The fact that Chief AI Officer is now a real title at many Fortune 500 companies tells you almost everything you need to know about where enterprise AI stands today.
Not long ago, the biggest questions were about what AI could do and the tools themselves. Now, the tools are in place and leaders are wrestling with how to scale, govern, and effectively measure AI’s impact. Not to mention keep their organizations prepared for what comes next.
Ahead of Ai4 2026, we sat down with Michael Weiss, Co-Founder of Ai4, to talk about the conversations he’s hearing from thousands of enterprise AI leaders and what they reveal about where enterprise AI is headed next.
What are the biggest shifts you’ve observed in enterprise AI conversations over the past year?
Michael: The whole vibe changed pretty dramatically. A year ago, many leaders were still in “help me understand what this stuff even does” mode, kind of kicking the tires. Now they walk in already having built something, and they’re a little frustrated, because their pilots work fine but they can’t get them to scale across the org. That’s basically the conversation now, across the board.
The other big shift is agents. A year ago it was this interesting thing people talked about on panels, and now everybody’s trying to put real money behind them and figure out how to make them work inside their business, which is a totally different level of commitment. Now people are starting to uncover challenges with rolling out AI across the enterprise (e.g., “these tokens are expensive”).
After bringing together thousands of AI leaders every year, what separates organizations that are successfully scaling AI from those that seem stuck in the pilot phase?
Michael: I’ve watched a lot of these leaders sit in a room and compare notes, and the funny thing is it almost never comes down to the model or the tech. The ones who actually get to scale tend to have a few boring things in place. Somebody owns AI, with a budget and exec team buy-in. Their data is in decent shape and they can get to it without a 12-month data wrangling project.
The ones who get stuck are usually running 20 little experiments with nobody really in charge of any of them. None of that is glamorous, but the difference is mostly integration, data, and change management, not whatever frontier model everybody wants to argue about.
What do you think organizations are still underestimating when it comes to AI transformation?
Michael: The people side.
Everybody funds the technology and almost nobody funds the change management, the training, and getting folks to actually change how they work. That’s usually what decides whether anyone even uses the thing you built.
They also don’t focus enough on the “plumbing.” People massively underestimate how much of this is just data and wiring it into the systems they already run. The model itself is a relatively small slice of the work. Most of the effort is everything that must be true around the model.
We hear a lot about AI technology, but much less about leadership readiness. How has that conversation evolved among the executives attending Ai4?
Michael: It’s matured a ton. A couple years ago our exec content was basically, “here’s what a language model is, please don’t be scared.” Nobody needs that anymore. The leaders showing up now are asking way harder questions: How do I organize my company around this? Who should own it? How do I actually tell if it’s working? How do I put guardrails on it without killing the momentum? The fact that Chief AI Officer is even a real title now that most Fortune 500s are adopting kind of tells you everything.
What questions are business leaders asking today that they weren’t asking even 12 months ago?
Michael: A year ago it was mostly, “What can this do for us?” Now, it’s more like, “How do I actually run this and prove it was worth the money?”
And from there, it’s all about the specifics: Do I centralize this or spread it out? How do I govern agents once they can take actions and not just spit out text? What does this cost me when the compute and token bills are real line items?
In a year’s time, the whole conversation has shifted from curiosity to being on the hook for results now.
If you could give one piece of advice to a leadership team just beginning their AI journey, what would it be?
Michael: Pick one problem that actually matters, hand it to a real owner with a real budget, and push it all the way into production before you go wide.
The mistake I see constantly is people launching a dozen pilots, nobody clearly owns any of them, and none of them ever ship. You’ll learn way more from pushing one thing all the way through, with all the messy governance and data and people stuff that comes with it, than you will from 20 fun demos that never leave the lab.
In your view, what skills are becoming essential for business leaders, even if they’re not in technical roles?
Michael: Being able to think creatively about AI without needing to build it yourself. That means knowing what it’s good and bad at, so you can smell the difference between a real use case and a fantasy. It means knowing enough about data and risk to ask your technical people the right questions. Going to a conference like Ai4 is a good start to build your “AI compass.”
On the technical side, we’re seeing growing interest in agentic systems and retrieval-augmented generation. Why do you think these topics are becoming priorities for engineering teams right now?
Michael: Because they’re the two things that make AI useful for a business instead of in the abstract.
