Microsoft adds CodeAct to Agent Framework, letting AI agents complete multi-step tasks in one go instead of many separate requests

Microsoft integrated CodeAct—a technique that combines multiple AI tool calls into executable code—into its Agent Framework. Instead of an AI agent asking for permission after each step (like 'should I fetch this data?' then 'should I process it?' then 'should I save it?'), CodeAct lets it plan and execute the full sequence in a single request to the AI model.

This cuts latency and token usage (the input/output units that AI services charge for). When an agent previously needed 10 separate model calls, CodeAct can collapse that into 1. For users, this means faster responses, lower API bills, and fewer seconds waiting for an answer.

Teams building AI assistants for customer service, data analysis, or automation can now build faster, cheaper bots. Developers can access this in the open-source Semantic Kernel framework without waiting—meaning startups and enterprises building internal agents can deploy tighter, more cost-effective tools immediately.