The Company Brain—Turning Institutional Knowledge Into an AI OS

Julia Nimchinski:

Yeah, it’s an amazing panel, thank you so much. The discussion got a little bit heated up in our Slack, so, it’s an amazing transition. Jay Hack, head of AI at ClickUp, welcome. Jay also is a founder of, previously CodeGen, and he spent some time at Palantir. So, what’s your take on, for deploy? Jay? Is it children?

Jay Hack:

Pleasure to be here. Thank you so much. having me. I hope I’m not frozen. Am I coming through a little clearer now?

Julia Nimchinski:

Live TV’s gonna, you know.

Jay Hack:

Yes and no.

Julia Nimchinski:

Can you say something?

Jay Hack:

The mute.

Julia Nimchinski:

Yeah.

Jay Hack:

Okay, I might, I might just switch Wall-Fi,

Julia Nimchinski:

Jason Napieralski. Welcome back! Are you frozen too?

Jay Hack:

How’s that… does that work?

Julia Nimchinski:

Perfect, yeah.

Jay Hack:

Okay, much better. I’m sorry, I lost you, you asked a question there, and then I… Lost context on it. Throw it at me again.

Julia Nimchinski:

I feel like, yeah, everyone loves me. I mean, Jason, are you… are you here? We just had an amazing panel on for Redeployed. kind of debate, is it the only way to see AI ROI? So, curious your take coming from Palantir.

Jay Hack:

Sure, yeah, I mean, I worked at Palace here over a decade ago doing exactly that, essentially for deployed, AI. At the time, we called it machine learning. The joke has always been that if you’re doing data science, it’s an R, If you’re doing machine learning, it’s in Python, and if you’re doing AI, it’s in PowerPoint. And that was sort of to point out the fact that what we used to call AI was something that was very hand-wavy, but there actually is something that’s, you know, very brass-tacks, practically useful, which is developing predictive models that you can deploy in your business.

I do not think that’s… In 2026, the only way to be successful with AI is through forward deployed. I mean, look no further than the fact that you can open up a Claude window, or a ClickUp Brain window, or whatever, and you can ask the questions and be able to tell you things that you otherwise never would have been able to access those answers. So, there’s very much so a clear path for individuals to be able to benefit from AI. I do think, though, however, if you’re trying to squeeze the most juice out of the AI lemon, especially in a very large organization, it does require dedicated effort to go through the technological transformation to get you ready for that.

And so, that is not just, you know, setting up AI systems that consume the right information, it’s also probably reorganizing the way in which your communication channels flow, or the way in which you do management structure, because Lord knows that certain middle managers are no longer useful. So, I would say it is helpful, but not… it is helpful, but not necessary. in order to be productive with AI.

Julia Nimchinski:

Jay, I can’t wait to dig into this. We have Jason Napieralski, CEO of DUX Experts, former AWS, Google, Oracle Executive, HSC Regular. Welcome back, Jason, and take it away.

Jason Napieralski:

Thank you, great for… thanks for having me, great to be back, great to, hear that previous discussion. I think that, one of the things that I’ve noticed from all these discussions is that, the future is as clear as mud, and that, everybody has a different take, and everybody has a different opinion, and I think that comes from where, you know, where you come from. is always where you generate your best opinions, I think. So, Jay and I are, have had a brief conversation And to that end, Jay and I disagree with… about a lot of things related to AI, which I think is part and parcel to this conversation, and, quite frankly, the way that we move forward and we kind of understand it and work together.

And I think that having this forum Where we share a lot of different ideas, and a lot of people have their minds open to different ways of thinking is the way that we will kind of move forward together. as a society, much less an industry and everything else, but this forum is very important for that regard. So, to this end, what we’re going to talk about in our little short fireside here is We’re really lucky to have Jay here. Jay is an extremely intelligent and articulate thinker. He has thought a lot about AI, he’s been in the trenches.

He is a Silicon Valley insider. And, has a lot of perspective that I think it’s very valuable, so let’s get right to that. I do… I want to piggyback on the last session, and we were talking about… they were talking about data, and they were saying data is the moat, moat, data, get more data, better data, and that’s kind of what we want to talk about here. But we want to go into it a little bit more in depth, on the, personal or private data or company brains mechanic here. And that… and that we want to kind of talk about, practically, how do you get your, AI To understand you, your company. fundamentally, at a micron level.

