# I started thinking about the future of AI agents after using Claude — how far can Claude actually go? — retained record

> Retained reddit record from reddit:ClaudeAI.

Canonical: https://highsignal.app/data/records/0ccd1a1ac6de18ec

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeAI)  published 2026-10-07 17:56 ingested 2026-10-08 02:36

[original source ↗](https://www.reddit.com/r/ClaudeAI/comments/1x03ers/i_started_thinking_about_the_future_of_ai_agents/)

## retained text



```
I started thinking about this after having a conversation with Claude about how chatbots have evolved.

Originally, chatbots were mainly built to answer FAQs, search information, and provide basic customer support. But with models like Claude becoming much more capable, I started wondering where the boundary between a **chatbot and an actual operating agent** really is.

For example, what if Claude wasn't just answering:

>“How do I return my order?”

but could actually understand the user's intent and perform the task — open the return flow, select the relevant order, submit the request, update the user, and handle the next step?

That made me think about websites and apps differently.

Imagine a business using Claude as an agent where users could say:

* “Add this product to my cart.”
* “Find me the cheapest option.”
* “Track my order.”
* “Raise a complaint.”
* “Take me to the page where I can change my address.”
* “I want to return this product.”

And eventually, potentially doing these actions through **voice**, rather than navigating the entire UI manually.

I'm curious specifically about **Claude's current capabilities and limitations here**.

How close are we to Claude or Claude-powered agents reliably operating real websites and applications end-to-end?

And what are the biggest blockers right now — **permissions, security, hallucinations, authentication, user intent, or giving an AI permission to perform irreversible actions?**

I'm not building a product around this right now. I'm genuinely curious after experimenting with Claude and thinking about where this technology could go.

**For people actually working with Claude/Anthropic: where do you think the line currently is between a Claude chatbot and a true operating agent?**
```



document fetched 2026-10-08 00:22 · canonical https://reddit.com/r/ClaudeAI/comments/1x03ers/i_started_thinking_about_the_future_of_ai_agents
