On 6 October, OpenAI introduced apps that people can use directly inside ChatGPT. It also launched AgentKit, a collection of tools for building agent applications. The announcements affect both marketing's contact with prospective customers and the development of self-service in businesses' own portals.

Norway was outside the initial app rollout

The first apps, including Canva and Booking.com, became available to users outside the EEA, Switzerland and the UK. Norwegian customers were therefore outside this launch. OpenAI also opened a preview of its Apps SDK, allowing developers to connect their services and interactive interfaces to conversations.

For a Norwegian marketing department, this initially changes the distribution landscape. Customers may encounter product information inside another provider's service before visiting the company's website. Accurate prices, clear product descriptions and consistency across channels become relevant even where the purchase still takes place in the online store.

More components for businesses' own portals

AgentKit includes visual workflow development and ChatKit for embedding agent-based chat on websites. ChatKit and the new evaluation capabilities became generally available. Agent Builder entered beta, while Connector Registry began a limited beta rollout. These different maturity levels affect what businesses can plan to adopt.

In our assessment, a clearly bounded portal process provides a better basis for decisions than a general promise of automation. An agent might prepare a proposed order, for example. Customer approval, checking the customer's price agreement and recording the order are distinct steps that need to be resolved before the application can complete a purchase.

Analysts distinguish assistance from action

In its 16 October analysis, Forrester describes a market where agentic commerce refers to very different services. Its analysts distinguish experiences on a company's own properties from those in external answer engines. They also stress that product recommendations and help with selection differ from an agent independently completing a transaction.

This qualifies the vendor's launch narrative. Our editorial assessment is that results should be measured through specific customer actions: can the customer find the right item, understand its description and complete an order on the correct terms? More conversations alone reveal little about profitability. October's announcements broaden the scope for experiments, while market availability, integrations and actual customer behaviour determine their value.

Sources

OpenAI: apps in ChatGPT and the Apps SDK (October 6, 2025)

OpenAI: introducing AgentKit (October 6, 2025)

Forrester: Is Agentic Commerce A Thing? (October 16, 2025)

Join the discussion

Which experience from your own organisation is most relevant to this story? Share what you learned.