Agentic commerce describes purchases where an AI assistant or agent does part or all of the buying: researching options, comparing them, and completing checkout on the user's behalf. In 2025, OpenAI introduced Instant Checkout in ChatGPT along with the Agentic Commerce Protocol, developed with Stripe, and Google announced the Agent Payments Protocol for agent-initiated payments. Adoption so far centers on retail, but the same ideas reach self-serve software purchases.
How it works
Protocols in this space define how an agent learns what a merchant sells, how it requests a purchase, how the user authorizes payment and how the merchant fulfills the order. The merchant typically stays the seller of record and keeps the customer relationship, while the AI platform hosts the conversation and the checkout step. Details differ by protocol and are still changing, so read current specifications rather than relying on summaries.
- Product data: structured feeds the agent can read
- Checkout: an API the agent can call with user approval
- Payment: delegated, scoped payment credentials
- Confirmation: explicit user approval before money moves
What SaaS companies should watch
For self-serve SaaS, agentic commerce raises a few practical questions. Can an agent find your plans and prices without a sales call? Can it understand seat-based or usage pricing? Is your signup flow usable by an agent with user approval? Enterprise sales will remain human-led, but low-touch plans may increasingly be compared and started through assistants.
This area is new, and adoption and platform rules change quickly. Treat it as something to monitor and test rather than a channel with known economics.
Example
A self-serve plan an agent can buy
Tidewell, a fictional status page tool, sells a self-serve plan by card. It publishes plan names, prices and limits as plain text and a structured product feed, and keeps its signup to a few labeled fields. When a user asks an assistant to set up a status page tool under a budget, the agent can compare Tidewell's plan accurately and hand the user a ready checkout to approve.
Common mistakes
Hiding prices for self-serve plans
Agents compare what they can read. 'Contact us' pricing for low-touch plans removes you from agent-led comparisons.
Assuming one protocol will win
Several competing approaches exist. Follow the ones relevant to the platforms your buyers use.
Questions
Mostly for self-serve plans and add-ons today. Complex enterprise deals still go through sales.
Some programs charge merchants a fee on completed purchases. Terms vary by platform and change, so check the current program terms.
