Headless Software: Why Companies Are Building Software for AI Agents, Not Humans and what it means

Added: 2026-05-06

[00:00:01]I've been driving a car for 25 years and it's been the same story every time I need to go somewhere. Even though I swear we are supposed to have flying cars by now, but whatever. But it's second nature now, no matter the car.

[00:00:14]You open the door, start the car, grab the steering wheel, shift into drive, hit the gas pedal, and you're off. But I'm actually in San Francisco right now and I'm seeing all these Waymos on the street and there the story is different.

[00:00:27]You don't get in the driver's seat.

[00:00:29]There's no steering wheel. No turning of the keys, no shifting, no gas pedal.

[00:00:33]It's not like the story has changed.

[00:00:36]It's a completely new playbook. When it comes to driving cars and the same shift is happening now when it comes to software. Because since the 90s we've clicked to launch the program. We've navigated with the keyboard and mouse.

[00:00:50]We've made decisions with our own human context and then we repeated.

[00:00:54]But headless software is changing that.

[00:00:57]So, what is headless software?

[00:01:00]It's a newer trend in both consumer and enterprise AI that might cut out the role that humans play in the human-to-computer interface and in the same way that Waymos are changing the role in human-led transportation.

[00:01:16]Now with headless software we just enter the destination and we orchestrate. We sit in the backseat.

[00:01:23]So, that's what we're going to be tackling today on Everyday AI. Uh the headless software, why companies are building software for AI agents, not for humans, and what it means. All right, welcome. This is the start here series, but let's first tackle the big picture before I tell you what this is.

[00:01:43]So, in the last eight days alone we've seen a seismic shift in the way that software is being made, which impacts how agents access it, which impacts how all of us humans actually work. Take Parker Harris, the Salesforce co-founder, who essentially asked, "Why should you ever log into salesforce.com again?" Yeah.

[00:02:06]Salesforce co-founder saying, "Why should you ever log into salesforce.com again?" And the three giants launched agent-first platforms within an 8-day range this month. So, yeah, that's Salesforce, OpenAI, and Google. And the uh a recent Reuters report showed that software stocks lost nearly $1 trillion in market cap as AI agents disrupted the price.

[00:02:34]So, that's what we're going to be tackling. So, on today's show, stick with me for the next 20 minutes and you're going to learn why Salesforce told 25 years of CRM users that the browser login is pretty much over. Uh how MCP and A2A protocols became the new open protocols that every enterprise agent now runs on. You're going to know why interfaceless software or headless software could likely be where the majority of work gets done in the future. And ultimately, where does this leave your company? Your CRM contracts, your IT spend, and all of those customer-facing workflows. All right, let's dive into it. Welcome to the start here series. This is the Everyday AI Essential podcast series to both help newbies learn the AI basics, but also if you are an advanced in AI practitioner, this is going to help you stay up to date and double down on your knowledge.

[00:03:30]Uh so, if you are new here, make sure you go to starthereseries.com.

[00:03:33]That's going to give you immediate and free access to our inner circle community. And in the start here series space, we have a playlist uh set up on Spotify, so you can go listen to every single volume in this series. So, this is volume, what are we on now? 23, I think, right? So, yeah, I I recommend you listen to them in order. It'll make a little bit more sense. So, if this is your first start here series, go ahead, finish listening to this one, but then go back and listen to number one and uh you know, just might as well go listen to that whole uh playlist in order there. So, yeah, make sure you go to starthereseries.com.

[00:04:10]All right, if you did miss our last episode, we talked about agentic context carry and the three steps to understand and use the hidden AI workflow. All right, but let's talk about headless software today. What is it? Well, before we get there, I actually want to build on something that we did in a previous start here series. Uh and this is how fast y'all the landscape moves. Uh because this was about two months ago uh right in volume 10. And we kind of tracked AI's phases. And I would say that now with headless software, it's not a completely new step, but it is a step that is technically going to impact all of the other phases of AI so far.

