This topic was automatically generated from Slack. You can find the original thread here.
Hi, I’ve not had a chance to use String.com yet, however, I was wondering if it actually builds AI agents, or does it build workflows instead
Thanks for your question! Based on the information provided, String.com is not referenced in the Pipedream documentation, so I can’t speak specifically about that product. However, I can clarify how Pipedream works in the context of AI and workflows: - Pipedream is a platform for building workflows that connect APIs, apps, and services. You can use it to automate tasks, integrate with AI APIs (like OpenAI, ChatGPT, etc.), and build complex automations. - Pipedream lets you build workflows, not standalone AI agents. However, you can use Pipedream to connect to AI APIs and build workflows that interact with, or even orchestrate, AI-powered tasks (like chatbots, summarization, classification, etc.). - If you want to build an AI agent (like a chatbot or automated assistant), you can use Pipedream to connect to AI APIs and manage logic, but the “agent” itself is really a workflow that leverages those APIs. If you have more details about String.com or want to know how to build AI-powered workflows on Pipedream, let me know! If you need more support, you can always visit Support - Pipedream.
maybe you can help here?
, from my understanding, String builds workflow, which you can use as an AI Agent.
For example, for the Q&A AI Agent on Telegram you can use String to build a workflow that:
- Triggered from Telegram command with a question,
- Fetch your internal data from Airtable
- Uses OpenAI - Chat action with the data context from Airtable to produce answer to the question
- Then send the answer back to Telegram using Telegram - Send Message action
Ahh ok got it. To me, an AI agent is an LLM layer that has access to tools (such as workflows) and can orchestrate the right way forward without it being limited to Step 1, Step 2, etc. that a workflow would. Does that make sense?
Regardless, cool tool!
Ahh ok got it. To me, an AI agent is an LLM layer that has access to tools (such as workflows) and can orchestrate the right way forward without it being limited to Step 1, Step 2, etc. that a workflow would. Does that make sense?
I understand your thinking. There’re quite some frameworks that allow this such as autogen or crewai, though they usually requires coding experience and quite a lot of thinking on the prompting.
From my experience using LLMs, they are not really good at solving complex problem that requires multiple steps and lots of context fetching (even with multiple LLM agents in place).
I usually use LLM for small tasks (such as consolidate data into report, summarize, labeling, etc…) in a bigger automated workflow, where we now the exact steps that will be executed.
- I think @UN1BV42B1 did a good job of outlining our perspective on the spectrum from tasks to self-directed, autonomous AI agents in this podcast that dropped today https://youtu.be/t3q5hCv_Kyo?si=jkdJLt6S2I1BW6Qk&t=1260
String can build self-directed, tool-calling agents where the LLM decides which tools to use, but we’re in the early days and we’re investing in the foundations to make that experience better. E.g., you can try this prompt in String (or the code I shared in this thread) to generate an agent in code that calls remote MCP tools. I’ll share a demo soon for a tool calling agent I generated with String to create and edit Notion docs by chatting with the agent from Slack.