Four Predictions For How AI Will Change Software in 2026

Let's get into our 2026 predictions. I

think AI progress is still really fast and it's kind of mind-blowing where where we are now versus where we were a year ago. And I think that will be the

year ago. And I think that will be the case in at the end of 2026. I think

we'll be sitting here being like, "Wow, it's like actually a lot better." And

the world and AGI, however you define that, is a lot more complicated and hard than we thought. And we will not have AGI in 2027. I think for the most part if you are an engineer who wants to

incorporate AI into your process but doesn't want to reinvent the skill of software engineering there's going to be a home and a job for you for a very long time. Actually the big AI CEOs like

time. Actually the big AI CEOs like Daario and Sam Alman in 2024 said that 2025 would be the year of agents and it really was like we got there. We did it.

I think Sam said like it's less than a thousand days away maybe like a year ago. So that's in 2 years and now their

ago. So that's in 2 years and now their timelines have shifted.

>> [music] [music] >> For people who don't know, Brandon is our fearless COO at every >> It's true. It's true.

>> I'm super psyched to do this episode. Uh

I'm psyched to do it with Brandon and al also just psyched to get to reflect on 2025 and make some predictions for 2026.

2025 was just like it was the year for us. It was it was this breakout year.

us. It was it was this breakout year.

And if you look at our our like revenue chart, um it's so interesting because uh like we're six years into this business.

Um the first like four or five months it was like straight up and to the right and we were like this is [ __ ] amazing. This is business.

amazing. This is business.

>> What was that business? Uh it was like a it was a bundle of newsletters and so it was just people just subscribing specifically for newsletters. We would

add new newsletters to the bundle and it would grow. It was like at the beginning

would grow. It was like at the beginning of co and so we were like this is amazing. We're

going to be rich. And then basically for like another four and a half five years it was just like more or less flat. Like

if you really zoom out it was more or less flat. Um which was difficult. And

less flat. Um which was difficult. And

there were there were a lot of like false starts and moments like um where it looked like it was getting better and then it would start getting worse or just stay stay the same. And um I think

like year over year it did kind of grow a little bit but like for all intents and purposes like there was no progress made in the business uh from a just a revenue perspective. And then this year

revenue perspective. And then this year it's just like we just totally bent the curve. And specifically starting in like

curve. And specifically starting in like I think starting in June >> we basically doubled the business. Um,

so we're at like about we're about 150k MR. Um, and we were at I think in the 60s, like mid60s to 70s starting in June. So there's just something like

June. So there's just something like really big happened and it's so fun to um watch this thing that has been in this justational period for 5 years like

suddenly start to break out.

>> Well, I think it was our was it our first Claude Code Camp in June?

>> That was one. I think honestly I think what what happened was um in June we really had built this um engine that was

ready to get people to subscribe. So

basically we had the newsletter >> um and then we had we had a couple different of our AI software apps in market. So I think we had launched Kora

market. So I think we had launched Kora in June.

>> I think Kora was June.

>> That was in June. And then and Kora is our our AI email assistant. Um we had Sparkle. We had um a version of Spyro

Sparkle. We had um a version of Spyro which is our um our AI ghost writer and then I don't think we had Monol yet but we had three we had three products in market and so it was it was this really meaty subscription because you pay one

price you get access to everything you make um but we needed like a couple moments to like almost reintroduce ourselves to the business and so we had that launch moment with Kora. Um, we did

a cloud code camp which is like basically that was like when cloud code was breaking was starting to break out and we sort of noticed it and we started to do these live webinars called cloud cloud code camps for subscribers.

But the big thing was I went on Lenny's podcast. Um, and that was like in June

podcast. Um, and that was like in June and that was one of those moments where it was like oh um it res I think it reset some of people's expectations about what every is and what you can get

from it. Yeah. Uh and and so we grew

from it. Yeah. Uh and and so we grew like 20% that month and it has just like basically been that way since then.

>> We grew 22.63% that month after consistently growing between 1 and 5% for years.