[00:04:50]So, in volume 10 we traced AI's five phases from chatbot, right? The original chat GPT, to now the autonomous desktop co-worker, right? All the way up to the Claude code and Claude co-work and chat GPT's Codex.

[00:05:05]And at that point we essentially declared the AI assistant era over and the proactive AI worker era officially here. That's obviously still the case, but now headless software or software made for AI agents, not humans, really expedites where we're at on that trajectory. Not only that, I think it obviously raises the ceiling in terms of what teams can accomplish, but I think it can help people get along that adoption journey a little bit quicker, right? Because every single individual, team, and enterprise is on the, you know, kind of zero to five uh scale, right? And you have to work your way there. And I think headless software might actually help some companies get there faster because once you get that piece figured out, uh it does create less friction. So, essentially these big companies are just reworking how their software works in general just to make it easier for companies to get more data out of those systems and ultimately to make their data more valuable, right?

[00:06:07]Because if there's too much friction, you know, if if if there's too much I I talk a lot about, you know, the human duct tape and you know, recently talked about the context carry. All right, but if there's too many uh buttons, too many switches, you you know, eventually that agent is going to break. Right?

[00:06:26]Until the agents get much more capable, until their context windows get much better, until the the the harnessing and the tool calling uh becomes more specific, right? Um yes, we have those capabilities today, but the headless uh kind of era makes it so much easier and it allows, right? Going back to our car analogy, uh it really opens up the highway, right? Now we're on the auto the Autobahn going, you know, 150 miles per hour instead of, you know, 290 here in Chicago going 30 miles an hour in the middle of traffic.

[00:07:02]Okay, so what does headless actually mean? You're like, "Okay, I get it. It's a these companies are changing some things. They're making it easier." But here's what it actually means. It is interfaceless software.

[00:07:15]Right? So, they are literally just rebuilding their software at the core uh to make sure that it works agentically.

[00:07:23]Right? So, headless just means the software still works, but there's no actually uh you know, there's no real user interface required to use it.

[00:07:31]Right? And who knows? Maybe, you know, the fact that um you know, we're almost four years into the generative AI era and the fact that we still have um you know, normal user interface software, right?

[00:07:47]Software built for humans, maybe we'll look back at this time and be like, "What took so long? This is crazy." Or who knows? Uh maybe headless software will be something that is more of a trend or more of a drop in the bucket. I will say probably not, right? Because, you know, whether you're talking about and there's different ways that headless software kind of presents itself, right?

[00:08:05]And I'll break this down in a little bit, but uh I think once you start using it, whether it's through command line interface, uh APIs, right? Been around forever, or the new agent-based, right?

[00:08:16]Um model context protocol, A2A, agent-to-agent protocol, right? The more you use this, once you get it to work, you're like, "Why would I ever log into my software again?" Right? There's so many pieces of software I just don't use anymore. Right? I've I've used This isn't an exaggeration. I Yeah, at one point you could say I was maybe addicted to software. People who know me personally, uh you know, or professionally, I guess, you know, know that about me. I have a list of software. It's it's very long that I, you know, have had subscriptions to. Uh I don't log into software much anymore.

[00:08:49]And I'm actually making business decisions based on, hey, does this piece and it's much easier, right? For me, I'm a small business. Uh you know, a little more agility, so it's not as easy for, you know, SMBs and enterprise companies, but, you know, this does have to become a part of your um procurement, right? You have to say, "Hey, do you have A2A? Do you have MCP?

[00:09:14]You know, what kind of headless states do you support? Do you have command line interface?" Uh right? And for the non-technical folks out there, let me let me think of something. Well, let's just use Salesforce, right? Um because they're kind of one of the ones leading this headless charge. Uh right? And now, you know, to make sure I hit my last talking point here about the app becomes the infrastructure and the agent becomes the actual operator. Because right now, well, us humans for the past, you know, since the 1990s, since computer software became popularized, it's always been us humans the one who were the operator. Uh and that's really not the case anymore.

