AI recommending something for you to buy is one thing. Giving it permission to actually buy it is something else entirely.
That’s where agentic commerce starts to get super interesting.
Imagine telling an AI agent: Find me the best nonstop flight to New York next Thursday, keep it under $700, make sure I can change it, and book it.
The technology is increasingly moving toward being able to do exactly that. But the moment an AI goes from recommending an action to actually taking one, an entirely different infrastructure has to sit underneath it.
How does the agent prove it’s acting on your behalf? What exactly is it authorized to spend? Which payment methods can it access? How do you prevent a compromised agent from making unauthorized purchases? What happens if it makes the wrong decision? Who is responsible?
Identity, permissions, authentication, spending limits, audit trails and human approval are paramount.
This gets even more interesting when you take it beyond consumer shopping.
An enterprise AI agent could eventually reorder inventory, purchase cloud capacity, renew software, book logistics or transact with another company's agent. At that point, we’re not just talking about AI assistants anymore. We’re talking about software with real economic authority.
For the last few years, the big question has been: What can AI do?
I think the next question might be: What are we comfortable letting it do for us?
Where would you draw the line?
PS: Yes, this is an AI generated image of me shopping. And yes, it gave me an extra toe 😂