Nate B Jones
When to Automate, Build, Buy, Hire, or Wait on AI
Ask the tutor about this video
Sign in with a verified email and ask the tutor anything about this video.
TL;DR
Research summary
▶ Start with the work, not the AI
The speaker opens with Gartner’s projection that more than 40% of agentic AI projects will be killed by the end of 2027, attributing likely failures to cost, unclear business value, and inadequate risk controls rather than agentic technology itself. They describe a finance leader whose CFO wanted AI for orders-to-cash, while three vendors proposed different solutions without describing the actual work. The speaker recommends first discussing internally where and why to invest. AI investment, they argue, is fundamentally a question about how work is shaped; model, vendor, and dashboard choices come later. An accounts receivable team may have distinct workflows such as collections prioritization, invoice matching, customer follow-up, exception handling, cash application, dispute resolution, reporting, and escalation. Bundling these into one RFP can produce a poor fit. They make a similar point about product work and say the unit of AI investment decision should be a specific piece of work, not a department or role.
▶ Define workflows and assess automation
The speaker defines a workflow as the full operating loop: incoming information, permitted system actions, good outputs, checks, escalations, and ownership and accountability for results. They say the model is only one part of that loop, so investment should focus on workflows. To assess a workflow, they suggest considering how often it repeats, the cost of mistakes, the judgment required, how company-specific it is, whether the market has a solution, whether future model releases could absorb it, and where its output goes. They outline five levers: automate or delete the workflow, build, buy, hire, or wait; a solution can combine levers, such as hiring before building. Automation is appropriate when work repeats frequently, follows a clear pattern, has definable exceptions, and can be checked cheaply. IBM AskHR and Intercom’s Finn are offered as examples involving combinations of levers. The speaker cautions against automating when exceptions contain most of the value, and says vendor demos may show routine cases that do not represent exception-heavy production traffic. A detailed scoring rubric is mentioned as available on the speaker’s Substack.
▶ When to build
Building is presented as an option when work is unique, has many edge cases, or depends on company-specific context such as data, standards, approval gates, and risk thresholds. The speaker describes a build as a repeatable agentic loop that may involve skills, connectors such as MCP, plugins, data calls, sub-agents, and vendor tools; a purchased component may still be part of a mostly built workflow. Before investing, leaders should define the required data, what good output means, workflow boundaries, and how output quality will be assessed. The speaker says this clarity is often missing from build discussions, even as agentic capabilities have improved over the preceding four or five months. Executives need to know whether a result is acceptable and be able to reject it, revise the mission, or hire people if needed. Asking a team to build something without defining success is described as no solution.
▶ When to buy
The speaker says buying depends on whether a company can apply the purchased capability to its workflow and get value, including whether the underlying data or system substrate and integration path are suitable. They distinguish buying reusable primitives or services that can be stacked into workflows from buying a complete workflow solution. Stripe is cited as offering agentic primitives that teams can begin using, and the speaker describes tools for passing context between agents as another reusable component that could be integrated and built around. Harvey is used as an example of a product selling a fuller agentic pipeline for legal work. For a vendor workflow solution, the speaker recommends assessing whether the company’s work and the vendor’s envisioned workflow overlap by roughly 80–90%; less overlap may require more adjustment work than expected. The speaker says this is more complicated than buying bounded traditional software.
▶ Hire for the missing capability
The speaker says companies often seek an ideal AI hire combining domain expertise, AI building, systems architecture, executive experience, and change leadership, but argues that this person may be hard to find. They recommend identifying the human capability that the target workflows will need in six months or a year and hiring for the team’s specific gap, such as domain trust, workflow engineering, evaluation design, or executive ownership. They describe the AI hiring market as difficult, with AI-generated resumes, possible video deepfakes, unclear job descriptions, and roles taking months to fill. Defining workflows can clarify what to hire for and whether current staff can develop the needed capability within six months. The speaker mentions Talent Board, a community being established with Substack to connect people around AI roles and demonstrate AI talent; they say it is in its early days. Their recommendation is to define roles specifically and, if hiring is chosen, hire clearly and quickly.