[00:09:50]Right? So, it started off with, well, humans use the software. We click, we type, we navigate. Then, right? With generative AI, humans we prompted the AI, right? And sometimes that integrated with software, but that's what led us to kind of stage three, right? So, stage one, humans use software. Stage two, humans prompt AI. Stage three, more of the co-working phase.

[00:10:12]Right? Humans can co-work with AI that's connected to software.

[00:10:16]Right? And now we're getting into this phase where I think the fully autonomous um that would be the next final step, I think. But now we're in this phase of humans, we're just overseeing headless software. Right? Whereas before in the co-working phase, you know, a lot of times, you know, for literally talking Claude co-worker, if we're talking Codex, yes, you know, they support headless states now, you know, using these tools through the through the command line, through different MCP protocols, etc. But in many cases, it's still opening apps, right? It's opening browsers for you, right? The Claude co-work can, you know, open apps. It can open your browser. Same thing with Codex, right? And as more and more platforms offer headless support, I think that even concept of the co-working, right?

[00:11:05]Using computer vision to, you know, open apps, using computer vision or browser control to, you know, navigate websites, I think those websites that, you know, maybe, I don't know, in a year, 18 months, if they're not offering MCP A2A from, you know, Google's protocol, if they're not offering command line tools, right? APIs, obviously, you know, most companies didn't make it very far into the 2020s if they didn't have APIs, but it's the same thing. I think that makes the software less valuable.

[00:11:40]So, let's [snorts] talk a little bit more about what these three giants, Salesforce, Google, and OpenAI released over the course of eight days and that one big bet that they are betting on.

[00:11:50]So, first, you know, and they literally called this platform headless 360. This is Salesforce. And they launched the headless 360 software, I guess, right?

[00:12:02]April 15th at the trailblazer DX. So, they also alongside that shipped 60 new MCP tools. But essentially what they did is they turned every CRM workflow into a callable through an API MCP tool or CLI command, right? Command line interface.

[00:12:21]So, what does that mean for our non-technical audience?

[00:12:25]You don't have to go into salesforce.com, click the login button, look at the interface, did anything change? Yes, it did. All right. Where are my deals? All right, here's my here's my deals. Let me scroll through all my opportunities. All right, let me put some filters on. Oh, this filter didn't save from the last time. Oh, someone on my team, you know, added a new deal and they didn't connect it to an account, etc., etc., right? This probably sounds like the day for a lot of people if you're involved in sales, right? And you spend a lot of time, the human, working with the interface. And that's in general can be very time-consuming, especially for more legacy software that's not exactly user-friendly, right? But now it just becomes a command to an agent. And then once the agent has figured it out, then it just becomes a routine that a human oversees, right? In the same way that, you know, Google kind of popularized this teach and learn concept in their original um what was it? The Mariner the Mariner agentic browser, right? So, they were kind of the first to have one of the first, I think, at least in the big players to have this concept of, "Hey, you teach an AI something once, it it sees it and understands it. It looks at the, you know, the DOM or the the dev side of the website, sees what you're clicking, and then it tries to repeat that." Right? So, that's essentially what we are able to do with headless software, right? Agents, yes, they're not deterministic, they are generative, so they're going to make some weird mistakes sometimes, and that's why I preach all the time, spend your time going through the chain of thought, right? Cuz agents are going to go off the rails even if you put guardrails, they're going to go off them. Yet that's what we're going to spend our time doing. And you might say, "Okay, why?

[00:14:04]Isn't it better for a trained human to go in there and go through all those manual steps?" Well, probably not eventually, right? Because as this headless software becomes better, you know, you're probably going to be able to do 10 20 50x what you can do via an interface because agents don't sleep and they don't need to sleep, and right? And that time will be spent on, you know, front end direction and, you know, kind of back end, you know, observability and traceability. So, that's what happened with Salesforce.