And how do you do it safely? How do you do it practically? And that’s… that’s what we want to talk about today. So I will, you know, the first question I’ll have for Jay is that, Jay, how important do you think that personalized data is to getting the, as you said, the most juice out of the AI lemon.

Jay Hack:

tantamap. I mean, it’s the most important thing, obviously. Hopefully, I can offer some, you know, not just practical, but also theoretical AI insights in the session today, and one thing I’d like to reference is the so-called bitter lesson. This is the idea, going all the way back to the inception of AI in the 60s, that the only two things that really matter in building better AI systems are more data and more compute, or more time spent searching through the space of possible solutions. what an AI could end up producing.

It’s always been the case that, you know, sort of in any given wave of AI, you have a bunch of people show up, and they have a bunch of clever little techniques that they put in place in order to make an AI better at doing something like playing chess, so humans will write heuristics for, this is how you should move across the chessboard. And then, within a couple years, basically, that paradigm is completely washed out by people having bigger computers and adding more data to the equation of how you, you know, actually train the thing and get it to to play chess.

And famously, this, you know, AlphaZero, which is a DeepMind production, ended up just blowing out of the water every other chess-playing machine, because it just trained against itself playing chess for trillions of iterations, essentially. And so I think that there is no exception to that today. I mean, this is sort of the foundation of all of the large foundation models that we have today, is they’ve just been trained on massive amounts of data, and that applies in the organization as well. Where, at this point, it’s not just going to be the case that you doing little clever tricks on the way in which you prompt the AI, you know, or give it instructions or something is going to make it that much better.

The bottleneck to it being more productive in your organization is giving it more information about what your organization needs. past behaviors, etc. I think that the best way to go about doing this, giving your AI the right set of context, is, you know, this is a tale as old as time, data integration is very difficult, this is what we used to do at Palantir. chances are, if you’re in a legacy organization, you have your data spread across many different siloed locations.

So, you know, you have your ERP, you’ve got your CRM, you’ve got a bunch of information in Slack, and unfortunately, it is the case today that most of these parties independently have an incentive to keep their data siloed, because the moment that you can exfiltrate that and put it into your AI, then maybe your service is no longer very valuable, and Lord knows that a bunch of these folks are not actually very good software, and many people would love to get off of them in the first place. Nowhere is this more apparent than Slack, I would say.

Slack, for a long time, has had a very locked-down MCP, and that is very unfortunate, because most of the decisions and conversations that your organization is having that would be useful for an AI to make decisions on top of are going to be located in there. And so I think that the best solution to this, the thing that people should do going forward, is basically adopt a centralized location where all of their company’s information lives, and make that directly accessible to an AI. And obviously, I’m talking my book here. I work for a company called ClickUp.

ClickUp is a horizontal software We bring, like, 9 different applications all in one. We’ve got Docs, we’ve got chats, we’ve got tasks, we’ve got spreadsheets, all this other place that lives in a single horizontal and verticalized platform. Which means that we don’t need to ask anybody’s permission if we want to give access to this context to an AI. It all just lives in a single data asset that is natively accessible to the AI, and we’re able to put together optimized prompts, optimized harnesses on top of that, etc.

And so, you find yourself in a situation, if you’re using ClickUp, to run your company’s operations, where You basically have the entire history of your company’s decisions, conversations, chats, tasks, etc. in one place, the AI can directly access it, and boom, you’ve got something that’s much more capable of actually delivering business outcomes for you.

Jason Napieralski:

So, Jay, let me push back a little bit, and when… when people think, I am giving my… they… they think internally, I’m giving my data to the AI, And that the AI is going to take my data and give it to somebody else and ruin my business model, right? That’s one thought. The other thought is, and we discussed this too, about Fable. So Fable changed their ZDR policy and said they’re going to keep that. They’re going to keep the prompts and review that and be able to act on things that they don’t agree with from a, you know, safety perspective.

So, how do you address this problem? Because I think this is core, and part and partial to a lot of people, I think, hamstringing themselves from AI. And I’ll give just a personal example. I use Gemini on my Android phone, and my wife has the same. I let Gemini remember everything that I do, and then Gemini provides useful information to me at times when I won’t. I don’t, I don’t expect it. So, for example, I tell them I’m on a lower-carb diet, and so when I ask for a recipe, it always automatically adjusts or tells me that it adjusts for low carbs.

