It’s hard to walk into a meeting today — whether with a platform, a payments provider, an enterprise customer, or a government — without someone asking about “agentic” solutions.
‍
Commerce is no different.

That doesn’t mean agentic commerce is already mature or operating at scale. It does mean the conversation has shifted. The question is no longer if software will be authorised to act on behalf of people and organisations, but how that delegation should work safely and responsibly.

That shift matters more than any individual demonstration of what an agent can do on its own.

For years, AI in commerce has largely been advisory: search, recommendations, comparisons, customer support. These systems influenced decisions, but humans still executed them.

Agentic commerce marks the transition from influence to execution. Software systems are now being authorised to discover products, assemble carts, initiate payments, and complete transactions — sometimes with a human in the loop, sometimes within predefined limits, sometimes autonomously.

It’s important to be precise. Agentic commerce is not yet large in terms of transaction volume. Most real-world deployments today are constrained, assisted, or tightly guarded.

What has changed is the level of commitment underneath.

When commerce platforms redesign catalog and checkout for agent access, when payment providers engage on delegated authorisation and liability, and when standards bodies accelerate work on interoperability, that signals something more durable than hype. These are not experiments that get quietly rolled back.

This is the phase where experimentation is widespread, architectures remain fluid, and ecosystems form in parallel. History suggests it’s also the phase where foundational choices matter most.

Agentic commerce can evolve in many directions.

Trust has to work in all of them.

Once that shift is accepted, the next question almost always follows: if everyone is building agentic systems, won’t this all fragment?