▶ Wait selectively and define the work
Waiting is described as a valid lever when a workflow is lower priority and change-management resources are limited. The speaker is not advising businesses to avoid AI transformation, but to direct resources first toward workflows that offer the most leverage and help people learn to work in an AI-native way. They recommend stacking investments rather than waiting indefinitely. As an example, a company rewriting its analytics system might leave working, deterministic SQL query pulls alone for now and prioritize AI-enabled natural-language analytics that could provide more upstream leverage. The speaker’s central principle is not to automate work that cannot be described. Leaders should be able to state inputs, outputs, standards, exceptions, and ownership in plain English; otherwise investment decisions are difficult. Broad requests can conceal multiple workflows and ambiguous outputs, and similarly vague job descriptions can lead people to interpret and interview for a role differently. The speaker urges specificity and points to a Substack deep dive on operational aspects of useful agent work.
▶ Use a workflow investment matrix
The speaker presents an investment matrix based on two axes: how specific the work is to the company and how mature the AI market solution is for the relevant industry. Common work in a mature market is an obvious buy; examples include Workday for commodity HR and payroll, Stripe for payment primitives, and standard help desk solutions. For common work in an immature market, they recommend narrow prototyping or waiting, and caution against a five-year contract for a category that could look different in 12 months. Company-specific work in a market with useful primitives can be built using purchased building blocks such as connectors, models, or orchestration, while the company owns the workflow and its standards. Company-specific work in a thin or immature market is presented as a strong case for building. The speaker says hiring can apply across the matrix, especially when work requires trust or someone to define the standard and what good looks like.
▶ Allocate capital around workflows and people
The speaker says executive work is changing because capital allocation requires understanding adjacent workflows well enough to set outcomes, prioritize problems, allocate talent, and prepare teams. Executives do not need to personally evaluate every tool; they should understand the workflows behind decisions to build, buy, hire, or wait. Rather than treating AI as a single strategy blob, the speaker advocates making investment decisions after examining specific workflows. They criticize framing the issue simply as AI versus people and recommend discussing where people should focus their time, where they can develop, which talent gaps require hiring, and how jobs should change when task bundles are automated. They argue that remaining human work can become more impactful as agentic systems take on work, and frame successful AI investment as a people problem as well as a workflow problem. The closing advice is to understand workflows, describe them specifically, and hold discrete investment conversations about those that matter most; the speaker closes by wishing viewers luck and saying they will return next time.
Full transcript
[00:30] That's why so much money is flowing into this space. This video is about how to think about your investment logic in your AI projects so that you adequately invest in the parts of the project that truly drive value versus investing in levers that are likely to make you disappointed and you end up on that 40% Gartner list. Look, I had a finance leader tell me last month that her CFO wanted to do AI
[01:02] in orders to cash, and three vendors had quoted her three different shapes of solution. None of them had described the actual work that she was doing. And I got to tell you every conversation in this space feels like that where like the people inside the business are saying, "This is what we need." They don't fully understand what that looks like from a workflow perspective, and there's about 10 million vendors knocking down the door saying, "Here, we'll sell it to you. This is what you need. I promise you this is what you need. You don't understand AI, but this
[01:32] is what you need." We need to take a minute. We need to shut the proverbial door to all the vendors for just a second and have a conversation inside the house about where we invest, why we invest, and what is likely to yield success from an agentic workflows perspective. And that is what this video is about. So, here's the first thing that I see going wrong in the conversations I've had a peek in and conversations executives have with me privately, etc. First and foremost, AI investment is not an AI question. It is actually a
[02:04] question about the shape of our work. The model question is downstream of that, the vendor question is downstream of that, the dashboard or whatever you would want to build and show is downstream of that. What sits at the root of the whole conversation is how the work itself is shaped and accomplishes value, but it's really hard to talk about that. We don't have a good vocabulary for it, and I find in practice most teams skip that step, especially if their vendors encourage them to do that, which so many do. Let me give you an example here. An accounts
[02:34] receivable team does not have one singular AI problem in the space. They have, you know, half a dozen, maybe eight. They have to tackle collections prioritization. They have to tackle invoice matching, customer follow-up, exception handling, cash application, dispute resolution, reporting, and escalation. Those are all very different shapes of work. They route to very different investments. Like you might buy some, you might build some, etc. If you pile all of them into a single RFP, which I see happening a lot, you're
[03:06] going to get a mediocre tool that does maybe one of them well and isn't adequately covering what you really need and maybe covers a bunch of other stuff in another department. It's just not going to be a great fit. What if we look at product? It's a similar situation, right? User research synthesis is in one shape of work, spec drafting is in another shape of work, backlog grooming and design review and experiment analysis and roadmap judgment and launch coordination and customer escalation are all different shapes of work. Maybe some of them are builds, maybe some of them are buys.