>> Well, here's here's what's really interesting. What people should know is

interesting. What people should know is we started growing like about 5% month over month starting in like maybe let me see let's look at this. It's like

>> we started growing about like a couple percent month over month between like two and 5% month over month starting in um May of 2024. That's actually when the growth

>> what else happened? [laughter]

>> Brandon joined.

>> Wow. You can really see the date that I joined [laughter] here.

>> And so once Brandon joined um it just the curve like started to consistently bend up. it was like starting to go up

bend up. it was like starting to go up and then and so I think the business at that point was starting to like reconfigure around this whole um thing that that we started to started to build

together and um and then it needed like a moment to catalyze it and like sort of light it on fire and I think that was like the Lenny podcast moment >> and what's been interesting after that

is it has consistently delivered. So

after the Lenny podcast, we've grown at least 10% each month. We've had a bunch of other launches since then, but I do think the biggest thing was um a

rejiggering in people's brains of what every actually is from a newsletter to a bundle. And we're not even there yet.

bundle. And we're not even there yet.

Like that's going to be what 2026 is really really about.

>> Totally. Um and I [clears throat] think um yeah, it sort of um it sort of became a real company, too. So we were I think at the at the beginning of the year we were

>> like maybe five people or something like that. Five people full time and now

that. Five people full time and now we're almost 20.

>> And um that has been that it's honestly >> have a huge consulting business.

>> We have a huge consulting business that's not even counted in the MR >> and that's honestly been um it's been so fun.

>> Yeah. because I think we're really figuring out what it means to be an AI first business ourselves and then we get to bring that to readers to like people use our products because we build tools

for ourselves that we can sell to sell to subscribers um and and bring it to big companies in the consulting business and it's just like it's the most exciting thing I've ever done.

>> What's what's one word to describe 2025?

>> Uh the word that came to my head was fire. [laughter] I don't know.

fire. [laughter] I don't know.

>> In a good way or is that like >> in a good way? A good good fire. Like we

start we like lit the engines, right?

And so like >> the engines are lit.

>> Liftoff. Let's say liftoff is is the word.

>> I don't even know. Yeah, I guess so.

>> Yeah. I really think we've like I don't even think the engine is on. I think

we've like primed the engine [clears throat] >> and 2026 is going to be where we really like, >> you know, in the grand scheme of every we're just going to be putting it into first. We're going to think we're

first. We're going to think we're putting it into, you know, fifth. That's

the thing about these curves is every time you hit a new inflection point, the old curve looks really small and shitty.

And um so I think right now the one we're on, we're like this is amazing.

But I definitely agree in 2026. The

thing that we're going to do is >> every has been this very bottoms up place for a long time where we're doing the newsletter and we're writing whatever we're interested in. We're

building tools and we're just sort of building whatever tools come to our head that we kind of need ourselves. And I

think we're starting to um build a more much more cohesive platform. So um we brought in someone uh really talented, his name is Willie who is build building

the every platform where we have um we have single sign on so you can log into any one of our apps. We have now we have unified billing so you don't have to enter your credit card information for each one of the apps. But that's sort of

like the groundwork. What we're doing next year is um we're going to build this like unified memory and context layer where we have um we have a bunch

of uh basically we can collect everything everything that you care about and who you are into one place and all of the apps and all the content can take that into account as they work with you. So it's like a it's a much more

you. So it's like a it's a much more custom personalized AI experience across all the different surfaces. I think the [snorts] biggest one of the biggest challenges for us next year will be

>> I describe every as like the most genuine work environment that I know of.

Yeah.

>> The things that we build are the things that like any one of us is like this is interesting. I need it. It's cool. Yeah.

interesting. I need it. It's cool. Yeah.

>> And then we build it.

>> And that has like served us really well.

We get to build cool things but then write about it and and we build things that we need.

>> Uh and we're going to like start to have like a real strategy. Yeah.

>> Next year, which like we haven't had, but we've been doing it like >> based off of what's genuinely interesting.