[00:14:38]They're essentially like, "Yeah, you probably aren't going to need to log into salesforce.com anymore if you're using this headless 360 product." And they want Salesforce to become the infrastructure for agents that live elsewhere. Like in ChatGPT or in their product Slack. All right, next, OpenAI. So, it's actually interesting, right? Cuz if you listened to our show yesterday, OpenAI had one of the biggest weeks in AI since like 2024, low-key dropping GPT-55, images two, and workspace agents. So, that's what we're talking about here. Workspace agents were launched for all team accounts. All right, so, you know, if you're logging into your ChatGPT account ChatGPT account, you're like, "Where are these?"

[00:15:20]Well, even if you're on a paid But this allows you to build drag-and-drop agents, really good ones, that can connect natively to all of your software, >> [snorts] >> right? Slack, Salesforce, Google Drive, Notion, whatever you use.

[00:15:40]And and then you can run scheduled cloud agents at scale.

[00:15:44]And yes, it does require a new identity and governance layer, so this does change, you know, how humans insert their their information or contacts or requests on the front end and what they're looking for in the back end. But this is another step, right? Yes, we could do that individually, right? We've had these connectors and we've had these apps and we've had custom GPTs, right?

[00:16:08]But now with by being able to build these and scale them across your team, it changes everything, right? That this is a huge step from proactive or that, you know, phase two of human asking AI or co-working, right? Cuz co-working, you are still kind of the human duct tape or you're overseeing an agent that's working every single step of the way.

[00:16:32]Not with, right? Not as an example with the workspace agents. You're doing a better job on the front end and doing a better job on the back end. And yeah, we'll see long-term success. I'm bullish on the workspace agents. I've been using them. I really like them. Even if you're an individual solopreneur, entrepreneur, it might just be worth getting a team account, you know, it's two licenses, it's not that crazy.

[00:16:54]But we'll see because it's only kind of included through May 6th, so we don't know what's going to happen after May 6th, whether it's going to be, you know, certain limits on those, if it's going to be paying for usage, I don't know. But yet, Salesforce um first, OpenAI and Google were kind of, you know, right at the same time. So, you know, we just talked about what OpenAI did. Here's what Google did. They essentially rebuilt their Vertex AI platform. So, Vertex AI, it was kind of like uh I would say it was more for technical people, right? I didn't even use it a ton, right? They have their model garden and, you know, you can build, you know, more advanced enterprise workflows.

[00:17:34]But now they've taken this Vertex AI, one of the most popular platforms, right? Used by some of the largest enterprises in the world. And at their Cloud Next Conference last week, they rebranded the Vertex AI as the Gemini Enterprise Agent Platform. See what's happening here, right? Everyone is bringing in more agents that connect with third-party data, and it runs on their cloud instance, right? Their secure sandbox cloud instance. So, everyone's like, "Hey, bring all your data in here, >> [snorts] >> connect all of your, you know, third-party softwares, and well, they're just going to run in the cloud." So, you know, now, you know, you have your direct API usage. Google said hits 16 billion tokens per minute. So, yeah, that just means people are using this a lot more than they were previously.

[00:18:29]All right, and Google does obviously host manage MCP servers native for Cloud SQL, Spanner, and Big BigQuery.

[00:18:38]So, in the span of just over a week, you saw three of the biggest players in the space, right? One traditional software in Salesforce, ChatGPT, kind of the consumer leader in AI, and I would say still the enterprise leader in AI. We'll see what the rest of 2026 has to say about that. And then Google, one of the biggest names well in software and in tech, all making this big step toward let's try to get rid of the software layer completely.

[00:19:11]And that's able to happen in large part because of some of these platforms.

[00:19:17]Right? So, you obviously have the shift toward desktop AI agents, right? Which changes what they can do by using the terminal, by using the right? So, using the command line. But then you [clears throat] also have kind of these new connectors, right? The USB-C of AI that is the model context protocol, which was built by Anthropic. So, that essentially lets agents call any tool across any vendor like US like a USB-C, right? So, yeah, the easiest way to say say this is think of, you know, two different AI tools in the cloud, and think of the MCP or the model contacts protocol as a wire that just tethers the two together and allows them to connect data.