[03:36] The unit of decision for AI is not your department head. It's not a particular role. It is that work that I am naming. Now, quick definition before I go further. When I say workflow, I don't mean a prompt. I mean the entire operating loop. What information comes in, what the system is allowed to do, and what good output looks like. Who's checking what? What gets escalated? Who owns and is accountable for what the result is? The AI model is a tiny tiny part of that loop. It does
[04:07] make the whole thing go. It's like the brains of the business. I get it, right? It's a big deal to have an AI model in a loop. That's why we're having this conversation, but it's not the only thing. And if you want to understand how to invest correctly, you have to think less in terms of model and more in terms of the workflow because the workflow is what you are actually investing in. That's what gives you leverage if you do it better. Once you get to that level, the question becomes much easier to follow. Every workflow can be evaluated. There are handful of obvious inputs, how often a workflow repeats, how costly a
[04:38] mistake is in that workflow, how much judgment does that workflow need, how specific to you is that workflow. Does the market have a solution here? Is the next model release going to eat this workflow? Uh where does the workflow output go? So, I want you to think about your workflows. Think about your high priority workflows in that sort of deeply enmeshed detailed understanding. And I want you to then walk into your investment decision from there. Then you walk into, well, do I build? Do I buy?
[05:10] Do I wait? Do I kill the workflow? Do I hire for it? You really only have five options or five levers when it comes to your workflows. You can either automate it away. Uh you can also call that eating or deleting the workflow. It's similar. You can build that workflow in detail with AI and it's a complicated workflow and it's not fully automated, but there's big AI components. Uh you can just buy a solution off the shelf and take care of it. Uh you can hire and those smart people are supposed to help you make the right calls. Or you can just do nothing and wait. And by the
[05:40] way, often times the solution is a mixture of those. So, you have to think about moving more levers at once. Like I've seen cases where someone wants to build, but they need to hire to build first, for example. Of those levers, the easiest call is to automate. If you're going through your workflow priorities, automation is the one that most teams understand. Automation, deleting, eating the workflow, it's the right call when the work repeats often, follows a clear pattern, has recognizable exceptions that you can define, and you can check if it's good really cheaply. So, IBM AskHR is a great example. They're
[06:10] another one, if you're thinking, "Well, that's build, Nate. You're doing multiple levers." Yeah, we're doing multiple levers. Another one is a buy lever plus an automate lever, and that might be Finn. So, Finn is an agent from Intercom, and is really tasked with sort of tackling repeatable customer support case volume. Now, there are folks that build that. Finn is a case where you can buy that, but it's a similar idea. But, whatever it is, you cannot be religious about automating. You need to be focused on where the
[06:40] value is in the system, and where AI can handle it versus where a human can handle it better, whether you're dealing with internal or external audiences. So, regardless, automation makes sense where routine cases dominate and exceptions are easy to understand. Don't automate when the exception is where most of the value is. And by the way, I think this is where a lot of bad enterprise AI demos start to fall down. The vendor shows you the routine case in the deck,
[07:10] and the buyer signs the contract because the routine case is impressive, but the buyer never realizes that their production traffic is a lot of exceptions, and the executive team is staring at an accuracy number that's very low, wondering why they were lied to. Well, nobody was lied to. The buyer just bought the wrong thing, which happens a lot right now, which is why I'm making this video. If you want to dig into what is a full scoring template, how do you understand how to have a good conversation around whether to buy, whether to invest in building,
[07:41] whether to hire, how do you balance all of that for particular applications? I have a very detailed rubric that I put together on the Substack for that. But for now, I want to focus on these big levers in this video, so you understand how to start to pull them. So, automation I think is the easiest one. The next category is one that a lot of executives get excited about, and they don't really know how to allocate money efficiently here. So, I'm talking about building here. I'm talking about the idea that the work shape that you select is not suited to purchasing because it's unique, because it's something that has
