases and the ROI and the risk management. Now, the next topic

risk management. Now, the next topic that I want to talk about is jobs. Uh

because this is one thing that people are always asking me, which is how hard, how fast, when are the jobs going away?

And in one conversation that I had, something kind of crystallized for me.

And that is that, you know, right now unemployment is nominally good. Although

when you look closer at the data, you see that like the gigification and and the um and the uh u oh crap, what's the word? The uh procarity is rising. So gig

word? The uh procarity is rising. So gig

gig work is rising. 65 million Americans are free freelance workers or gig workers, which is insane. The American

dream, it does not include precarious work. The American dream was solid,

work. The American dream was solid, reliable full-time employment. And when

fully a fifth of the population is precariously employed, that is not that is not the American dream. So that's

getting worse. Then you look at marginally attached workers and discouraged workers. Both of those are

discouraged workers. Both of those are on the rise. Youth unemployment is on the rise. So even though topline

the rise. So even though topline unemployment is good, the rest of it is not looking as good. So when you look closer uh at the data, it's it's more

alarming. Then you combine that with the

alarming. Then you combine that with the fact that u they're forecasting a recession within the next 12 to 18 months. Of course, you know, people have

months. Of course, you know, people have been calling for that for a while. We

were supposed to have a recession this year. uh but you know the insane

year. uh but you know the insane investment in artificial intelligence driving the magnificent 7 has prevented that at least on the stock market. So

people are saying ohh but there's a bubble uh and you know the thing is the companies are still profitable. Um, so

it's not as, you know, there is some speculation driving up stock market prices. And I pointed out that like

prices. And I pointed out that like having a stock market pullback or a stock market correction, you know, like yes, that's to be expected. And some

people see, aha, that's a bubble. That's

not what a bubble is. A bubble is when you have speculation built on nothing, but Microsoft and Google and Nvidia are all wildly profitable companies. Um, so

what you're seeing is some ir some irrational exuberance around uh the hype and speculation, but it's not it's not a house of cards. It's not built uh on on

sand. Um, now with that being said, you

sand. Um, now with that being said, you know, after you have a long stretch of all-time highs, which I think we're about what 37 months into a bull market and the historical average bull market

is like 65 months or something like that. So, we're a little bit we're a

that. So, we're a little bit we're a little bit past the halfway mark of a bull mark bull market. So, you're going to see some corrections, you're going to see some pullbacks, and this is not financial advice, by the way. This is

just my hypothesis based on all my reading. Um, but at the end of that, so

reading. Um, but at the end of that, so maybe within the next, you know, year or two or three, we're going to see a major recession and there's going to be, of course, you know, a lot of cooling off

on the job market. And what we're probably going to see is more of what's the new normal, which is more jobless recoveries. Because what's going to

recoveries. Because what's going to happen in the next year or two is that AI is going to get better. enterprise

strategy and corporate adoption is going to get better. And so then when uh you know the stock market uh does have a major pullback, we have a recession, when interest rates and credit crunches

and all that other stuff locks up, uh companies are going to shed a lot of employees to to survive the recession or the pullback, whatever happens to come down the pipeline. And then when things

loosen up, they're going to say, "Well, why don't instead of hiring people back, we just deploy more AI and more agents."

And we saw a very similar rapid technology adoption during the recession and that was the adoption of remote work. So here's an example. I worked at

work. So here's an example. I worked at Cisco Systems uh in the mid20s and one of the ethos, one of the values of Cisco systems was that you should be

able to work anywhere at any time. And

so guess what? You know, we had WebEx accounts and we had at home VPN. So it

was just really easy to work to to live and work as though you were in the office at all times. Even so, they wanted you to be in the office uh as much as possible just because of work

culture and you know so so on and so forth. But what happened with the

forth. But what happened with the recession or sorry not the recession the pandemic is that that proved to every company you need to have the ability for employees to be just as effective at

home as in the office. So there was a huge investment in things like WebEx.

And of course WebEx died because Cisco was resting on their laurels. And Zoom

came in and ate their breakfast, right?

They ate their lunch to say, "Hey, we can do remote uh conferencing and remote video better than WebEx. Great. Go for

it." And then of course you know uh bring your own device and other cloud you know uh ver you know uh thin app virtualizations all that kind of stuff took off because of the pandemic. What I'm here to say is

that the next time we have a recession or a depression and a and companies are forced to lay off a bunch of employees.

They're going to do the same thing but they're going to say how can we get this work done without human bodies? And

that's when the surge of adoption of AI agents is going to really ramp up.