[00:20:01]A2A, very similar to model contacts protocol, that was launched by Google.

[00:20:06]And that's lets agents from different platforms hand off work to each other.

[00:20:11]Where um the model contacts protocol is maybe a little bit more of data sharing or a little bit more of tool access across different AI platforms.

[00:20:24]A2A is a little bit more of handing off entire workloads, right? It's like, "Hey, I'm done with this. I'm handing it over to you." Where MCP is a little bit more of, you know, traditional information sharing like an API.

[00:20:36]All right. And then both of those platforms are now Linux Foundation governed, breaking the vendor lock-in that once fueled the per-seat pricing.

[00:20:44]So, speaking of per-seat pricing, that's the big thing to figure out here, right?

[00:20:51]The software industry is a multi-trillion-dollar industry when it comes to market cap, right? Software, for the most part, has in one way, shape, or form fueled the modern American economy, right? For the last 30-plus years, right? We obviously saw it with the, you know, the dot-com boom and bust, but obviously um the internet and software has continued to propel us forward.

[00:21:22]Um so, what does this mean? Because the entire success of most software companies is predicated on, well, they want to get big enterprise clients, and then they charge them by the head, right? By the seat. Um so, hey, if you can get a company with 100,000 employees and charge them 10,000 seats, you're doing great. But what happens when maybe that 1,000 seats and the output, right?

[00:21:48]If we if we look at it shifting toward outputs, what if you only need 10 humans overseeing 100 agents each, right? Or 10 humans overseeing 10 agents each?

[00:22:02]How does this change things?

[00:22:04]Right? And and who's going to figure this out first? That's the big That's the big deal here. And and what we're going to be continuing, and I think what will be one of the more compelling storylines in AI and technology in 2026 is who figures out this headless pricing first, right?

[00:22:20]Obviously, keep an eye on Salesforce since they're the ones, you know, really, you know, not to use their own language in a very PR way, right? But it fits.

[00:22:30]They're trying to trailblaze this path forward. Uh but this is where on the software side it's going to, I think, send a shockwave, but there's going to be the same type of shockwave for uh companies trying to make decisions on what software they should or shouldn't use. Because the per-seat software pricing model is going to quickly become mathematically broken. Because right now, like I said, vendors are charging per human seat, but one agent can do the work of 10 or 20 or 50 or 100. And the IDC predicts that 70% of software vendors are going to abandon per-seat pricing by 2028.

[00:23:14]All right? A lot of times these shifts, right? You could say the shift is like the shift to cloud, right? And the shift to cloud, it took at least 8 to 12 years, right? Especially for smaller businesses.

[00:23:27]Um this is fast, right? To go from, you know, essentially a new category being created to 70% of software vendors abandoning per-seat pricing in the course of 2 years, that is a frenzy. So, right now, vendors are testing out and they're trialing pivoting to credits, tokens, actions, workflow runs, and outcome-based contracts. And I think that that will largely become the norm in the past or sorry in the next few quarters. I think you're going to see new challengers come up to take on, you know, kind of enterprise, old, slow, legacy software.

[00:24:09]You're going to have some of the big players, like we already talked about three of them, but you're going to have some of the biggest players spin off their own kind of side brands on this, right? Still keeping their legacy per-seat. And then you're going to have some big ones that are maybe just going to abandon the whole thing and do a major shift, right? Because let's if they're a public company and their valuation is going down by, you know, a couple of points a week, they might have to make some seismic shift, right?