[08:11] a lot of edge cases, it has a lot of exceptions. It's something where you have company-specific context that matters, your data, your standards, your approval gates, your risk thresholds, whatever it is. It's your team's way of doing the job. It's the secret sauce. And obviously, you are ready at that point if you're building to invest, right? This is not a chatbot solution. We're talking about having the right repeatable agentic loop. Where you may have skills involved, you may have connectors like MCP, you may have plugins, you may have
[08:41] data calls, you may have sub agents. You may have even vendors with tools that you call inside this. Remember how I said you'd have multiple levers? Well, you might buy a lever that is a tiny part of that loop in that workflow, and then you build most of it. So, the hard part here is making sure that you understand what is the data that you need to put into this workflow? What does good look like for this workflow? And how do you
[09:11] know that the output at the end of that workflow is going to be up to snuff, going to be great, going to be fantastic if you try to build it? Because right now, presumably if you are in business, the humans can already do it pretty good. If it was easy to automate and delete, you would have automated and deleted it. There is something complicated enough here that you need agents and data and tools and connectors to do all of that. And I'm asking you, and this does not get asked enough, do you have bounds
[09:42] around that task? Do you understand the edges of that workflow? Do you understand all the bits that go into it? I named a bunch of them. And do you know what good looks like at the end? Because your team is going to come back to you, and they are going to be incentivized to tell you, "Yep, this is good. Yep, we built the AI thing, the AI thing the executive wanted, we did it for you. It's amazing." Okay. Do you know if it's good or not? Can you be the honest third-party eyes that say, "You know what? This actually
[10:13] works." Or "You know what? This is terrible. It's unacceptable. Go back and build it again." Or "I gave you the wrong mission and you couldn't build it. You need to have people. I'm going to hire people." Whatever it is. That level of clarity around value is missing from most of the build conversations that I get told about, that I've sometimes been in the room for. And even though we've had this massive gain in agentic pipeline of value, agentic pipeline skill, agentic pipeline impact in the
[10:44] last four or five months, this conversation keeps repeating. The people in the room buying have not up-leveled their conversation in response to what we can actually do, and they aren't realizing how important it is that they, the executive, understand what good looks like if they're going to be suggesting building. So often it's like, "You go build it. We can't afford to buy it, and we can't afford to hire people, so you go build it, and it better be good." And they can't tell you what good looks like. That that is not a solution. That So that's the build level. Now we come
[11:14] to buy, right? It's often build versus buy, so I thought putting them next to each other made sense. Buy is really a question of whether you have the capability to take the thing you're purchasing and apply it to your workflow in a way that gives you value back right away. And that's a lot more complicated when you're talking about workflows than when you're talking about traditional software. You need to understand what is the underlying substrate that your purchased solution will sit on. Is it a
[11:44] data substrate? Like what is it Where's it going to sit in your system? And then how do you know that your dev team is going to be able to integrate it well and actually get you value? And that's always been a very high-level question software, but because software traditionally has been so bounded, it's been a really clean box, you can at least have that conversation cleanly. With workflows, if you're buying a solution that is effectively a part of the workflow, it's going to be more complicated. And that's why when I talk about buying, I often will separate it out and I will say you want to either be buying primitives, like basic components
[12:15] or services that you can stack into a lot of agentic workflows that you build, and those are fairly easy because then if your dev team likes them, they'll use them a lot. Um I think Stripe has put a lot of agentic primitives out there right now, uh that are actually very easy to get started with. You're not making a big purchase decision, you're just starting to play with them and build with them and your dev team can put solutions together, uh and that's great. Or you want to be in a position where the primitives are something that allow you to build directly in your system. And so there's some tools out
[12:46] there where you're starting to have like uh tools that help your AI that you build communicate with other AIs in your system. So it's sort of a a notebook tool or a context tool that's focused around tickets. There's different kinds of solutions there. Those are things that you can imagine re-applying because they're basically about how does the AI agent communicate context to other agents in your system? You may buy that one piece, it may integrate with your particular tools, and you may build around it. And then there's the folks who kind of like Harvey, for example, for legal. They sell the whole thing. It's a whole
