VS Code · Claude Code · Markdown

AI Agent Workflow in VS Code

Build an AI agent that thinks, plans, and executes tasks autonomously.

Install Claude Code Download Visual Studio Code

The workspace

Two-column layout: Explorer on the left, Claude Code chat on the right.

  • File
  • Edit
  • Selection
  • View
  • Go
  • Run
  • Terminal
  • Help
my-first-agent — Visual Studio Code
CLAUDE.md
research-report.md
ai-agents-2026.md
◆ Claude
# Project Context
This is my AI agent workspace. I use it for research, content
creation, and productivity workflows.

# About Me
I create content about technology and productivity. My audience is
people who want practical, no-nonsense tutorials. I prefer clear,
jargon-free output.

# Rules
- Always ask clarifying questions before starting a complex task
- Show your plan and steps before executing
- Keep reports and summaries concise - bullet points over paragraphs
- Save all output files to the output folder
- Cite sources when doing research

# Project Structure
- workflows/ - Workflow instruction files (plain English recipes the agent follows)
- output/ - Finished deliverables (reports, drafts, analysis)
- resources/ - Reference docs and templates

# research-report.md

## Objective
Given any topic, research it thoroughly, organize the findings,
and produce a clean, structured report.

## Steps
1. Ask clarifying questions (scope, audience, depth)
2. Plan the research
3. Search the web and gather sources
4. Organize notes in /output
5. Synthesize the report
6. Add key takeaways + recommended next steps

## Rules
- Cite sources inline
- Bullet points over paragraphs
- Save outputs to /output

# AI Agents in 2026 — Hype vs. Reality

## Executive Summary
- Agents have moved from demos to daily workflows in research, ops, and coding.
- The biggest gains are narrow: scheduling, research, and code review.
- Hype outpaces reality on "fully autonomous" general agents.

## What people actually use them for
- Research & report generation
- Inbox + calendar triage
- Coding assistants in IDEs
- Customer support deflection

## Top 5 AI Agent Tools
See comparison table below.

## Key Takeaways
- Start narrow. Pick one workflow.
- Always add a human check-in step.
- Memory + context files (like CLAUDE.md) outperform clever prompts.

## Recommended Next Steps
- Pick one repetitive task this week.
- Write a workflow recipe in plain English.
- Run it. Refine it. Save it to your library.
I want to build a workflow where I can give you any topic, and you'll research it thoroughly, organize the findings, and produce a clean, structured report. Before you start researching, you should ask me clarifying questions about the scope, the audience, and what I specifically want to know.
◆ Claude · Plan mode

Got it. Here's the plan I'd follow for any topic you give me:

  1. Ask clarifying questions (scope, audience, depth)
  2. Draft a research outline
  3. Search the web and collect sources
  4. Organize notes into /output
  5. Write the structured report

Shall I save this as workflows/research-report.md?

I also want the report to include a section at the end with key takeaways and recommended next steps.
◆ Claude

Added. The workflow now ends with Key Takeaways and Recommended Next Steps.

I want to research the current state of AI agents in 2026 — what are people actually using them for, what's working, what's overhyped, and where things are heading.
◆ Claude

Before I start, a few clarifying questions:

  • How broad or narrow should the scope be?
  • Who is the intended audience?
  • How in-depth should this be?
  • Any specific tools or platforms to include?
  • What time range should I focus on?
Keep it broad. I want a hype vs. reality breakdown. The audience is tech-savvy people who are curious about AI but not necessarily developers. Medium depth — 3 to 5 pages. Include specific tools and platforms where relevant. Stick to the last 6 to 12 months. And format it as a structured report with clear sections and bullet points.
◆ Claude
Agent Todo
  • Search: "AI agents 2026 adoption"
  • Read 8 sources, save notes to /output
  • Draft outline
  • Write ai-agents-2026.md
  • Add key takeaways + next steps
The executive summary is a bit long. Trim it to three bullet points.
◆ Claude

Done — trimmed to three bullets.

Add a comparison table of the top five AI agent tools mentioned in the report.
◆ Claude

Added to output/ai-agents-2026.md:

ToolBest forStrengthWatch out
Claude CodeDev workflowsPlanning + filesIDE only
ChatGPT AgentsGeneral tasksBroad toolsInconsistent
CursorCodingInline editsCode-focused
DevinLong tasksAutonomyCost
n8n + LLMAutomationsIntegrationsSetup time
Plan mode ⇧ Tab
◆ Claude Code Plan mode UTF-8 Markdown Ln 1, Col 1

Three levels of using AI

From chat to building to true agentic workflows.

