AI Agent action now upto 50% cheaper to run
The AI Agent Action has just become significantly more affordable to operate. We have implemented a series of cost optimisations throughout the agent execution pipeline that reduce the cost of every agent run, without altering agent behaviour, output quality, or setup. During internal testing, we found that agent executions are now 40 to 50 per cent cheaper.
- Up to 50 per cent lower cost per AI Agent execution
- Prompt caching: repeated context across agent steps is now cached instead of being resent on every call, eliminating a significant amount of redundant token expenditure.
- Lean system and tool prompts: system and tool instructions have been revised to be more concise and precise, thereby reducing the token cost of every single run.
- Right-sized models for supporting calls: internal operations, such as summarisation, structured output, and memory generation, now employ lighter models, while the core reasoning remains unchanged.
- Fewer tokens consumed per execution overall, in addition to the savings made per call.
The AI Agent Action has become one of the fastest-growing components of the Workflow AI suite, and this enhancement propels it further. Cheaper executions empower users to utilise the agent in a wider variety of contexts and more frequently.
The decision between using an agent and a lengthy manual workflow now hinges on the suitability for the task, rather than financial constraints.
The most substantial benefits are realised by the heaviest agents. Long conversations and multi-tool agents experience the greatest cost reductions.
High-volume workflows are now easily achievable, enabling agents to operate across a much larger portion of a user’s automations.
New users face a significantly lower entry cost when starting to work with agents.
No action is required. The optimisations are already live and automatically apply to all new and existing AI Agent Actions.