[13:17] agentic pipeline effectively under the surface, and you have to decide if that works with your workflow as a legal firm or a legal department. And so when I when I look at all of that, the the hardest and most complicated one is sort of the Harvey case. How do you decide that buying Harvey makes sense? And I think the heart of that question is if you were to look at the work that you're doing today, is it kind of Harvey shaped or not? Do you know their product well enough to answer that question? If they're selling you a workflow and I In
[13:48] this case, I don't care that it's legal. Harvey's just an example. If you're buying a vendor's workflow solution, do you know that solution well enough to be confident that there's like an 80 90% overlap with the shape of their work and how they envision the workflow and yours? Because if there's not, you're going to do a lot more work than you think adjusting it. And it's more complicated than than in the age of AI than it was in the age of deterministic software. Now hiring. I'll be honest with you, a lot of companies right now are trying to find the impossible hire, the purple unicorn, the domain expert
[14:18] who's an AI builder, who's a systems architect with executive experience, and a change leader. Sometimes that person exists. More often than not, the market is going to clear out a lot of AI talent from under you while you figure out what you actually want. And so I think the better question if you're looking for AI talent to hire is to ask yourself what kind of human capability the workflows that you're putting together actually need in 6 months or a year, and then hire for that missing piece that you don't have on
[14:48] your existing team. Maybe it's domain trust, maybe it's workflow engineering, maybe it's evaluation design, maybe it's executive ownership. I don't know what it is, but that is a much more sustainable way to hire than to just say we need the perfect AI unicorn. And I think that this is something that's a good sort of reminder for all of us because I get that we have talent issues with AI. People are coming to me saying, "Can you help me find talent?" And the answer is I know lots of folks. I'm looking for lots of folks. I bring folks together where it makes sense. But
[15:19] ultimately, you need to be in a place where you understand your work well enough and how you're investing in your team long-term well enough and how your team's gaps align to the workflows you're building to be able to make a coherent job description that is for a specific person, not a purple unicorn, and who you can hire for in that market and really assess. One of the things that makes hiring really hard right now is the people are trying to sift through all the noise from the AI-generated resumes and maybe AI deepfakes on video and all of the fluff around hiring and
[15:51] the complexity of the market, really a broken hiring market right now. And at the same time, they have never had less clarity on what they want to hire for. And so they're trying to sort of wade through the fog of their own job description and also wade through the fog of the market and it's just a disaster. Like they they it takes months to clear roles. It's frustrating for candidates, it's frustrating for companies, it doesn't make sense. Hire more specifically. And again, the workflow is a key. The workflow helps you unlock that hiring definition and
[16:22] what what higher gaps you really have versus what your team can grow into. And you really should be at a point where you can say, if my team or someone on my team can level up in 6 months to get to this particular talent set I need for where I want to go with my workflows, keep them inside the house, train them up, level them up, do not hire for that because it's just it's so painful to hire right now. And I know that there will be people who are listening to this video who are like, "Ah, you know, it's me. I'm raising my hands. I I'm here." And I look, I get it. But on the hiring side of the table, it
[16:53] looks like 10,000 people saying, "I'm here. It's me." and raising their hands and how do you know which is which? And so, we've talked about that. Uh I've talked about Talent Board, which is a community that we're standing up along with uh Substack to help folks to connect over hiring roles and over uh proving that they have AI talent. It's early days, but I'm excited for that. And ultimately, where we want to go here is we want to be in a position where we start to clear the fog out of the market. And if something like Talent
[17:23] Board or another tool helps you to understand who actually has AI skills versus who says they do, and then if you take this video seriously, and you start to actually define your role as a hiring manager, and you start to understand how your role aligns with workflows, I think you'll be in much better shape to go wherever you go with your hiring journey, and actually get clear answers on the AI talent, and actually pick up AI talent before the market clears it out from under your feet. And that's really my goal, right? If you're going to choose to hire, hire clearly, and hire quickly.
