As we build larger, more capable models, and distill them into smaller, more affordable models, intelligence will only grow in the long term. At that point, creating intelligent agents that operate within your environment is an engineering problem, not a research one.
This approach has been working well so far. By exposing OpenAPI specs to Bedrock Action Groups and leveraging some prompt engineering, the agent can call APIs on behalf of support. I’ve found that even relatively inexpensive models can handle this use case effectively, given its simplicity. I’m not trying to parse data and insert it in MongoDb like the author is here, but rather simple support tasks like please let this customer change ownership of a previous submitted report.
The business value here is significant. We can offload trivial support tasks to less expensive team members, freeing up engineers. Plus, I can deliver this capability without the overhead of building an entirely new UI, which makes it even more practical.