Level 01

Chat.

You ask a question. You get an answer. One and done.

What is an AI agent?
An AI agent is a system that can autonomously perform tasks…
Level 02

Building.

You direct every step. The AI writes the code.

You
The Director → AI
The Builder
do this→then this→then this
Level 03

Agentic.

You describe a goal. The AI figures out how.

"Here's what I need to accomplish…"
PlansAsksResearchesAdaptsDelivers
THIS IS WHAT WE'RE BUILDING TODAY

What makes an agent, an agent

It follows a process.

Not just answering a single question. Working through a series of steps to get to a result.

UnderstandResearchOrganizeSynthesizeDeliver

It makes decisions.

Chooses what to do next based on what it finds. If one approach isn't working, it pivots.

Your goal
Try A ✕ pivots Try B ✓
found something unexpected → adjusts

It asks before it assumes.

Instead of guessing what you want, it asks clarifying questions first.

  • What specific aspects should I cover?
  • Who is the intended audience?
  • How in-depth should this be?
That one change alone makes the output 10x better.

The building blocks

Workflows.

Plain English files that describe a process. Think of them like a recipe.

# research-report.md
## Objective
## Steps
## Rules

The Agent.

Claude Code itself. It reads, thinks, and decides. You don't program it. You just give it clear instructions.

Workflow → Claude Code
reads · thinks · decides → Result

Tools.

What the agent uses to get things done. Built in. No extra setup needed.

Read + Write FilesBUILT IN
Run CommandsBUILT IN
Search the WebBUILT IN
Want more? Connect to Gmail, Notion, databases via MCP — next video.

Build your first agent

Click any step to expand.

Grab VS Code from the official download page if you don't already have it.

Open Extensions in VS Code and search for Claude Code, then install.

A clean folder per agent keeps memory, workflows, and output organized.

This is the agent's automatic memory. Drop it at the root of the project.

Add sections: Project Context, About Me, Rules, Project Structure.

Plan mode forces the agent to show its plan before touching files.

Tell the agent what you want, not how to do it.

Add requirements. Iterate until the plan looks right.

Choose "Yes, and auto-accept" or "Yes, and manually approve edits".

The agent will check in. Answer specifically.

Files appear in the Explorer as the agent works.

"Trim the executive summary to three bullets." "Add a comparison table."

Hover an earlier message to rewind and rephrase your prompt.

5 mistakes to avoid

1 / 5

No CLAUDE.md

Without

  • Generic output
  • Guessing
  • Repeats mistakes

With

  • Structured output
  • Asks first
  • Learns over time
2 minutes of setup saves hours of frustration.
2 / 5

Too vague

Vague

"Write me something about AI."

Specific

"3-page hype vs reality report on AI agents, last 12 months, bullet points."

Specificity is everything.
3 / 5

No plan mode

Skip planning

Goal → Edits files → Surprises

Plan first

Goal → Plan → Approve → Edits

30 extra seconds. Always worth it.
4 / 5

No questions

Assumes

Guesses scope → wrong output

Asks first

Clarifies → exactly what you need

Always build in that check-in step.
5 / 5

All at once

All at once

10 workflows, no testing

One at a time

Ship one. Refine. Repeat.

Don't automate your entire life in one weekend.

4 pro tips

1 / 4

Organize.

my-first-agent/
├── CLAUDE.md
├── workflows/
├── output/
└── resources/
Organized as you scale.
2 / 4

Watch it work.

Agent Todo List
  • Research
  • Draft
  • Polish
Catch mistakes early. Before they snowball.
3 / 4

Be specific.

Start over

"Redo the whole thing."

Targeted fix

"Trim section 2 to 3 bullets."

Don't start over. Tell it exactly what to fix.
4 / 4

Build a library.

library/
├── CLAUDE.md
├── research-report.md
├── content-brief.md
└── weekly-digest.md
Copy your best files into new projects. Build a library.

Start building your first agent

Two minutes of setup. Hours of leverage.

Install Claude Code Download Visual Studio Code