RAG (retrieval-augmented generation) is basically how you point a general model at your own data so the answers come from your reality and not just whatever it picked up off the internet, which is the whole reason an enterprise can start to trust the application.
Agents are the next rung up, where it’s not just answering you but actually going and doing things, chaining tasks together. The catch is both are hard to pull off in production, all the evals and reliability and permissioning and cost, and that’s why engineering teams are heads-down on them right now.
Conferences often inspire people, but attendees still have to return home and execute for their organizations. How can they turn their excitement into lasting capability?
Michael: Leave with one thing you’re actually going to act on or build, not 10 things you’re going to “think about.”
The energy at these conferences is real, but it kind of evaporates on the flight home unless it’s attached to something concrete. So come in with your real problems, go find the two or three sessions and people that actually speak to them, and leave with a specific commitment: this use case, this owner, this first milestone by this date.
We also find companies who bring their team are more likely to act on something they learned at the conference because everyone is on the same page, versus the one “AI champion” having to bring everyone else up to speed.
Looking ahead to the next 12–18 months, what conversations do you expect to dominate AI leadership?
Michael: Agents in production, and everything that comes with letting software take actions on your behalf: governance, security, keeping it reliable, whether you can actually trust it.
We’re moving from AI that answers questions to AI that goes and does things, and that raises the stakes a lot. Right next to that, ROI is about to get very real. We’re already seeing token cost concerns. The “let’s just experiment” budgets are drying up and boards want to see the money, so measuring value is going to go from a side conversation to the main event.
What’s one trend you think is getting too much attention, and one that deserves more?
Michael: Overhyped: everybody obsessing over which frontier model is a hair ahead this month. It matters way less than people think when it comes to actually getting value out of this, because the model is almost never the thing holding you back.
Underhyped: World Models. Once a “foundation world model” hits, which I think we’ll see before the end of 2027, an entirely new class of AI applications will open up. Compute demands will grow, justifying the AI infrastructure build out spend.
What are you most excited to learn from attendees this year?
Michael: The war stories from people running this for real. We’ve got 12,000 people coming, and more and more of them are past the experimentation phase and actually running this stuff at scale. I want to hear what broke, what governance approach held up, and how they got their people to buy in and use it.
GA has partnered with Ai4 for three years now around workforce capability and AI education. Why do you think partnerships between AI communities and learning orgs are so important?
Michael: Because getting inspired and actually being able to do the thing are two totally different animals, and you need both. Ai4 is great at the first part of showing people what’s possible, putting them next to the folks who’ve already done it, and creating that “wait, we could pull this off too” moment. But it’s only a few days. Turning that into a workforce that can execute takes real structured learning, and that’s exactly what a partner like General Assembly is there for. Three years into our partnership and we’ve seen Ai4 bring the spark and the network, and GA build the skills.
If attendees could leave Ai4 with one new capability, not just one new idea, what would you hope it would be?
Michael: The ability to take an AI use case from “idea” to something real inside their own company. Scoping it, making the case for it, getting the right people and data around it, and getting it shipped.
If everybody left Ai4 able to drive just one project from idea all the way to production, I’d feel like we did our job. At the end of the day, the AI industry that we serve is about building AI and all that goes with it. If Ai4 doesn’t help people build, we’re not doing our job.
Turning conference conversations into organizational capability
Our conversation with Ai4’s Michael Weiss makes it abundantly clear. The broader conversation isn’t about whether AI matters anymore. It’s about whether organizations have the leadership, skills, and operational foundations to turn promising ideas into measurable business outcomes.
That’s exactly why General Assembly is partnering with Ai4 to offer hands-on, preconference training on August 3.
Whether you’re leading AI strategy or building AI systems, there’s a track designed for you:
Business (nontechnical) track
Gen AI & AI Agents for Business & Product Leaders
Learn how to identify high-impact AI opportunities, prioritize AI initiatives, evaluate risk, and lead successful AI adoption across your organization.
Technical track
Agentic Ops & Retrieval-Augmented Generation (RAG) in Practice
A hands-on workshop exploring advanced RAG architectures, retrieval strategies, evaluation frameworks, and production-ready AI systems.
These four-hour workshops are available as an add-on to your Ai4 conference pass and an awesome opportunity to make sure you and your organization are ready for the now and the next.
Learn more and reserve your workshop seat.