The 5 Levels of Agentic Commerce

ChatGPT Instant Checkout launched on February 16. Within days, 800 million users had a direct checkout path through an AI agent. Glossier, SKIMS, Vuori — already live.

Most people read this as AI helping you shop. That framing is already out of date.

The agent isn’t the assistant anymore. It’s the buyer.

What that actually means plays out across five distinct levels of autonomy — and most DTC brands are only prepared for the first one.

Level 1: Assisted

The agent helps. You decide everything.

It fills forms, compares prices, applies coupons. Your intent is explicit at every step. The agent has no autonomy of its own — it’s a tool you wield. This is where most people assume we still are.

Level 2: Intent Translation

The agent interprets. You still approve.

“I’m traveling with a toddler. I need a car seat by Friday.”

At Level 2, agents don’t match keywords — they reason about context. They translate a situation into a structured product query. The human still confirms before anything moves.

Level 3: Preference Memory

The agent remembers. Trust starts to matter.

Agents at this level know your preferred brands, your ethical constraints, your budget norms. They shop according to who you are, not just what you typed this morning. This is where the DTC brand relationship starts to shift — brand loyalty is no longer between a person and a company, it’s between an agent’s memory and a set of encoded preferences.

Level 4: Budgeted Delegation

You set policy. The agent executes. This is the inflection point.

“Keep household essentials under $300 a month. Preferred vendors only. Reversible purchases.”

The agent buys independently. You review the receipt — not the decision. The infrastructure for this launched last month. Most DTC brands aren’t ready for it: their entire stack (landing pages, CRO, scarcity tactics, loyalty UX) was designed for a human making a deliberate choice. None of that reaches an agent operating inside a policy.

Level 5: Anticipatory Action

The agent acts before you ask.

Restocking before you run out. Switching vendors when something better appears. Renegotiating when your contract is up. No prompt required. Last week, Circle ran a live agent-to-agent payment with no human in the loop.

The Gap

Most brands are building for Level 1. The infrastructure for Level 4 went live last month.

Everything designed for the human buyer was built for a different world. The real question isn’t whether agents will buy from you — it’s whether your product is even visible to them when they do.

The bottleneck between Level 3 and Level 4 isn’t technology. It’s trust, standards, and liability. Agents need to know they can trust a vendor’s data. Vendors need frameworks for agent-readable product information. And someone has to be accountable when an autonomous purchase goes wrong.

That’s where the next layer of commerce infrastructure gets built. I’m writing about all of it.