[17:53] And that takes a lot of work on your part up front to define up against the workflow you're looking to get to. Okay, waiting is the last motion, and it's the most counterintuitive when the world is shouting at you to do AI. I am not saying, by the way, that you should not be leaning into AI transformation. That is not what this is about. I am saying you should be deliberate about where you apply AI for leverage in your business first. Let's say your whole business probably needs AI transformation, which is a fair basic assumption for most businesses.
[18:23] Think about where you get the most leverage from going first, and when you have things that are workflow-shaped that are lower on the priority list, think about waiting. Maybe it doesn't make sense to start there. You have limited resources for change management in your firm, by definition. Apply those resources where you get the most bang for the buck. Apply them where you get the most leverage from transforming the workflow, the most leverage on your people from learning how to work in an AI-native way, and then you can spread
[18:55] out from there to drive the change. And so, I'm not saying wait forever. I'm not even saying wait for a year. I'm saying stack your investments, and make sure that the right ones are up top, and deliver disproportionate leverage, right? I'm not Cuz there's a lot of people who will tell you, well, if you're waiting, you're too late, etc. Look, in the larger sense, if you are saying, I don't want to do AI, I would fully agree. You are too late. It is a problem. You got to get on board with that as a company, as an individual.
[19:25] Fully agree. But, within that, and this is true for individuals as much as firms, you have to understand where you want to prioritize that time, and there will be some things that you don't want to prioritize because it's just too too much work to change, right? Like, I'll give you an example. If you're trying to rewrite your analytics system, you may not want to prioritize changing the SQL query pulls too fast because SQL just works for you, and you can get deterministic pulls and data, and it's just not broken. And so, it's not the first thing you're going to fix, and you're going to focus on natural
[19:56] language descriptions of the analytics and things that you can only do with AI that give you a lot more upstream leverage. And that's a very small example, but it illustrates what I mean when I talk about waiting and why why waiting makes sense in certain cases. Now, there is one principle that I have been dancing around in this whole video that I want to say very, very plainly. If you remember nothing else from this video, remember this. Do not automate what you cannot describe. And this is a line that should sit in every single AI investment review. If
[20:27] you cannot describe the work you're doing in really plain English, well, what's your inputs? What's your outputs? What are the standards here? What are the exceptions? Who owns it? It's going to be really, really hard to make good investment decisions. And a lot of AI projects right now fail before that conversation ever comes to a conclusion because the team that wants to propose the AI thing doesn't have clean, clear words like I just described to talk about what they mean. They're hiding 20 workflows inside the same broad ask to
[20:58] the vendor. They are talking about broad outputs that could mean six different things, and everyone inside the room is reading that differently. They're putting broad words in the job description that they're hiring for and everyone's talking about that differently and interviewing for it differently. Be specific so that you can get impact for your AI investment. Please, please, please. Now, if you want to dive into the operational layer, what actually makes agent work in practice useful and high yield for the firm, I have a whole
[21:29] deep dive on that in the Substack today. I think it goes with the whole conversation here. I didn't have time for it in this video, but it's it's a good reminder that you need to be thinking about agent capabilities and agent workflows end to end as a leader if you're going to make a build by higher weight decision. And so I wanted to dive deep on that. All right. Before we go further, let's look very classically at an investment matrix because I love these, right? Like this is I've seen these in so many board decks. We're going to look at an
[21:59] investment matrix and and how you think about decisioning. So let's do that here. So what you have is two axes. First an axis about how specific this work is to your company. Is it super specific or is it pretty pretty general and everyone in the space does it? An example of that would be everybody does customer service in telecom. Like that's not a specific thing, right? That's not special. But on the other hand, if you have very, very specific plan switching incentives that are driven by a long-term marketing motion and nobody
[22:29] else has those, well, that might be specific. The second axis is vertical and it is how mature the market solution is for your vertical on AI. And that obviously changes by industry a lot. So very, very high level, right? That these always have four boxes and I want you to get get the takeaway here. If the work is common, general, right? And the market is mature, it's an obvious buy, right? You can think of this outside AI is like Workday for commodity HR and payroll, right? You can say you're buying Stripe for payment primitives, right? You can