>> And it's going to be I think one of the big challenges will be like continually being genuine and also like having a strategy and having personas of the types of people that we want to like attract towards every

>> and not losing that like genuine motivation that we have to just like make cool [ __ ] >> I agree. What was like for you like so this is actually this is the first time

I've run a business with 20 people.

>> Um but you've run a business your last business you got to about like 100 employees I think. Yeah.

>> What was the transition like from like 20 to 50?

>> You have to rebuild the business every time the number of people doubles. And

we felt that. I mean we've we've done that a little bit from like 5 to 10 and now we're at 20 and like you know we just went through and changed everything.

>> So every time the business doubles it changes. you have to rethink

changes. you have to rethink operationally how everything works and just >> sometimes the whole strategy and then you just sort of need to have

like people do their job really really well and when you're smaller everybody's sort of doing everything because you have one person >> doing support and one person doing

>> design and then as you scale like >> uh what I mean by that is like sometimes you need the design person to also help with support.

>> I see. And um and then as you scale like people start to just do very specific things.

>> Um >> and I found it that like it's a real challenge to keep the culture going where everyone feels like equally as excited about the company and like they

play a very big role in it because >> you end up just hiring some people where they're just like >> I just do support for this company, >> right?

>> You know.

>> Yeah. Yeah. Um, and so keeping that genuine interest in building stuff and exploring things is what makes us special and will be the the big challenge as we scale.

>> I totally agree.

>> But I I don't really think we're going to like I think we'll hire next year, but I don't think we're going to hire a ton. Like the whole thing is like keep

ton. Like the whole thing is like keep it small. I hope so.

it small. I hope so.

>> Yeah. I hope so. Cuz I love just >> it's amazing. I love all the people that we work with and every time I go into a meeting I'm like, "Wow, all these people are so cool and interesting and have interesting backgrounds and stuff to say and I really don't want that to change

>> and I don't think you need to anymore because AI is allowing us to scale the same number of people essentially."

>> Yeah. Um, speaking of which, uh, let's get into our 2026 predictions. I know

you have some doomer takes, which >> I got a couple doomer takes, >> which is good because I'm I'm usually like pretty sunny and optimistic, so it'll be a good good balance. Uh Rachel,

our producer, said um we're going to do a good cop bad cop routine. So do you want to start with >> I don't want only want to do bad.

[laughter] >> Um I'll I'll start with a positive take and and this is actually inspired by a conversation that we had which is that

>> agentic development will be maybe by the end of the year inside of the app that you're actually using.

>> You said that to me and it like completely blew my mind.

>> Yeah.

>> Um and we're already doing that. So what

I what I mean by that is like uh the inside of every right now when we build a product if something isn't working you can DM Dan's building a tool that does

this right now you can DM this tool or mention it in a uh in a Discord channel and say hey I want this feature or this bug is broken and it will um write some

code and then submit a PR um for it. So,

it's sort of like working inside of Discord but the next step of this is that I can do it inside of the app itself and then it can like write the code in a way that

doesn't break it and then actually like relaunch the app itself with a feature that I might specifically want >> and it just feels really [ __ ] hard to do but kind of inevitable.

>> Yeah. No, I agree. I think like the the way that I've been I've been talking about it and thinking about it is we reached this place at the end of 2025

where um coding agents are really good and mostly the code they write works which is was not the case last year. it

was um you could you could get pretty far but eventually you get stuck into a bunch of errors and bugs and so if you're not a professional developer it would like just break and now you can kind of with Opus especially with Opus

45 you can kind of like keep going forever almost especially if you're using stuff like our compound engineering plugin and you're you're actually like having it plan and think and all that kind of stuff it's like

it's really good so I think that ch that fundamentally changes software engineering um and it also fundamentally changes how apps are built and who they are built by. And so I've been kind of

playing with this new idea for how software is architected that is very inspired by cloud code that I've been calling um agent native architectures.

