Friday, July 19, 2024

Update for 19th of July on the dynamic agent workflow system I am writing.

 Hi everyone! 👋 I wanted to give you all an update on the dynamic agent workflow program I've been working on. It's been an exciting week with lots of progress! 🚀

🎉 First, I'm thrilled to share that my work on this project led to a job offer from someone I know! It's always motivating to see hard work pay off, and I can't wait to see where this new opportunity takes me.
Now, onto the updates:
  1. Proc Agents 🔗: You can now run a function stored in the JSON config file within the agent itself. These functions will be sandboxed and can access input data, read/write files, and interact with legacy systems using the power of Python scripting. I was able to implement this feature in just 5 lines of actual code! 💪
  2. Template Agents 🛠️: This feature enables you to fill in templates by replacing {keywords} with input from the agent, previous results, step values, or the config file. It's more powerful than using a Python function, as the Proc Agent doesn't have access to internal program variables.
  3. Replicating API Agent Functionality 🔗: I managed to replicate the abilities of the API Agent using 4 Proc Agent steps within a workflow. By starting with the exec_api_call agent and answering questions about its input, I was able to create a solution that maps inputs to either a literal value or a value from the system.
  4. Step Agent ⏭️⏮️: I'm currently working on a Step Agent that can iterate and navigate the workflow forwards and backwards. This will enable skipping over result handling if it's bad, jumping back for multiple results, or retrying network connections before moving to error handling.
  5. Nested Workflows 📦: My next focus is to enable workflows to contain other workflows. This will help replicate the abilities of the API Agent I mentioned earlier, allowing me to remove that class altogether.
The concept of agents with inputs, outputs, and transformative actions applies to workflows as well. I call them agents because AI is an integral part of everything this tool does. When you create a workflow, you're essentially building a new tool that can be used to create even more workflows. This fractal design and self-similarity at every level mirror the complexity of life itself. 🌱 Once nested workflows are supported, it will be much easier to build complex workflows that integrate traditional data stores, AI, and automation in companies. 🚀
That's all for now! Stay tuned for more updates as I continue to refine and optimize this program. 👀

No comments:

Post a Comment