AI Agent Action Now Up to 50% Cheaper to Run

Modified on Tue, 11 Aug at 12:16 AM
Overview:
The AI Agent Action just got a lot cheaper to run. We have shipped a set of cost optimizations across the agent execution pipeline that cut the cost of every agent run, with no change to agent behaviour, output quality, or setup. In internal testing, agent executions are now 40 to 50 percent cheaper.
What's New
Up to 50 percent lower cost per AI Agent execution
Prompt caching: repeated context across agent steps is now cached instead of being resent on every call, removing a large chunk of redundant token spend.
Leaner system and tool prompts: system and tool instructions have been rewritten to be tighter and more precise, which lowers the token cost of every single run.
Right sized models for supporting calls: internal operations like summarization, structured output, and memory generation now run on lighter models, while the core reasoning stays exactly where it was.
Fewer tokens consumed per execution overall, on top of the per call savings.
Why It Matters
The AI Agent Action has become one of the fastest growing parts of the Workflow AI suite, and this change puts it on an even steeper curve. Cheaper runs mean users can reach for the agent in more places, more often.
The choice between an agent and a long manual workflow now comes down to what fits the use case, not what fits the budget.
The heaviest agents benefit most. Long conversation and multi tool agents see the largest drop.
High volume workflows are now well within reach, so agents can run across a much larger share of a user's automations.
New users get a far lower entry cost to start building with agents.
How to Use
Nothing to do. The optimizations are already live and apply automatically to all new and existing AI Agent Actions.