>> And in an agent native architecture, the like first level of of an agent native architecture is anything inside of an app that the user can do, the agent can do. So in cloud code anything you can do

do. So in cloud code anything you can do in cloud code where you're like oh I want to uh please go you know uh or in cloud code it's anything you can do on your computer cloud can do. So like if

you want to change files or um you know write write a file or run a server or whatever cla can do all those things and mostly most of the agents that we use right now that's not the case. So, if

you go into um for example, one of our apps, Kora, uh the the agent that you can talk to, there's many things you can do inside of the Kora app that the agent can't do itself. Um

uh and same thing with a lot of other like sort of chat agent like chat apps basically. And so I think the first step

basically. And so I think the first step of agent native development is anything you can do the agent can do. And that

gives um this like power and flexibility to the application because um you can just tell the agent to go do stuff. So

like you know, oh I want to change my settings, right? I've been doing that

settings, right? I've been doing that all the time now with with the Atlas browser. It makes kind of makes every

browser. It makes kind of makes every app a little bit agent native. So if I I get requests all the time where people are like, "Oh my god, like can you update this like setting or like add me to this team or whatever?" And I'm like, "I don't want to figure out this

interface." And I just ask Atlas to do

interface." And I just ask Atlas to do it. I think every app like every agent

it. I think every app like every agent native app will have that. That's the

first level. Second level is any code in the app can do because there's many pieces of code that the user cannot um trigger themselves. Any piece of code

that um the app can do the agent can do.

So that's a lot of like so for example in Kora users can't really like generate their briefs and briefs are the summaries of their emails. Um uh

eventually what should happen is the agent should have the ability to like regenerate and redo an entire entire brief an entire summary in whatever way they they want.

>> Maybe triggered by the user, maybe triggered by the developer. The third

one is yeah anything a developer can do um the user uh the agent can do. So the

user can be like I want to customize my software in this way. the agent's going to go submit it and maybe it's a bug fix that goes out to everyone or maybe um there's maybe there's some system by which you can um

uh you can just have your own custom version of any app >> and we're already starting to do this inside of every and it seems like it's sort of starting to work and I just I think that that's going to be broadly distributed like especially that first level is going to be a broadly

distributed paradigm in a year that no one's really talking about right now except for except for anthropic I think anthropic has nailed this but everyone else is going to be doing it >> I think notion is doing that pretty well too. Yeah, I mean you can you can use

too. Yeah, I mean you can you can use the notion AI to like do to >> create almost anything that you would create in notion.

>> Yes, there I think they also get it.

They're like really thinking about this in the right way where um it's not just software that's made for humans to collaborate that they've like pasted agents into. It's like actually agents

agents into. It's like actually agents and humans are like first class citizens in the app.

>> Yeah.

>> Um let's keep going.

What's my what's my prediction? Let me

see. I have a note file. Okay. Uh oh,

this is a good one. Um

I think that designers are the next big um superheroes of the AI age. like they

have been uniquely enabled in this world because >> um forever designers have what they're good at is they have a lots of taste and they have like a vision for how they want an app to work and they've just

bumped up against like well I have to convince this developer to do this for me um for a long time and I just see so a good example is Lucas our creative uh

our creative director inside of every um he has turned into this machine where he's not only like this incredibly creative designer, art director kind of

person, he's also like vibe coding little apps that let him do his work better. And I just think there's like

better. And I just think there's like there's a whole class of people that are sort of like Lucas where they're these highly creative, highly visual people who have been held back from making full

experiences because they can't code that now are going to be able to code. And

the stuff they can make um because code is cheap. um being able to make a really

is cheap. um being able to make a really beautiful really um evocative well thought through experience is going to be a huge deal and I think all these designers are just going to like start

going crazy and it's going to become this new archetype of like really interesting builder.

>> Cursor I mean I think is already inching towards that. You can do you can you can

towards that. You can do you can you can do front-end updates inside of cursor with like a visual medium. I I sort of think I get a little bit worried that

[snorts] like um so I I used to be afraid of the terminal and like a lot of people are afraid of the terminal like it sounds like a scary word.