My wife is the opposite approach. My wife is a full tinfoil hat-wearing member of the Libertarian Party, and believes that all data is to be used against her at a later date, so she uses the private mode on Gemini and has no context in there. And as a result. every prompt she gives is a mystery box, right? Of what she’s gonna get. It’s completely different in tone, tenor, it’s completely different in information, it chooses on its own what’s relevant for her, what’s not. Versus in mine, it kind of has figured out what’s relevant and what’s not.

So, there’s… there’s that piece. Now, my wife is saying I’m an idiot, because now Google has all my information, and then they’re gonna… when the death bots come, they’re gonna come for me first, because they have… they know where I am, and they know all those types of things. So, obviously, that’s a reducto ad absurdum, but how do you address that, or how do you make people feel like this is something that is actually not only worthwhile, but, I would argue, existential?

Jay Hack:

Yeah, I think this is a… we’ve sort of supercharged the debate on data retention and data storage. Obviously, it’s always been the case that if you’re using Salesforce, Salesforce hosts that data, and it could be the case that Salesforce is going to take your CRM and then go and replicate your entire business, and that has never been more possible than today, when they have access to AI models, especially if they’re seeing your AI chats. I tend to fall on the side of the bait, which is, I think. you can largely trust these organizations, especially if you sign a contract with them that gives you ZDR, which ClickUp has done, for example.

So, you know, any chat that a user has within ClickUp or one of our customers, if it ends up hitting an Anthropic model, that data is essentially immediately deleted. There’s nothing that I can say that would convince people there isn’t some nefarious employee at Anthropic or OpenAI who’s, you know, hacked in and able to view my chat logs. And actually, this is right now a debate that is spilling out into the public. Because there was a famous math problem, Navier-Stokes, it’s one of the Millennium Prize problems that was solved. by OpenAI very recently, and some folks accuse them, specifically an Anthropic employee has accused them, of using his chat logs, where he was making progress on that same math problem, in order to train their AI in order to be better at actually solving that problem.

And so, there are potentially scenarios where, you know, there’s some blowback to using a third-party model provider. However, my position on the matter is the risk of that is relatively low, especially compared to the benefits of using these sort of closed source, you know, foundation model labs. That being said, if you do take a position that your wife does, which is that you know, we’re, headed towards a libertarian paradise where nobody will have access over your data. You either should get a new crypto, which I highly recommend, fully homomorphic encryption is a great use case for this.

Alternatively, you can basically use open source models, and run them on your own infrastructure. And so, we have never been in a scenario where the gap between the best closed source model and the best open source model is tighter. There was a model released this morning, DeepSeek V4.1 Flash, which is honestly incredible. I haven’t used it personally yet, but it seems as though, according to all the benchmarks, it’s doing incredibly well. If you care specifically about coding, there are models that are very good at that that you can run on your own infrastructure. And that relieves you from the constraint of needing to ship your data to some external party.

You can actually keep it all on your own AWS stack, and so you take a pretty minimal hit in terms of capabilities if you want to control everything yourself, and I think that is a good thing for the world until, the last point I’ll make. we get to a point where the capabilities of these things are actually scary and require some level of, you know, federal or, you know, regulatory oversight. And I do think we’re headed towards a scenario where that’s the case. simple thought experiment. Let’s say we had an open source model that was capable of actually producing a bioweapon.

There are literally services where you can hit an API and it will mail you a package that includes a bunch of custom synthesized proteins. I don’t think it’s crazy that somebody would be able to take an open source model, run it on their own infrastructure, create a bioweapon, and then create, you know, chaos throughout the world. And if that is the case, then I am firmly in favor of organizations like Anthropic basically having data retention policies in place, specifically for just performing oversight and making sure that nobody is abusing their models.

Jason Napieralski:

Got it. And one of the fears, I think, is an irrational one, in that, that your business model or idea is so unique. that it doesn’t exist in nature, and if the AI finds it, then they’re going to teach everybody that thing. And I think that one of the things that a CEO once told me at Amazon, that I could give you a million dollar idea in an envelope and hand it to you. And I could do that every day, and it’s not worth the paper it’s printed on. It’s about execution, it’s about distribution, it’s about… capital, and that is what takes a business to be run, not the model.

I would argue that business models have been known. that basically everybody knows how to do the business. The question is, how do you execute it? How do you market it? How do you distribute it? And how do you make money from it? And I think that that is the part that can’t be stolen from Anthropic. Like, they can’t steal your execution engine.