That's my current prediction is that the next time we have a jobless recovery, that's going to prove to a lot of business leaders, oh, AI agents are the way to go. So, just be keep your mind

out your your your keep that in kind of your back burner. Uh, you know, let that let that ferment in your in your mind and and just, you know, pay attention to

to the trends. And then the last thing that I want to talk about is uh humanoid robots. So, a year ago, almost a year

robots. So, a year ago, almost a year ago today, maybe maybe uh December last year was when I made that prediction that 2026 was going to be the the year

of the robot. Now, that holds true. And

what we're seeing, we're already seeing kind of the first clues with all the the huge number of robots being released, uh, America, China, even some in Europe

and Japan, where humanoid robots are really getting close to, uh, product market fit. And so product market fit is

market fit. And so product market fit is is different from an MVP or a PC. What

we've been seeing for the last couple years is proof of concept. You know,

when whenever a Tesla Optimus has walked their robot out to say, "Hey, look, it's walking on its own." Or when Boston Dynamics shows that, you know, it can do back flips and can do some parkour.

That's a proof of concept. And so then this year, 2025, what we saw was MVP. So

minimum viable product. That is the first iteration of something that is useful. It's not necessarily good yet,

useful. It's not necessarily good yet, but we hit ro we hit humanoid robot MVP this year. And so next year, what we're

this year. And so next year, what we're hitting is product market fit where one of these guys, I don't know who it's going to be, um maybe it's Figure, maybe

it's Boston Dynamics, maybe it's one of the Chinese companies like Unitry, one of these companies is going to hit product market fit where the capabilities of the robot are good

enough to justify the price tag and demand will surge. Now, of course, everyone who's building robots is counting on this. Everyone who's

building robots sees that this is a future hundred trillion dollar industry and they want to be a first mover. Now

the fact that we have a lot more than just one first mover here. We've got

Tesla Optimus, we've got Figure, we've got Atlas, we've got Unit and and that's just the top four off the top of my head. There's a whole bunch of other

head. There's a whole bunch of other ones out there. There's more than a dozen h but now there's probably more than two or three dozen companies trying to enter that robotic market space. So

we're going to hit product market fit next year. Now, just because you hit

next year. Now, just because you hit product market fit doesn't mean that that's the final form. It doesn't mean that it's the best version. That just

means that you can justify the price tag and that it is suddenly good enough that that that uh demand scales.

So, that's what I expect to happen by the end of 2026. By this time next year, we will have hit product market fit for humanoid robots. Now, does that mean

humanoid robots. Now, does that mean that we're going to have shortages because of, you know, the actuators and the onboard computers? Who knows? So,

here's a bonus. Here's here's a bonus topic that I didn't plan on talking about, but I got some time left on the recording, and that is the not just the AI bubble, but you know, the shortages.

So, I asked a question on Twitter. Um,

and of course, I know Twitter is not the best source of information, but I also asked AIs and other people, but I asked, you know, is anyone genuinely actually concerned about, you know, not having

enough power or chips for artificial intelligence?

And by and large the consensus is yes there might be a squeeze over the next 1 to 3 years. Uh and the reason is just because power grids take time to adapt

it takes time to install solar. If

you're going to build nuclear that takes even longer.

But what we're seeing is a rise in collocation of power. So whether that's natural gas, whether it's solar, whether it's building data centers near existing nuclear power plants as Microsoft is

doing. Uh there's all kinds of ways

doing. Uh there's all kinds of ways around that power constraint. So even if the grid is outdated, where there's a will, there's a way. If you need the

juice, someone can produce it. So that

is probably the easier problem to solve.

Now, ramping up chip production is also very hard because building chip fabs takes a while. Um and it's due to how delicate the process is because you need

ultra pure water. uh just that as step one if you don't have a source of ultra pure water you don't have a chip fab sight unseen full stop. So you know that's one of the things and then also

however just because it takes time to spin up new chip fabs that doesn't mean that existing chip fabs aren't also churning out chips. So you know even if there is

out chips. So you know even if there is a shortage Nvidia is still churning out millions of GPUs a year. So yes, there is hypothetically a little bit of a

squeeze, but that doesn't mean that like we're going to run out of AI. It doesn't

mean that we're going to shut down data centers. It means that over the next uh

centers. It means that over the next uh you know 1 to 3 years, electricity prices near data centers are going to go up by 20%. Which, you know, yeah, I'm I'm not going to beat around the bush.

That sucks for people who live near data centers. Uh but at the same time, that

centers. Uh but at the same time, that is a price signal. Both of these are price signals. And so we see uh other

price signals. And so we see uh other entrance entering particularly in the chip space where Intel, AMD, Nvidia and others ARM are all competing and you

know we even have Grock the GRQ um and and others entering into this space because they're like hey there is a there is such a huge demand for chips that it is now worth it is now

cost-effective to build a new startup or to pivot and produce more GPUs and TPUs.

Um, so while there is a short-term squeeze, and I'm saying squeeze, not shortage, because again, it's not like we're going to run out of power or run out of chips. So, there's a short-term

squeeze on those two resources, but the price signal is very clear and the market is already pivoting. So, whatever

squeeze we have, whatever short-term squeeze we have is going to be erased and it's just going to fade into the background and people aren't going to talk about it anymore. Um, now, you know, that's my opinion. People are

welcome to disagree, but when you look at the amount of investment going into power generation and chip production, like you know, where where money flows, problems are soon solved. So, with that

being said, thanks for listening and I'll check you all next time. Take care.