>> Um and I kind of wonder if like doing design inside of cursor >> a lot [clears throat] of designers will be like no like that's scary to me like I can't touch the code I don't want to

mess anything up. So I wonder how that will play out um next year so that like you know the best designers in the world will be doing this but like you need to

build a product that like every designer can you know can access. So

>> um I wonder how they'll abstract it the cursor you know if they want to expand their market away from just >> engineers to designers like they might need to abstract it away from the code

so that like the average person doesn't get scared like going near code.

Um, what's next? I know I think >> um, okay, I have one.

>> You have a hot take coming.

>> I have a bit of a doomer take, which is that I am really worried that Okay, I So like I just have a bunch of friends, my mom, let's say my mom, who's also my

friend. Um, [laughter]

friend. Um, [laughter] if she sees something on Instagram and it's I'll tell you the exact story I'm thinking of. She sent me a video of some

thinking of. She sent me a video of some gorillas that were taking care of their children.

>> And the point of the video was like, "Wow, gorillas, they're so humanike.

Look at the way they're taking care of their kids."

their kids." >> And it looked totally real. My mom sent it to me and was like, "This is so cute." And

I was like, "Mom, that's just not real.

That's AI." And she refused to believe me. She was like, "That's,

me. She was like, "That's, >> you know, what are you talking about?

This is real." Yeah.

>> And I just think it's really scary that a lot of these social media companies, all of them, which are heavily invested in AI working, >> aren't made by AI, um, you know, as a

thing.

>> And I just think that that is going to be something that will have to happen, >> if not by the midterms, than by um the next election year, presidential election.

>> Mhm. But yeah, I think midterms are going to be like kind of scary. Um

because I think there's going to be deep fakes, which used to be a really scary word and now nobody talks about nobody says deep fake anymore.

>> So what is the like specific? So I agree with you. I think this is a huge problem

with you. I think this is a huge problem and I think it's right to point it out and that example of your mom like actually not believing it's AI is really interesting to me because for me I look at it and I'm like I I can tell but I

know that that's I look at this [ __ ] all day. So that makes sense.

day. So that makes sense.

>> Um, what's like the specific doom or take? Like what do you think?

take? Like what do you think?

>> The doom or take is that shit's going to go bad during an election cycle and that's going to be the inspiration.

>> But the really the really doomer take is that it's going to go well for the team that, you know, then has to make the rule to not >> I see, >> you know, to label stuff.

>> And what specifically do you think is going to happen? Is it going to be um like uh uh they're they come up with a video of a politician like getting caught, you know, doing something that

they do. Is that it's like the P tape

they do. Is that it's like the P tape [clears throat] but like real.

>> Yes. Yes.

>> I think that I think that scary stuff is going to happen where fake videos are made like deep, you know, the nature of the word deep fake came from like people doing like really grotesque things that

like they didn't actually do, but AI made it look like they they did that.

And um I think that like we are just like so there's just like a thin wall that's blocking the average person from making videos like that.

>> Yeah.

>> And if you really want to, you can like I've never tried, but you probably could easily go make like really messed up videos.

>> And uh that is going to happen 100%. So, I

think like that's my doom or take is something's going to happen with elections that like that's that's that's what will trigger um you know some sort

of label that's like made with AI. I

also think another interesting thing that's happening is this deal that Disney just did with uh with OpenAI to put Disney characters into into Sora

>> which is really just going to like sort of ch train all of us to expect it's a little bit confusing because Disney characters many of them aren't real anyway.

>> Yeah, [laughter] >> mostly mostly they're not.

>> Mostly they're not real. Uh but like we're just going to be like really used to like not real stuff soon.