[00:24:35]Because without getting into the whole, you know, vibe coding and is it a legitimate thing for enterprises? Well, here's the thing, even if the answer is no, it's getting closer and closer to maybe every single week when we see all of these new updates. So, something has to give, right? And we've talked you know, we've talked a little bit on the show in the past couple of months, this concept of, you know, token maxing, right? And you know, all these companies want more and more tokens because they're seeing the outcomes that you can derive from these tokens, which is why I think the same expectations that enterprises have when it comes to the, you know, let's just say allotment of compute or, you know, allotment of AI light, right? Let's let's make it easier to certain people or departments. I think they're going to have that same um expectation with software vendors. And I think you should too. If, right? If if you are someone in your company who has a say in that or you're the one leading the charge, you have to be looking at, "Okay, is it time to pivot?" Because what you're going to see, I think, is you're probably going to see an option that might cost 5% of what you're paying ultimately for predictably better results that are likely faster. Will there be bumps along the way?

[00:25:48]Absolutely, [clears throat] right? But for companies that, you know, a single piece of software running seven, eight figures, you have to be pricing this out. You have to be paying attention to this headless momentum. So, there's also a whole 'nother fascinating side, which we're not going to get into a lot, and we did have an actual podcast on this a couple of months ago, but Cloudflare projects that automated bot traffic will exceed human traffic online by 2027. And they say that agents visit roughly 1,000 times more sites than human shoppers, right? Yeah. The whole like, you know, agentic commerce space is wildly fascinating, but yeah, we'll have to have much longer show to tackle that.

[00:26:32]But I think that marketing funnels that were once built for human clicks must adapt to agent-driven shopping behavior.

[00:26:38]You know, maybe this decade we might see a phasing out human marketing made for humans, right?

[00:26:47]That would be a wildly hot take, but I do think, right? When you think of if agents are going to be the ones ultimately browsing the web, right? And I actually did talk to the on the podcast here, the head of AI agents at Cloudflare, and we talked about this literal same thing. If there's, you know, more AI agents, you know, than humans, and we concluded collectively that yes, there were, right? So, if there's more agents out there and they're seeing way more web pages, and the commerce capabilities between agents are increasing, you know, where does that even leave humans in the end in terms of, you know, making these buying and purchasing behaviors, right?

[00:27:29]I I don't know. If my If I have a bunch of way more capable agents in a year that have all my contacts, all my goals, all my priorities personally, professionally, and, you know, they they have access to a safe way to to transact online, does it need me? Maybe. Do I want to be involved?

[00:27:48]Possibly, but I might not want to be, right? I might just want to be having a conversation with my wife, and, you know, we're talking about buying a certain something, and I just speak it into existence, and it shows up at my door. Uh right? Or, you know, I don't know, drone drops it off. Um but I mean, we do have to understand that it's not just AI bot traffic that we um need to be thinking about here. It is the fact that these, you know, AI agents are going to have our business contexts.

[00:28:21]They're going to have commerce capabilities.

[00:28:24]And well, the software layer might just kind of be slowly disappearing or just, you know, reshaping itself as headless software made strictly for agents and not for humans. In the same way we've talked about maybe that's what websites in general will be, right? Maybe websites aren't going to be about appealing to humans. Maybe it's going to be about appealing to, you know, GEO or AIEO, right?

[00:28:51]You know, these generative, you know, optimizations to show up more in AI chats, right? So, maybe human websites won't be for humans anymore.

[00:29:01]They'll be for agents. And the same thing can be said for all software across the board. So, what does this mean for you and your business? Well, first and foremost, as anything, unlearn.

[00:29:12]I'm I'm I'm doubling down on unlearning in 2026, y'all, cuz I'm I'm so tired of people trying to double down on on upskilling.

[00:29:19]Upskilling won't work. You have to unlearn working without AI, right? So, what I mean by that is uh you have to unlearn all of your processes. You don't sprinkle AI on the top because that's what so many companies are doing, and then they're doubling down on that. So, don't just do the, "Oh, we upskilled with AI, so that's good." No. Blow it up. Start over because look, software, you know, software may not really exist in its form that it is now in a year or two. You have to start unlearning these old antiquated SOPs now, all right? Uh next, you have to audit your top vendors today to see what they support. This is something I'm doing all the time. I actually have a a working markdown file that my different agents are always checking because if one of my new pieces of software as an example, we use Beehiiv for our email newsletter and they just came out with an update in their MCP we use Circle for our free everyday AI inner circle community. They just came out with an MCP software, right? So, I have my agents constantly going out and then updating that file and then updating our workflows. Um you know, depending on what happens, right? And you should be doing the same and then with that comes a new way to work almost every single week. But you need to be auditing your top vendors for MCP tool support, API access, and agent readiness overall.