[23:00] say you're buying any standard help desk solution for standard help desk needs. It makes sense. The work is common. Now, if the work is common and the market is not mature, you want to prototype narrowly or you want to wait. The category is still defining itself. You don't want a 5-year contract for a tool category that will look different in 12 months. And if you want to invest here, you want to be prototyping and building cuz you have a chance to win the category. So, this sort of depends on your vision for the company, where you want to go here. Now, if the work is company specific and the market has some
[23:32] useful primitives, buy the primitives, buy the building blocks, buy the tools you can call, but you own and set up the workflow that everything runs on where the value lives. This is where most ambitious teams ought to be living right now. You want to buy the connectors, you want to buy the model, you want to buy the orchestration, and you want to own the standard. Uh, now some people, of course, are standing up their own open weights models. That's another way to do it. But regardless, uh, you build it and you just bring in the building blocks as you need it. Now, there's a bit of a blurry line here, but if the work is company specific and the
[24:03] market is very thin or immature, you want to definitely be building because there's value there. Um, and you want to make sure that you're building in a way that enables you to own that category. That's a very easy call. If you're wondering, hiring cuts across this grid. You need to define your workflows to get the hiring right. And by the way, pro tip, if the work you're trying to define is something that nobody is able to define well, if it requires trust, if you need to frame it up, if someone needs to define the
[24:33] standard and what goods looks like and you're like, "Oh gosh, Nate, I don't have that." That's a clue that you should be hiring. Your next investment is probably a person. Um, and you need to think about what person that is and how you can define their role in such a way that they're set up to enable that larger workflow decision to be helpful, correct, and long-term impactful for the business. And if you're wondering, does this mean the executive role is changing? Absolutely. We're talking about a change in how we allocate
[25:03] capital and a need to understand adjacent workflows in more detail to do that well. The job is not to become the person who personally evaluates every single tool. I'm not advocating that. The job is to understand enough about the workflow you're investing in to make really good capital allocation decisions, which is what we're doing, right? If you're hiring, if you're building, if you're buying, if you're waiting, those are all capital allocation decisions. And you have to define what are the outcomes that matter, what are the problem frames that matter, what are the
[25:33] workflows that you should prioritize, where do you want to allocate talent in that mix, and then how do you set up your teams to be successful as you think about these larger workflows that you want to unlock for AI and you want to prioritize for AI. And so what I'm advocating is essentially, instead of looking at AI as a sort of a singular blob that you want to have a conversation about where your CEO says, "Let's do an AI strategy." No. Don't do that. Instead, look at the workflows and make good investment decisions once you
[26:04] understand those workflows. One last thing before we close. There is a very cheap version of this conversation I hear a lot that tends to turn the debate into AI versus people. I don't think that is a serious version. It's not a helpful version in most cases. The serious version of this conversation asks where people should be maximizing their time, where should they should be up-leveling, where there are talent gaps you can hire for, and where we have cases where people need to transform their allocation within a job family because certain bundles of tasks
[26:35] are getting picked up in automation. That is a much more serious conversation. It's one that is productive. It's one that's useful. It's not one I hear nearly as much as the drama-filled version, AI versus humans. That It's not useful. Ultimately, the human work that remains in this version as we look at these workflows is getting more impactful and more leveraged because we are putting AI and powerful agentic systems at the heart of the business and we need to get that right. And that is that is a people problem. And so this is not AI replacing workers.
[27:06] This is figuring out how to make investment decisions that unlock disproportionate value that get you out of that Gartner 40% figure and get you into a place where you're getting real value back on that agentic investment. And and it starts and ends with a workflow. If if I if you take nothing else away from this, understand your workflows, be able to talk about them specifically, and have discrete investment conversations about those particular workflows that matter most to
[27:36] your business. And that's going to set you up for success in a way that most conversations that start with we need an AI strategy will not. Best of luck and I'll see you next time.