>> Yeah. Um okay let's keep going. So um

my next prediction is I have been observing this split in software

engineering. Um and it's uh I think

engineering. Um and it's uh I think there's like now basically three types of software engineers. Uh I mean there's four types. One type is just traditional

four types. One type is just traditional engineer never use AI which I think is getting less and less but like they're definitely still a thing. It's

definitely they're definitely still there are super skeptical all that kind of stuff. But the more interesting thing

of stuff. But the more interesting thing are the are the other three I think um uh mostly what people are talking about

and think about are um two groups. One

is uh traditional engineers who use AI in their process and they're kind of they're using cursor, they're probably using cloud code, but like really they they still think like traditional engineers um but they're using AI as

this sort of like accelerant to their their normal engineering process. So

they're still reading code, maybe they're still writing it, but like um they're they're deeply involved in every aspect of the software. So that's one.

Second one um is vibe coders who are at the opposite end of the spectrum which is um I don't even know how to code. I'm

just like making something that I don't understand. Yeah. And I think that that

understand. Yeah. And I think that that that's like the the classic dichotomy between the two types of coding that's happening. But I think there's a third

happening. But I think there's a third way that I think we're starting to see inside of every and is really um I think Enthropic is really leading the way on this and OpenAI has not yet really

caught up to it. I think OpenAI is really right now for a lot of their coding tools is serving um the like traditional engineer who's incorporating AI. The the third way is what we would

AI. The the third way is what we would probably call compound engineers or um maybe more broadly as agentic engineers who are [clears throat] engineers that

are rein have reinvented software engineering as a skill for an agent world. Um where they are not looking at

world. Um where they are not looking at the code, they are uh not ever writing any code. um they are fully committed to

any code. um they are fully committed to delegating all of the actual programming work to um to the AI to an AI agent and they're like moving up a level in the

stack to essentially manage a single or multiple agents all at once and they've um sort of they've decided to make that tradeoff of because I think a a lot of

programmers and I understand this like a lot of programmers love writing code and they've decided to make that trade-off of like getting rid of this old skill which has I think for them a lot of them they feel like it has started to atrophy

in in exchange for this other skill and I think many traditional engineers who are using AI don't want to do that because they love writing code um and I think many companies are not aware that

um they have a choice here because they're trying to serve if you're if you're trying to sell enterprise software contracts right now you're going to run into a lot of traditional engineers who want to uh want to

incorporate AI into the process but not reinvent it and so I think a lot of companies are missing this group that I think we're a part of and a a couple other people are where um it's small companies that are reinventing what

software engineering is from the ground up and that will be its own scale. It's

it's not vibe coding. So, and I I would say vi coding is like very amateur-ish.

Um and it but it's definitely not engineering with AI. It's like a it's a third thing and I think that's going to be a big deal in in in 2026.

>> And what's going to happen to all of those engineers that refuse to get on board? Um

board? Um uh I I think it's a good question. My my

sense is that this kind of thing takes a while to move through an entire economy.

>> Mh.

>> And so I think for the most part if you are an engineer who wants to incorporate AI into your process but doesn't want to reinvent the skill of software engineering, there's going to be a home and a job for you for a very long time

actually.

>> For sure. Um and I think if you if you then zoom out though that kind that style of engineering is going to be on a pretty slow and steady decline. Yeah. Um

in the same way that I don't know like the newspaper industry was affected by the internet. It wasn't like all at once

the internet. It wasn't like all at once but like over time it's become like less less and less uh right >> less and less profitable as a as a as a business uh for specifically for like

the old cost structures of newspapers.

Um, and if you look at the growth, the growth will be in this new kind of engineering in this new kind of engineering work. And so if you're, for

engineering work. And so if you're, for example, a new grad and you're choosing between what what kind of stuff you're going to do, I think you're probably going to be more likely to to choose this third path.

>> But it will be interesting like I wonder how many schools are still like I studied architecture and they still had us uh draw by hand. Yeah. Which was like a good skill to have but so useless in

the real world. So like we all had to teach ourselves Revit and AutoCAD.

>> Yeah.