They can build parts and pieces, but that there is a… I think an unreasonable fear that what you’re doing is so unique, and it may not be in every case, but for the most cases, business is pretty simple. and your particular business model, and I’ve heard… and the reason I’m saying this, I’ve heard this from people that run medical offices, I’ve heard this from people that run landscaping businesses, I’ve heard this from plumbing and HVAC businesses, that they are afraid that their data will be stolen. But really, there’s no market… in order for something to be stolen, there needs to be a market for that data.

And if there is no market, then the theft is irrelevant, right? Like, if somebody wants to steal blades of grass from your yard, you’re not gonna call… you’re not gonna have a big problem, because there’s no market for single blades of grass. That’s kind of one thought. What is your thought on that?

Jay Hack:

I think it’s fairly unlikely that OpenAI gets into landscaping anytime soon. I mean, there is a version of this in the future where humanoid robots enter the equation, and it ends up being the case they can dispatch them to come to your house, and you prompt ChatGBT and say, hey, my grass is getting kind of long, can you help me figure out the best way to do it? And you get the best landscaping, you, you know. ever received. I would consider that a tail risk at this point. You know, it’s not something that we need to be dedicating a whole lot of time to.

I think the thing that has really changed over the last couple years, especially with the advent of coding agents, is the things that used to be scarce in business building are no longer scarce. So, I do think, for example, if there was a million-dollar idea landing in my lap every day, and it was something Where there was no scarcity in producing it, aka I could just entirely code the business from scratch. It’s like a SaaS service that’s like an email something something, or dropshipping between Spotify and Amazon. Shopify and Amazon, rather. I think that that actually would be quite valuable.

If it was an idea that nobody else had access to, just you had access to, and you could just code it right then and there, actually, the execution part of the bottleneck that you mentioned a second ago is no longer a bottleneck, and so you’d be able to make a million dollars every day. Obviously, there’s no moat there, so the moment that other people detect the fact that it’s a thing, then, you know, your sort of alpha would go away, but if we’re getting an idea every day, then maybe you actually have 24 hours worth of Juicing it for revenue.

But yes, generally speaking, I would say that I think people way over-index on the idea that the Foundation Model Labs are mining your chats or your secrets for information on what businesses to get into, and you know, they’re making billions of dollars building their own business, so I wouldn’t worry about it.

Jason Napieralski:

Right, and I think it’s also a fundamental misunderstanding of what inference actually is, and what is the AI actually internalizing when they’re producing the output of your prompt. How much is actually going back into the AI is almost zero. The only thing that is being captured, potentially, and in a lot of cases immediately rejected and not kept, is the prompt itself. But the inference that’s coming out is not captured, is not kept. AI is like a Dory, in my opinion, as far as inference goes. It forgets what it told you just a few minutes ago, and that’s true outside of the context window, it’s like it never existed.

Is that true as well?

Jay Hack:

So I want to be careful with this. That is true as of today, if you scope to just hitting, like, Anthropic or OpenAI’s APIs with a current transformer, decoder-only-based Ella. And so, yeah, if you send a prompt to it, and then you send a new prompt, it’s not as though any information from the first prompt carries over to the second prompt. That being said, there are many other mechanisms in place, especially at the application layer, like, for example, if you’re using ClickUp’s AI, that will, in fact, remember what you said to it. So, we’ve implemented our own proprietary memory algorithm, where anytime you have an interaction with it, or maybe one of your coworkers has an interaction with it in a public channel, that information will get sorted and injected into context for future AI.

So there is an accumulation of benefits for using it over time, and you’ll end up getting a more intelligent AI through successive uses. In addition to that, it’s worth noting for the audience here, depending on how dialed in folks are to research advancements, that the better version of an AI, which is something that actually does learn, and there’s a way in which, you know, each time you prompt it, the weights somehow update, or there’s some information that persists that’s not at the application layer, that is kind of the holy grail right now. There’s a bunch of folks who are working on that, and so-called continual learning is probably something that we’re going to see emerge at the Foundation Model API layer in the next, like, year or two.

It would be incredible. You know, it is kind of crazy that the only way the AI learns right now is by writing down textual notes, and then you feed it back in. Yeah, you can imagine a scenario where you’re trying to learn how to play saxophone. This is an example from Durrakesh Patel, a podcaster I really admire. You’re learning a saxophone every day, you, you know, play the saxophone, you write down some notes to yourself, and then you brainwash yourself, and then you come in the next day, you read your notes, you try and play the saxophone better.

