We're open-sourcing Claude Commerce Agents.
This is a blueprint for building shopping and merchant agents, with reference implementations across retail, travel, telecom, and entertainment.
Well thought out third party integrations into agent system (that work) or integrations w legacy, beauracratic software.
Ease of developer adoption (and stickiness/switching cost once in codebase)
Well defined, proprietary “architecture” for building something (ie Agents) that only runs on your platform. Once you build, you can’t switch.
Imagine you have a tool for “send email” but now your AI has a list of 1000 people it needs to email. Relying on the AI to call the send_email tool 1000 times is inefficient, unreliable, and probably wouldn’t work. In code, it’s a for loop.
HOW TO BUILD AGENTS IN 2026
* DO NOT use tools to call api’s/do RAG
Instead, use code.
One tool takes in code, executes in VM that maintains own file system and namespace between executions (Jupyter notebook style)
This is by far the superior way.
Today, we're excited to introduce you to Kafka, the first AI employee.
Kafka is the first step towards our goal of creating an AI worker that is virtually indistinguishable from a remote human employee.
Kafka is the new state-of-the-art on the GAIA Level 3 benchmark which measures AI's ability to complete economically viable knowledge tasks, scoring 77.2% on Level 3 and converging to near-human performance.
We've achieved this by creating a high-fidelity virtual environment for Kafka to work in, giving him his own email, phone and Slack account, and connecting him natively to 1000+ third party applications.
Try Kafka today at https://t.co/dx3faG3hp5.
Check out some of the things Kafka can do in the thread below.
@Jason you aren’t really getting much service, but you are recieving deflated food prices because of low base wages for employees. Sure, you don’t have to tip, I get it, but other cultures don’t tip because other cultures pay livable wages to service workers by increasing food prices.