[00:30:46]And if they aren't, well, you need to start potentially looking at other vendors. And then you need to renegotiate seat-based contracts.

[00:30:55]Because I don't know, >> [clears throat] >> if if you're buying 10,000, 20,000, or even 500 seats for an expensive software, but it has agentic capabilities, well, why are you still paying for all of that? Obviously, there's a a bigger piece of that puzzle in that data is going to become the most important thing and yes, the pricing might change to make it to where, you know, these companies can still kind of keep their revenue the same or potentially grow it. But, you know, I think a lot I think some people have written off certain software, including myself. But I I still do think that there's a certain type of software that will slowly die off. I think the software that is maybe actually going to become more valuable is the software that agents can use, but that humans need. You know, it's not something that makes our life more convenient. Oh, this tool does this and it's it's really nice and it helps us collaborate, right? No.

[00:31:55]What proprietary data does that software have? And if that software has proprietary data, I do think that that software there's a bullish case for you to continue to still obviously, you know, keep working with that because agents need data, right? Agents don't need a better way to collaborate, right?

[00:32:12]So, I'm I'm not saying, you know, project management tools and all that are going to go away because there's data that lives in there, right? Your your project statuses, communications with accounts, etc. But, I think that you have to shift more than anything.

[00:32:25]Shift from training employees on software. Get away from that. To finding the most efficient and reliable ways to get the same outcomes with AI agents and then reverse the human piece, right? So, don't say, how can we retrain 10,000 people on using software A? You look at software A, you see what outcomes you get, then you see is software A the best thing to get that outcome? Is there proprietary data that only software A has? Can we take that data and bring it to a new platform that's more agentic by default? That's what you have to do.

[00:32:58]Find the most reliable and efficient outcomes first that you are accomplishing with X number of humans on software and then you have to refigure out how to, you know, unlearn. This is the unlearning process, right? And then you rebuild by teaching these employees from the ground up. Here's how we orchestrate this set of agents. Here's how we observe. Here's how we trace and here's how we create the expert-driven loops.

[00:33:26]All right. That is a wrap. Headless software, interesting. All right, so I know this one was a little bit more technical, but I actually thought it was worth updating one of our earlier shows.

[00:33:37]So, go listen to volume 10, like I said, if you want to. Uh because this is how quickly things change, right? We've seen the recent trend toward kind of desktop co-workers, right? Autonomous agents that work on our desktop, they can access our files, they can read and write. And now, headless software just makes that process easier. Um if anything, it is going to expedite the adoption rate for desktop software because when you're using the command line, when you're using the terminal, you know, that opens up. If the software is headless, you don't have to use slow, clunky computer use. You don't have to use token-heavy browser use, right?

[00:34:20]Again, we'll see how this is ultimately priced, but at least right now there is an opportunity to before companies, you know, get this pricing piece figured out, there is a competitive advantage here, right? Because eventually, one of the big companies is going to figure it out, they're going to set the pricing structure and then in a matter of quarters to maybe a year or two, the entire software industry will have to follow suit. But in that interim, there is a period where I think you can still get an almost unfair amount of outputs with maybe way fewer seats. If you know what you're doing, if you keep up with the trends, the tools, and what matters, and that's what here for That's what we're here for. So, all right. If this was helpful, make sure to go to start here series.com. Go sign up for free access to our start here series space in our inner circle community. That's going to give you a playlist of every single start here series in order. So, you can go listen to them and be the smartest person in AI at your company. So, thanks for tuning in. I hope to see you back tomorrow and every day for more everyday AI.