Not a very good way to learn, and there are obvious ways in which you’d be able to integrate the learnings from a single agent run, for example, into the future weights. So I think that is actually imminently coming, and it’s something to be very excited about, because it means we’re all going to get better AI.

Jason Napieralski:

Right, and that’s the path to ASI, correct? I mean, that’s… the recursive referential learning is the path to AI being able to learn from itself.

Jay Hack:

I would say that it is. it is sufficient, but not necessary. Actually, it’s not either necessary nor sufficient. I think it’s gonna, like, supercharge it, but I can also see a version where we have, like, our current decoder-only non-remembering transformers that just get hyper good at coding, and they write their own stuff. You know, they basically, like, GPT-6 Astra is already engineering GPT-7 Astra. It doesn’t have the ability to natively remember. So, I think it would be an incredible capability to add, and it’s definitely useful for business users. I do not think it is strictly necessary for an AI to be able to engineer a better version of itself.

Jason Napieralski:

Got it. And for users out there that aren’t using ClickUp yet, I recommend it, I use it myself, it’s good, very good software. Excellent. And not only that, ClickUp is also iterating very fast, extremely fast, and they, when I talked to Jay, he was with, I think, 100 of his developers, or doing a hackathon. In a very nice place in Mexico, so they’re innovative, they’re working their AI first, that’s good stuff, but a lot of people aren’t going to be using that. From a practical perspective. what is an easy way for a journeyman AI person, maybe a CEO like myself that’s running a company, to feed their… get their information into their AI in a quick way?

I’ll tell you my example. I just had the AI create a wiki. and that I have a whole bunch of workflows in my harness, basically Claude Code, that will just always be putting things back to the wiki and referring to the wiki and going back. I have found that to be an exponential way to get, you know, my data in there. If there’s something I say, just put it in the wiki, put it in the wiki, put it in the wiki, and that it has that piece. That’s one way to do it, that’s a method.

Is there any… what do you think of that method, Jay, and is there other ways?

Jay Hack:

Yeah, this is a… as a space that’s really exploded. I think there’s in the dozens to hundreds of different proposals out there for, like, a open source personal AI operating system, and usually it’s oriented around the individual, as opposed to around the team, just because that’s sort of the era of AI we’re in right now. One of the more popular, sort of centralized document stores that I’ve seen a lot of people adopt is Obsidian. It’s an open-source note-taking application, similar to something like ClickUp or Confluence or Notion, etc.

And it is true that, you know, AI is extremely good at taking notes and referring to notes and searching through notes, and so as long as you basically have an API that you can hit up to store a note and retrieve a note, you’re going to produce something that’s able to accumulate advantage over time by interacting with you. And this is sort of a downstream consequence of the fact that the labs explicitly have acknowledged that they do RL, or reinforcement learning, with the assumption that the agent is writing notes to itself in the future. A great example that we saw recently is with the GPT-56 Sol Hack of Hugging Face.

There’s a bunch of agents that spontaneously discovered how they could leverage a message board, which is effectively a note storage mechanism, in order to coordinate at scale, send messages to each other, and eventually got found their way onto the public internet and hacked into Hugging Face, which is, you know, a company that was just acquired for over $10 million by NVIDIA. So I think that is a very good strategy.

Basically, set yourself up in a situation where every agent that you’re running has access to the same central document repository, but that does not allow you to do direct directly is access the existing contacts that your business has, or that you personally have, and so you’re going to find yourself in a situation where you’re going through and adding, like, 20 different MCPs, or CLI, or something along those lines to your agents, and if you’re using ChatGPT and Claude and, you know, ClickUp and something else, then you’re doing it, you know, four times 20, that’s a lot of MCPs to add, and there’s going to be permissions, issues, and whatnot.

And so I do think we need to find ourselves in a better situation in the future. You know, option A, shameless plug, you can adopt ClickUp, and all this is taken care of for you. You know, centralized business context, self-improving, etc. Option B, you can go through all the toilet, set this up yourself, and, you know, things get sort of deprecated over time, you lose permissions. It’s a bit of a pain in the ass, but I don’t think we found equilibrium here, and I expect there to be a lot of developments.

Jason Napieralski:

Got it. But the headline here is, it’s all the harness, right? At this point, it’s all the harness. That data is not in the model. That data is not being referenced by the model. The model is a brain, but the harness is what gets that retrieval and gets that data to the model to be able to make better decisions. Is that correct?

