Name: The Fortune 500 AI Forecast
Uploaded: 2025-12-09
Duration: 12 min 33 s
Description: What do the world’s biggest companies get right—and wrong—about AI? Corporate AI whisperer Allie K. Miller shares the most urgent insights from her work consulting across industries, revealing how enterprises and individuals should think about scaling AI responsibly, sustainably, and competitively.

What do the world’s biggest companies get right—and wrong—about AI? Corporate AI whisperer Allie K. Miller shares the most urgent insights from her work consulting across industries, revealing how enterprises and individuals should think about scaling AI responsibly, sustainably, and competitively.

The Fortune 500 AI Forecast

Conferences

Good morning. And amazing to see so many clients out there. Have I told you that you look beautiful today? We are going to be talking about change management, and I want to kind of start it off with a little bit of a hot take. I want to stir the pot. If we could get the slides on, we will be able to kick it off just yet. The hot take that I want to share with you all. Is that AI is not a tool, and I actually think that calling it a tool ends up being a little bit of borderline self-limiting behavior that is holding enterprise all around the world behind and that they're actually stuck. Uh, AI is a weird thing. AI is a little bit closer to like an alien or a beast, and it is actually leaning into that weirdness where leaders can gain an unfair advantage. So why is it weird? Unlike traditional software software 1.0, where you would put in one input, you'd get the exact same output done. AI allows you to reason, adapt and respond. Now that method of working opens up. Four ways of working with it. What are those 4 ways? What are those 4 interaction modes? The first is AI as a microtasker, AI as a companion, AI as a delegate, and the big one AI as a teammate, and I want to show you what each of these looks like. The first is AI as a microtasker. Every single one of you has used it this way, at least where we're asking trivia questions like which came first, Fortune magazine or the FIFA World Cup. We can give it a perfect 18 page prompt for the greatest M&A dashboard of all time. And we can ask it for songs and audio and video and infographics. We get back that trivia question. Fortune magazine came first in case anyone was curious. We get our beautiful dashboard that would make me cry with joy, and 90% of your employees are stuck in this mode, and so many employees think that they are an AI super user when all they are doing is asking AI to write their mean email in a slightly more polite way. And this is holding companies back and actually your annual subscriptions are made worthless because people are stuck in this mode. So let's talk about the others. AI as a companion which Joe talked about is the ability. To work real time with AI side by side where it codes with you, edits with you, talks live with you. I'm kind of obsessed with this one, sees what you see, and this is where AI becomes a lot less transactional, where I am assigning something and a lot more relational that real-time collaborative work. And some people have taken the term companion a little bit too far. Uh, I'm not advocating for work husbands, work wives that are digital, uh, but we'll save that one for like a new TLC show or something. The third interaction mode is AI as a delegate. This is when you are assigning large tasks, maybe they're 40 minutes long, and you're then getting it back. We can see things like inbox management. I know people who have fully offloaded their inbox to AI. They spend 1 cent a month and they don't even touch their own inbox. There are people managing workflows and of course very complex multi-step AI. This is me using an agentic browser and asking Chat CBT to build me an entire AI avatar video with me not even making a single click. This is really where job transformation is moving into real task reallocation, and most companies have not even gotten to this stage. We're going to talk about one more though. That is AI as a teammate. Now, I started in AI almost 20 years ago. And this is one of the most underrepresented, underutilized, and forgotten modes because it kind of only started about a year ago because of AI's performance. Two examples engineers from OpenAI are treating codecs inside of Slack as a teammate, pulling it into trap threads, working all together, and second, maybe AI is sitting in on your meetings, fielding questions, taking action. Now the big difference for AI as a teammate is that AI is lifting up a system or a group and not the individual. This is where AI is helping your entire engineering team or every single person in that meeting and lifting up the system. That is the big difference here. What are you optimizing for? And this is AI as operating infrastructure. We are going to have AI agents and systems that are sitting where we work slack Zoom Teams, what have you and they will be ambient they will be gathering information and importantly we will no longer be prompting AI. I don't know if people have quite come around to this AI will be prompting us because it will be in our systems and helping our team as a whole. So all of this, how do you actually operationalize it and what do I wish more leaders were working on? That is a brand new mental model that every single leader has to curate at a company level. You guys have heard of MVP, minimum viable product. I'm not gonna have you raise your hands, but I assume it's 100% because we're an SS. We are gonna take that methodology and we're gonna apply it to the world of AI agents. Because the capability that everyone has to now build out is carving out safe, minimum viable autonomy. What is that small safe bet that you can work on company wide, and I'm gonna break out how to carve out space for