For over 10 years, RIoT has partnered with All Things Open to host its annual Demo Night. I would argue that this is the best technology conference in the United States, and definitely the most accessible in terms of size.
What has always made ATO special is the intention behind it. Todd Lewis and his team have consistently built a stage where global experts and first-time speakers share equal footing, where early-stage startups rub shoulders with established companies, and where often overlooked voices in technology are not only welcomed but prioritized.
Of course, there are also vendors. Business is being done. But the center of gravity always leans toward education, access, and inclusion, rather than spectacle.
So last year, when the ATO announced a new conference series focused entirely on artificial intelligence, I was eager to attend. Due to scheduling conflicts, I was unable to attend the first event, which attracted more than 1,600 attendees and received rave reviews. This year, my schedule aligned and I attended the second annual All Things AI in Durham last week, joining an estimated sold-out crowd of over 4,000 people.
It did not disappoint.
What struck me first was not the technology, but the people. At a conference dedicated to artificial intelligence, every seat was filled with humans. Yes, many people actively used AI throughout the event. Yes, one keynote even included an AI agent that ran parts of the presentation. But no one sent agents on their behalf. The value of participating in conversations, listening to questions, and participating in shared learning was definitely human.
The scale of the event confirmed this. This was not a niche gathering of developers and researchers. It was a gathering of professionals from a wide range of fields trying to understand how AI fits into their work. I spent my first day at a business workshop led by Mark Hinkle, a longtime open source advocate and co-host of All Things AI with Todd Lewis, and the room reflected that change. These are not the people asking if AI works. They wanted a way to use it responsibly, effectively, and immediately.
This demand is no longer theoretical. It’s real and it’s urgent.
The most memorable moment of the event occurred during Mr. Whurley’s keynote address. After speaking, he left the stage, but an AI agent intervened and gave a second presentation that was not on the agenda. It was a planned reveal, but executed in such a way that the audience paused. Worley, a world-recognized expert in quantum computing and CEO of Austin-based Strangeworks, later explained that during a flight from Austin to Raleigh, he built an agent and instructed it to listen to a live keynote, incorporate advance talks, and generate a new presentation with slides in real time.
He claimed he had no foreknowledge of what the agent would say.
At one point, the agent demonstrated how to triage whurley inboxes across multiple companies, a moment that drew both laughter and a little discomfort from the audience. It was both impressive and disturbing. The ability was obvious. So were the risks.
It was also a glimpse of where AI is heading.
The pattern behind the noise
Conferences like this generate many ideas, but not all ideas become signals. Value comes from stepping back and identifying the patterns that connect them. What I’m seeing in Durham is not just that AI is advancing. That means AI is becoming more deeply integrated into the way we work, the way we make decisions, and the way we organize our businesses.
AI is becoming infrastructure.
Major platforms are no longer just tools you open when you need them. Those are the environments you start building within. Prompts, workflows, custom instructions, and projects accumulate over time and become more than just your usage. They become a system of work.
And that brings us to the second realization that is often overlooked.
True vendor lock-in is not a model. It’s a memory.
Every interaction with these systems contributes to the growing context that exists within a particular platform. This context makes the system more useful, more personalized, and more integrated into everyday work. It also becomes difficult to leave. In the cloud era, lock-in was about where applications ran. In the AI era, it becomes increasingly important which systems understand users best.
This raises uncomfortable but necessary questions about portability. If your workflows, thought patterns, and organizational knowledge are captured within one platform, how easily can you move them to other platforms?
At the same time, a different pattern is forming on the business side. The leading model providers are approaching the market in a way familiar to anyone who lived through the rise of cloud computing. They want to be your primary platform. They want their tools to be standardized. They want long-term relationships that anchor your business within their ecosystem.
But agent AI offers another possibility.
Agents can be tailored across multiple models, from large to small, specialized to general. Therefore, values may not apply to a single platform. Instead, they may move to a layer that understands how to apply these tools to real-world business problems. System integrators, consultants, orchestrators, and solution providers can become central partners, leveraging multiple AI systems behind the scenes to manage complexity on behalf of clients.
In this model, you are not purchasing access to a single platform. You’re investing in capability.
Here I am reminded of the most important lesson I learned in Durham and the one that will be most important in the future. Evolution is not leadership. Change doesn’t happen by chance.
One of the most pointed comments I heard during the conference challenged a pattern that has become commonplace in many organizations over the past year. I’m sure we’ve all heard variations of leaders encouraging their teams to experiment, explore, and be curious about AI. It was a necessary starting point. But that’s no longer enough.
Beyond the experimental stage
If AI is becoming infrastructure, it will be less about whether to adopt it and more about how. The habits you form now will create flexibility, efficiency, and competitiveness over time.
Here are five takeaways worth considering.
1. Treat your AI platform like infrastructure, not a tool.
If you’re using AI to build workflows, generate content, or make decisions on a regular basis, you’re no longer using tools casually. It’s built on top of a system. This means you need to consider durability and portability. Capture prompts, document processes, and save your work in a neutral format like Markdown. Store your most important agents, GPTs, and automated workflows in a neutral repository like GitHub. treat you Interaction as an asset It can be reused and tailored, rather than being a throwaway conversation tied to a single interface.
2. Start using audio as part of your workflow.
Voice fundamentally changes the way we interact with AI. Instead of typing a query and waiting for a response, you can have an ongoing conversation. I found this especially useful during my commute. What used to be passive listening can now be actively explored. Podcasts are inherently one-way. Voice-based AI is interactive, so you can ask follow-up questions, challenge assumptions, and refine ideas in real-time. This is a more efficient and often more productive way of thinking and learning.
3. Build with one model and validate with another.
Although large language models are powerful, they are not foolproof. Their output is driven by probability rather than certainty. For important work, it’s worth introducing a second layer of review. Use another model to critique, validate, or score the output based on criteria you define. This simple practice will significantly reduce errors and improve the overall quality of your work.
4. Be wary of long-term commitment to a single AI platform.
The instinct to standardize on one provider is understandable, especially given the precedent set by cloud computing. However, the emergence of agent systems suggests a more flexible future, where value is created by coordinating across multiple models rather than relying on a single model. Before committing to a long-term contract, consider whether it is limiting your ability to adapt to the evolution of the ecosystem.
5. Move from experimentation to execution within your organization.
If you’re in a leadership role, it’s time to move on from fueling your curiosity. Effective implementation requires structure. Identify champions within your organization who will be responsible for testing tools, developing workflows, and documenting best practices. Make training others part of their role. If such expertise doesn’t exist in-house, invest in external support. The organizations that succeed in this next stage will not be those that experiment the most, but those that learn the fastest and extend that knowledge most effectively.
Thoughts of the end
Durham wasn’t just a gathering place for people interested in AI. This reflected broader changes already underway. Technology will continue to evolve. New features are coming. Entire categories of work will be reshaped.
But the most important decisions are not made by the model. They are created by how people decide to use them. And as the responsibilities become clearer, so do the opportunities. Because evolution may be inevitable, but leadership is not.