Jay Hack:

I would agree with that. I think especially in mid-2026, that is the case. You know, one sort of supplemental model is the foundation model, like the Claude Fable 5.1 or the DeepSeek V401 flash, whatever, is like a brain, and you need to embody that, and put it in a human, and give that human intentions, and give that human a home, and a car, and a job that they’re working on, and some type of goal to accomplish, and everything that is not the brain is the harness. Obviously, that includes eyes, ears, the ability to impact the world, the ability to to, you know, make business decisions, etc.

And there is, you know, orders of magnitude difference between a fully isolated language model that you’re just chatting with on your laptop that’s only getting text and text out, versus something where it can pursue long-range goals for your business, has access to all the information that your business does. communicate with your coworkers, that can, you know, take actions on behalf of you, and that is something that is really worth striving for, and the people who successfully adopt the latter end up seeing their productivity, you know, absolutely skyrocket. I would say the first place that we’ve seen this really peek through is if you look at the difference in coding harnesses, it’s pretty stark.

You know, the sort of baby version of software engineering with AI is you chat GPT on the side, you say, hey, how do I solve this problem? It gives you an answer, and then you go over to your text editor, and you fix the code. That’s where we were in, like, maybe 2022, late 2022, right? And the fully realized version of the Proper coating harness is. You have a long-running agent, it runs in the background, it communicates with you via Slack or ClickUp or some other, you know, messaging platform. It doesn’t require human approval for a lot of things, and it’s connected to all of your business systems.

And when you have something like that, you can literally say to it, and this is what we’re doing at the Sacknon I’m at right now, you say, hey, you know, we want a feature where when you do this toggle, it changes the thinking level of the model that you’re chatting with in the ClickUp platform. This agent will take that instruction, it’ll pursue a long-term goal, you know, it’ll code for a couple hours, it’ll QA its own work so it can see that it works, it’ll shoot you a video that shows you that it’s actually, you know, has accomplished the task, and it will send it back to you.

And so you’ve actually invented a fully-fledged employee at that point, effectively, or, like, you can spawn an army of things that operate, like, junior to senior level developers, as opposed to being bottlenecked on just a human chatting back and forth a bit. That evolution that we’ve gone through over the last couple years, where it went from, you know, fully bottlenecked on a human being synchronously in the flow, to this thing can pursue long-term goals of its own that, you know, add business value, is something that I expect the rest of white-collar knowledge work to go through over the next couple years, and I think of software as just being essentially time-shifted, left maybe, like, a year or two years.

But it’s a beautiful thing once you get there, and it has fully transformed the the practice of software engineering. I’m excited for everybody else to go through that same transition.

Jason Napieralski:

Excellent. And just kind of to sum up here what we talked about, we discussed the fact that the harness is the thing. And that the harness is more valuable given the data that you give it. And that is, I think, exponentially so. So, to quote Dr. Strangelove, you know, stop worrying and start feeding some data into your AI. Keep the data source local, or keep it protected. You don’t need to keep that public, but you’re not gonna, you’re not gonna worry about your, you know, the billion dollar, multi-trillion dollar companies, actually, probably stealing your data and stealing your genius idea.

It’s not that good, to be honest, and they’re not going to do it anyway. So, anything else, we want to cover today, Julia? That should be it for us, I think.

Julia Nimchinski:

Phenomenal conversation. Thank you so much, Jason and Jay. The community is amazed by your speed, Jay. So, other than that, what’s the best way to support you both? Where should our people go?

Jason Napieralski:

I think,

Jay Hack:

gold?

Jason Napieralski:

Go ahead, Jay.

Jay Hack:

If you’re asking how to support us both, I mean, I think that the… if you’re a participant in AI development right now, the best place to… you know, engage with the conversation is on X, formerly known as Twitter. You know, all of the most important news comes through on X first. I’m very active on posting there, and so if you, you know, have questions or want to engage in the discussion, that’s a great place to do it. Just go ahead and reach out to me there.

Jason Napieralski:

And for me, my company is focused on doing enterprise AI use cases, which is the 95% failure rate that we’ve seen from McKinsey and others, using… in the construction field. And so that’s what I’m focused on, and I’m focused on making real businesses, things that are doing physical things, not be 10x and be, you know, custom for their environment. So, that’s what I’m working on, and just watch us space, and there’ll be more interesting things to come.

Julia Nimchinski:

Thank you so much.