>> So [snorts] I do wonder like how these schools are going to >> um >> adjust. It's going to be hard to

>> adjust. It's going to be hard to >> and hopefully you know that's why we do all this all these courses and training at every and we have a lot of like student discounts and stuff like that and it's not a replacement for any of that but um I I think there are a lot of

people will pop up to like serve that need for kids that are not being served by universities.

>> And also the best engineers that at least we've seen are also very very good traditional engineers. just adopted this

traditional engineers. just adopted this new way of building.

>> Totally. Yeah. Like Naveen, who runs Monologue, is a good example of that.

>> Yeah.

>> All of them. All of them are. Yeah. You

know, >> they really are.

>> So, what about um what do we think is going to happen with all of our non-technical friends out there that use JGBT and like use it to like pull up recipes, but don't really use it at

work? Maybe they like drop in a CSV and

work? Maybe they like drop in a CSV and like do some analysis, but like they don't actually. Will 2026 be the year

don't actually. Will 2026 be the year that like AI is just something that we as non-technical people natively use inside of products or are we still going

to be sort of like in this world where I mostly just use it to like make turn pictures into like like drawings for my kid to use

>> um because the and and I have I mean I got a bunch of friends. just had dinner with a friend yesterday >> who was like, "Yeah, like this AI thing, like it's just not working for me."

>> And I was like, "Well, you're not an engineer. Like that we know for sure

engineer. Like that we know for sure that is a thing."

>> Um, >> who was this person? Like where do they work or what do they do?

>> They run uh the deal desk at Twilio. So

they work at a really big publish company.

>> Um they do a lot of math >> or they have Excel that does a lot of math.

>> What's a deal desk? Deal desk is like um if a salesperson comes to you and is like I have a big deal but they want a discount. The deal desk at a big

discount. The deal desk at a big company's like tell them no yes they can do 10%.

>> Um >> so you know it's a lot of Excel work. Uh

and I it's probably more sophisticated than that if you're listening to this podcast Graham so I apologize.

[laughter] Um, I uh I I just wonder if next year cuz it feels inevitable like will we leapfrog to like

the point where Graham no longer has a job or will we steadily get there because uh he starts using AI a ton himself?

>> That's a good question. I

um Okay, here's here's you're you're opening up some like interesting threads for 2026. one is I think we will

for 2026. one is I think we will definitely see um layoffs that companies blame on AI um in 2026 and I think mostly they will not be actually AI related they will be about the

underlying business but they'll use AI as a good excuse. Mhm.

>> Um I think generally people like Graham and I don't know really that much about his job but um I would guess that they will find it valuable this year. Um

specifically because coding agents are good enough that once they make their way into Excel in a widely distributed way he'll be like holy [ __ ] I can't believe >> it needs to be in his place of work.

Exactly.

>> It's not really in his place of work yet.

>> Yeah. Um,

and uh I also think there's there are always these uh um tailwinds and headwinds. So like there's there's

headwinds. So like there's there's always sometimes it's like wow AI is like blowing up and everything we're improving so fast. And I'm we're starting to see right now this like oh is it improving as fast as we thought it

would? Mhm.

would? Mhm.

>> And I think um um we've started to see over the last year or so the the big AI CEOs so like Daario and Sam uh Alman

in 2024 said that 2025 would be the year of agents and it really was like we got we got there. We did it. Um but they've also been saying like AGI is like very uh it's it's I think Sam said like it's

less than a thousand days away maybe like a year ago.

in two years and now their timelines have shifted.

>> Yeah.

>> And so I think there's this really interesting thing happening where um two things are true and that tends to get lost which is

one I think AI progress is still really fast and it's uh it's it's kind of mind-blowing where where we are now versus where we were a year ago. And I

think that will be the case in at the end of 2026. I think we'll be sitting here being like, "Wow, it's like actually a lot better." And the world and AGI, what, however you define that, is a lot more complicated and hard than we thought.

>> Yeah.

>> Um and uh and so and we will not have AGI in 2027.

