The AI Questions That Matter Begin After the Demo

A few months ago, I had a conversation with someone who has been working around artificial intelligence since 1984, and it changed the way I thought about the current AI moment.

Most of us are experiencing AI through the lens of the last few years. ChatGPT, Claude, agents, copilots, vibe coding and multimodal models have made artificial intelligence feel like a new industry. But it isn’t new. What is new is the level of access, the speed of development and the number of people who can now interact with these systems directly.

The person I was speaking with had worked on early AI hardware, built and exited technology companies, invested in startups and spent decades watching different waves of technology rise and fall. Because of that background, I expected the conversation to be extremely bullish.

Instead, it was much more measured.

People tend to overestimate the impact of technology in the short term and underestimate it over the long term.

That feels especially relevant right now.

There is an enormous amount of excitement around AI, and much of it is justified. But there is also a lot of exaggeration. New models are released constantly. Basic features are marketed as autonomous agents. People talk about artificial general intelligence as though it is around the corner.

What I appreciated about this conversation was that it focused less on the hype and more on the architecture underneath it.

The product questions after generation

We talked about large language models and the fact that their outputs are inherently probabilistic. That matters because some of the behavior people describe as temporary bugs may actually be connected to the way these systems fundamentally work.

That changes the product conversation.

I spend a lot of time teaching people how to build with AI and working with founders who want to bring these tools into their businesses. Increasingly, I think the interesting questions are moving beyond whether AI can generate an answer.

We already know it can.

The harder questions are around what happens after that.

What information should a system have access to? What happens when it gets something wrong? What should be treated as the source of truth? How should memory work over time? When should a person be required to approve an action? How do you understand what an agent actually did once the work is complete?

Those questions may not produce the most exciting demos, but they are becoming some of the most important product questions in AI.

What a useful second brain requires

We also talked about the idea of AI becoming a “second brain.” I like that concept, but only when we are clear about what it actually means.

A useful second brain does not need to recreate human consciousness. It needs to retain context, understand your information, retrieve the right knowledge when you need it and help you make better decisions.

That distinction matters because AI conversations can quickly drift into abstract claims about intelligence, consciousness and what it means to be human. Those discussions are interesting, but they can also distract from the practical problems that are already here.

Memory is still a problem. Reliability is still a problem. Security is still a problem. Context is still a problem. Observability is still a problem.

And those problems create real opportunities for builders.

Easier building makes judgment more valuable

The conversation eventually moved into how companies should be built around this shift. We talked about venture building, education and the idea of identifying meaningful problems first, then bringing the right people together to build around them.

That also stayed with me.

AI has dramatically lowered the barrier to creating software. Someone with an idea can now build something in a weekend that might have required a small engineering team only a few years ago.

That is powerful, but easier building does not automatically create better companies.

In some ways, it makes judgment more valuable.

When the cost of building falls, deciding what deserves to be built matters more. Product thinking matters more. Distribution matters more. Understanding users matters more. Governance matters more. Knowing when not to add AI matters more.

That was probably the biggest takeaway from the conversation.

The future of AI will not only be shaped by the companies building the most advanced models. It will also be shaped by the people who understand how to build reliable systems around them, how to apply them to real problems and how to separate what is technologically possible from what is actually useful.

The more I build and teach in this space, the more I think that distinction is going to matter.