this. I have about, you know, 2 minutes to solve all your world's problems, but we're gonna talk about how you can get it not just with your engineering department but all departments, and I'll give you a template that you can steal as well. So this is kind of the core checklist for MVA or minimum viable autonomy. You are going to scope it tight. You are going to keep a human in the loop. Maybe it's just as a final check. You're going to keep it modular so that as AI changes every week or hour or minute that you can adapt with it. You're going to know your risk thresholds and we're gonna talk about that in a bit. And most importantly is logging and measuring everything and figuring out if AI is good or not, and that last one matters more than you think. Why? Because evals, figuring out whether AI is good or not is the new product requirement. We are no longer giving step by step perfect instructions to these AI systems. We are going to provide goals and boundaries and rules, and AI systems are gonna work from the goal backwards. They are going to check their work against our goals and so instead of defining these perfect 18 page prompts, we are instead giving it the sphere that it is going to work within. This is goal oriented AI. This is a fundamentally different kind of software than we are dealing with today. The capability that everyone should be working toward now until we have AI built into the hardware in our brains, which I'm sure is going to come like tomorrow or something, is agent protocols, and I want to give you some examples. Clawed skills is the ability to teach AI how your org operates in one way and it remembers it forever. I built out a. skills that uses an AI first index assessment. This is over 700 lines of directions, code, templates, resources, examples. I've trained this clawed skill, and I can now run an entire assessment on how AI first a company is in minutes. And this is something that used to take me weeks. Agents. MD, which is much more in the coding space, is from OpenAI. This has been used on over 20,000 GitHub projects, and what I love, one of their best practices as they're giving AI tasks, they group them into always do, please ask first and never do, and I love AI agents with healthy boundaries. My therapist would be so happy. So one template to steal, this is how I'm bringing it to not just engineers. This is a preview of something that I share with clients and as you're going through this template, realize that it actually takes hours or maybe even days to fill this out. This is not a quick one off prompt. This is architecting AI teammates throughout your company. I want you to be thinking about how to document the human process today, how to measure it. What the compliance requirements are that your general counsel becomes your best friend. I want you to be thinking about, uh, checkpoints for AI and what responsibility you can give these AI agents. I know companies that allow AI agents to spend up to $1000 without anyone checking it. I know companies that have an even higher spend limit for AI agents. OK, not only think about that, but when to escalate to a human. This is how you have safer and responsible AI use. And also, until we have self-learning AI, how are we optimizing these AI systems? Now I mentioned risk thresholds before. Importantly, as you're building out your agent protocols across your company and leaning into MVA. I want you to take about 70% of them on low risk tasks, about 20% on complex cross-department tasks, and about 10% on strategic tasks that change the structure of how you work. 70 2010 and this template can be used across all of them. Now, I'm gonna give you some final predictions here. These are 3 bets that I'm making and happy to argue with anyone during lunch as I'm eating some sort of sandwich. OK, 3 predictions. Number 1 is that AI will likely continue to get better. And what would I do about that? I would lean into MBA, build out that autonomy portfolio, and figure out how to move out of the polite email prison and move into AI as a delegate and AI as a teammate and lean into that weirdness. The second prediction is likely in the next several months AI will be able to work autonomously for over 8 hours uninterrupted. In order to take advantage of this, knowing that it's likely coming, I would pull together your employees and start thinking about use cases that take advantage of this and running some sort of MBA hackathon or agent hackathon and require that through that process they write out an agent protocol. You can even come up with your own template. The coming out of this, you are not only going to have an employee workforce that is trained on AI agents that knows how to use your tools and that is excited to be able to work on this, you are also going to walk away with a perfect stack of agent protocols for everything in your business, and that can be used to train new people, to use brand new AI systems immediately without having that start friction. The third is that the cost will likely continue to drop on today's level models as you are thinking about how to take advantage of what is to come, where are you using one query today? Where in the future maybe you instead use 1000 queries or 100,000 queries, and what would that change about businesses today? Maybe you're gonna build out hyper relevance, hyper personalization machines. Maybe you're gonna build out mass simulations for every market launch you do in the future. These are the three changes that I am taking a bet on. Now, of course, we don't exactly know what the future of AI will bring. We don't know how fast it will come, and we don't know what will happen when it does. But the bet that I am taking is that AI is not just a tool. And the organizations who continue to treat it like one are going to wonder over the next decade what happened. Thank you.