>> Also, what even is it? Like what even is AGI? I mean, I just remember a couple

AGI? I mean, I just remember a couple years ago I was like, "Yeah, AGI is a thing. It's when like they the robots

thing. It's when like they the robots like they know everything and they can talk to you." And now I'm like, "What was I thinking?" Like I would say we have AGI right now. It's just that like >> it needs to be able to continue to live

on and run and like propagate inside of my computer and do things that like I don't have to request that it does. Like

the models can do that right now.

>> Like cloud code can run for hours to do [ __ ] and you know fix problems that it makes. Um it's just I feel like it

makes. Um it's just I feel like it doesn't have the tools to like live on indefinitely. Um and like never really

indefinitely. Um and like never really turn off. To me, AGI is like a model

turn off. To me, AGI is like a model that never actually really turns off.

The conversation might just be the fact that like I have to go to a window in a browser to like engage with it. The

conversation never really ends. Yeah.

>> That to that's you know Brandon's version of AGI.

>> Yeah. Yeah. That's that's that's been my definition of it too is like um it uh uh you if you uh think about it from the perspective of human or childhood

development. Um as child as children get

development. Um as child as children get older they get more and more autonomous on a fairly linear curve. Um I don't know if it's linear but it's it's a predictable curve and you can like leave them alone for like if they're young you can leave them alone for like five

minutes um with a monitor. Um, but as they get older, they have more and more time um, where they're they can go play in their room for 10 minutes and nothing's going to like they're not going to die. Um, and I think AI has

been very similar. Like three years ago, Chadbt is like one turn. Now Cloud Code and and um, you know, CHT Pro like can run for 20 minutes to like an hour at a

time. Um, but that's actually quite far

time. Um, but that's actually quite far away from they just run indefinitely.

And it's worth it. it's economically

profitable and worthwhile for you to pay for it to be constantly doing stuff without you like always telling it here's what here's what to do. Um, and I think we're I think that's actually a

very complicated problem because it requires continuous learning. Um, it

requires the agent to have like a good sense of what its goals are and also to be able to modify its goals in ways that make sense. And these agents have these

make sense. And these agents have these little quirks in the way that they learn and the and their common sense that um start to add up over time. And so what they would do diverges a lot from what

what >> Yeah.

>> I would do basically or any human would do >> and it kind of like it starts with a seedling and like an hour later they're like the world.

>> Exactly. Um and so I think we're getting better and better at that. But there's

all this stuff that um you know we we for example we do a lot of um alignment training which is like essentially making the model be predictable and um

to do exactly what you say and um and flip side of that is like sick fency where it's it's just like telling you what you want to hear. And I think we we're gonna have to get into a mode where um

they need to have their own sense of like agency and almost like individual personhood in order to have more in order to be skilled at being autonomous because you have to like go let them play around and mess up um in ways that I think we've been hesitant to let them

do for like safety reasons. So I think that's going to actually happen a lot more in in 2026 is is we're going to um use training approaches and architectures that allow them to like go

off and do stuff. Um, that's a very vague prediction, but um, that's that's what I got. So, we're close to time.

Rachel, I just want to um, just check in with you, see how we're doing, see if like there's anywhere else you want us to go before we close.

>> Awesome. We are good to go.

>> Sweet. Well, um, happy new year. We'll

see you next year.

>> See you everyone.

[music] >> Oh my gosh, folks. You absolutely

positively have to smash that like button and subscribe to AI and I. Why?

Because this show is the epitome of awesomeness. It's like finding a

awesomeness. It's like finding a treasure chest in your backyard, but instead of gold, it's filled with pure unadulterated knowledge bombs about chat GPT. Every episode is a roller coaster

GPT. Every episode is a roller coaster of emotions, insights, and laughter that will leave you on the edge of your seat, craving for more. It's not just a show.

It's a journey into the future with Dan Shipper as the captain of the spaceship.

[laughter] So, do yourself a favor, hit like, smash subscribe, and strap in for the ride of your life. And now, without any further

your life. And now, without any further ado, let me just say, Dan, I'm absolutely hopelessly in love with you.