I am going through the reflection design pattern for agentic AI → lab chart generation. I wanted to understand with Claude in the picture now, isn’t it possible to prompt it in plain english with albeit the same flow but not the python code as suggested in the simulator lab?
I am not a coder and have been using claude code recently as a product manager. This might be a very basic question, but your reasoning would be helpful.
Hi Samina_Ali,
Thanks for your question. Here’s my two cents:
As coding models get better, for simple and non-critical applications, there will be less need to carefully control the process.
For complex or critical applications, it will always remain important to be able to control the process.
The lab constitutes some kind of middle ground. It is always good to know how to be able to control the process, even if it may not be essential in practice.
I hope this answers your question.
Thanks reinoudbosch, I had a follow up:
Lets say its a not very complex task, then by using claude code - how do I simulate agentic workflow of having say different LLMs as explained in the reflection pattern.
E.g: I am creating a chatbot for chart generation, I prompt claude to create a visual representation of the csv I give, claude generates. Now, do I then again ask it to reevaluate from a perspective of a visual expert on lines of clarity in the chart, simplicity in understanding… ’ ? OR do I somehow create these 2 agents by specifying in the beginning to act as a coding agent C, and then as a visual expert V, and then sought of lay down the workflow for it?
Hi Samina_Ali,
You could do both, although the second method (if using a single prompt) will have a larger chance to derail. If you are going to chain agents, you might want to have a look at the courses on LangChain/LangFlow or crew.ai.
I recommend you follow the course as designed, without using alternate tools.