Foundations Module 3 · Choosing Your AI Surface & Workflow

Browser Chat: Your Second Opinion

Last reviewed

Beginner

What you'll learn

~15 min
  • Use plain chat beside your workspace agent as a second opinion
  • Identify what answer-only chat does well, and what it cannot do
  • Tell answer-only chat apart from in-browser builders and workspace agents
  • Compare a browser chat and your installed agent on the same real file

Plain browser chat runs beside your workspace agent, not instead of it. In this lesson you use it as a second opinion: a quick place to brainstorm, learn a concept, or get an outside view on a plan, while the agent keeps working on your files.

Why this matters

In Lesson 3.1 you ran the build loop in your agent, and Lesson 3.2 showed when each surface fits. Chat is the surface most people already know: ChatGPT, Claude, or Gemini in a browser tab. For questions that are not about your files, it is often the quickest and right tool. Knowing exactly where it stops tells you when to switch back to the agent.

Mental model: a second opinion down the hall

Think of browser chat as a knowledgeable colleague down the hall. You can walk over and ask them to:

  • Brainstorm options and ideas
  • Explain a concept at your level
  • Review a plan and point out gaps
  • Draft an email, proposal, or summary

They have never seen your project, though. Chat relies entirely on what you tell it, so the more context you give, the more useful the answer.

Hands-on: three quick conversations

Open one of these in your browser, whichever you already have an account for. Each has a free tier; check current limits.

If you hit a usage limit, switch to another provider. The wording of answers varies, but the patterns are the same.

1. Brainstorm. Type:

I want to build a personal portfolio website. Give me a project plan with 5 steps, with a time estimate for a complete beginner.

Notice that it adapts to “complete beginner”. Planning and brainstorming are chat’s sweet spot.

2. Explain. In a new conversation, type:

Explain the difference between frontend and backend like I’m planning a restaurant. Frontend is what diners see, backend is the kitchen.

Notice how it runs with your analogy.

3. Review. Back in the first conversation, type:

Review this plan. What am I missing? What could go wrong?

Notice that it finds gaps you might not think of. That is the second-opinion role: an outside view before you hand the plan to your agent.

What answer-only chat cannot do

A standard chat is stateless about your work: it knows only what is in the conversation.

  • It cannot open your folders. You can upload a file, but it cannot browse your computer or see your project.
  • It cannot run your project. Some chat tools run code in a browser sandbox, but that sandbox is not your computer.
  • It cannot change your files. Any code it writes, you copy, place, and run yourself; if it fails, you paste the error back and repeat.
  • It forgets in long conversations. If answers drift, start a new conversation and restate the key details.

Your workspace agent from Lesson 1.2 does all four: it reads, saves, runs, and reads its own errors. Chat talks about things; the agent does them. Use both.

⚠Privacy note

What you paste into browser chat goes to the provider’s servers and, depending on your settings, may be used for training. Check the provider’s privacy settings first (Claude calls the switch Model Improvement), and keep credentials, proprietary code, and regulated data out unless your organization’s policy allows it.

Try it yourself: chat vs. your agent, on a real file

You have a workspace agent from Lesson 1.2 and an ai-class folder with the welcome.html page it built. Ask both tools about that page.

  1. In a browser chat (a new conversation), ask:

    What's in my welcome.html page, and how would you make it easier to read on a phone?

    It cannot see the file, so it asks you to paste it or gives general advice about a page it has never seen.

  2. In your agent, opened on ai-class (a new session in the desktop app, or cd ~/ai-class on Mac/Linux/WSL or cd $HOME\ai-class in Windows PowerShell, followed by your start line in a terminal), ask the same question.

    It opens the file, describes what is actually there, and can offer to make the change. If you say yes, refresh the page and check it.

  3. Back in the browser chat, ask something that is not about your files, such as “What makes a page easy to read on a phone?” This is where chat is quick and right.

The pattern: questions about ideas and concepts suit chat; anything about your actual files suits the agent.

🧬In Your Field: Biotechclick to expand

Second opinion in the lab: “I’m designing a CRISPR experiment to knock out gene X in cell line Y. What controls should I include? What are the common pitfalls?” Then let your agent build the analysis in your project folder.

📊In Your Field: MIS / Businessclick to expand

Second opinion in business: “I need to present quarterly sales to the executive team. Which charts tell the clearest story? What should I avoid?” Then ask your agent to build the dashboard from your spreadsheet.

🏛️In Your Field: Government Devclick to expand

Second opinion in government dev: “We’re moving a legacy .NET app to the cloud. What are the FedRAMP implications? What should I ask the security team?” Check every compliance claim against official sources.

⚠AI output is a starting point

Treat chat answers as a draft, not an authority. Check compliance, legal, scientific, or regulatory claims with trusted sources.

When things go wrong

🔧

When Things Go Wrong

Use the Symptom → Evidence → Request pattern: describe what you see, paste the error, then ask for a fix.

Symptom
The AI gives a vague or generic answer
Evidence
Response feels like it could apply to anyone, lacks specifics
What to ask the AI
"Your answer was too generic. Here's more context about my specific situation: [add details]. Give me advice that's tailored to this."
Symptom
The AI misunderstands your question
Evidence
Response addresses a different topic than what you asked
What to ask the AI
"That's not quite what I meant. Let me rephrase: [clearer version of your question]. Focus specifically on [the part that matters]."
Symptom
The AI gives you code but you don't know what to do with it
Evidence
Response contains code blocks but no instructions for where to put them
What to ask the AI
"I'm a beginner. Instead of giving me code, can you explain the concept first? Then tell me step by step what files I'd need to create and what goes in each one."

Where in-browser builders fit

Two other kinds of tool can build things, and it helps to keep the three apart:

Kind of toolExamplesWhat it works onWhere the result lives
Answer-only chatThe conversations you just hadThe conversation and anything you attachIn the chat, as text you copy
In-browser builderClaude Artifacts on claude.ai, ChatGPT Work & Sites on chatgpt.comYour request, plus files you uploadA live page or small app hosted by the provider, shared by link
Workspace agentDesktop agent apps, terminal CLIs, and cloud agents tied to a GitHub projectA project folder on your computer, or a GitHub project on a cloud machineFiles saved in your project, under version control

In-browser builders are legitimate building tools for a quick shareable page, or when you cannot install software. What they cannot do is work in the folders on your computer. Lesson 3.4 compares them with cloud agents and devboxes.

Key takeaways

  • Chat is your second opinion. It brainstorms, explains, reviews, and drafts, beside your workspace agent.
  • Three patterns: generate ideas, explain concepts, review work.
  • It cannot open, run, or change your files. That is your agent’s job, in a desktop app or a terminal.
  • Context matters. Generic questions get generic answers.
  • In-browser builders make shareable pages in the provider’s cloud, without touching your computer’s folders.

Next step: Lesson 3.4 covers zero-install and cloud paths for machines where you cannot install an agent.

KNOWLEDGE CHECK

You want to understand what 'responsive design' means before building a website. What's the best approach?

KNOWLEDGE CHECK

You need a Python script that cleans up a spreadsheet saved in a folder on your computer. What changes if you ask a workspace agent opened on that folder instead of a plain chat?

Search lessons