Welcome to session 43! This week Bala Panneerselvam walks through how he uses Claude Code — Anthropic's terminal-based AI tool — for non-technical operational work.
Bala is the Founder of Applied AI Studio and Applied AI Club. He's been using Claude Code daily for two and a half months and shares the real workflows that have made it his biggest productivity unlock since working with AI.
Resources
Session Overview
This session is a hands-on walkthrough of using Claude Code for operational tasks that don't require coding knowledge. Bala demonstrates real workflows from his daily work — processing invoices, generating sales decks, running customer research, and building content pipelines — all from the terminal. The session covers the mental model for thinking about AI assistance levels and why terminal-based tools unlock capabilities that chat UIs can't match.
Key Takeaways
- Claude Code works for non-technical tasks: Invoice processing, sales research, deck generation, and content pipelines all work from the terminal without writing code yourself.
- The 4 Levels of AI Assistance: Assistants (brain in a jar), Agents (brain + hands on one task), Cowork Agents (brain + hands + initiative — you set objectives, AI works, you review), and Autonomous Peers (independent colleagues).
- Terminal beats UI for compound work: The terminal lets you chain outputs, build on previous work, and create reusable skills — things chat UIs can't do well.
- Skills make work reusable: A CLAUDE.md file and skill definitions turn one-off prompts into repeatable workflows that improve over time.
- Brand consistency is automatic: Claude Code reads your brand guidelines, website content, and existing materials to maintain consistency across all outputs.
Topics Covered
The 4 Levels of AI Assistance
- Level 1 — Assistants: "Brain in a jar" — you ask, AI answers. ChatGPT, Claude chat, Gemini.
- Level 2 — Agents: "Brain + Hands" — you instruct, AI acts on one task. Cursor, CRM bots, task-specific tools.
- Level 3 — Cowork Agents: "Brain + Hands + Initiative" — you set objectives, AI works independently, you review outcomes. Claude Code, Cursor (agent mode), Devin.
- Level 4 — Autonomous Peers: "Independent colleague" — you share goals, AI finds its own way. OpenClaw and emerging frameworks.
Why Terminal Over UI?
- Terminal allows compound outputs — one task feeds into the next
- File system access means AI reads your existing docs, brand guidelines, and data
- Skills (reusable instruction files) make workflows repeatable and improvable
- No copy-pasting between chat windows and applications
Live Demo: Invoice Processing
- Dropped a folder of PDF invoices into the terminal
- Claude Code extracted vendor names, amounts, line items, and flagged items over $5,000
- Output: clean CSV spreadsheet ready for accounting — no code written
Live Demo: Customer Research & Sales Deck
- Single prompt: "Research this company and build me a sales deck"
- Claude Code scraped publicly available information, structured a research brief
- Generated a full HTML presentation deck incorporating Applied AI's brand guidelines
- The deck pulled brand colors, voice, and positioning from existing company files automatically
Skills and MCPs (Model Context Protocol)
- A "skill" is a markdown file with instructions that Claude Code follows
- Before/after comparison: one-off prompt vs. reusable skill that produces consistent results
- MCPs connect Claude Code to external tools (Slack, Linear, databases) for real workflows
Compound Learning Framework
- Every task Claude Code completes adds to its context about your work
- CLAUDE.md files accumulate institutional knowledge
- Skills improve over time as you refine them
- The system gets better the more you use it — unlike chat UIs where every session starts fresh
Q&A Highlights
- Security: Data goes to Anthropic's servers (same as any Claude product). You can opt out of training data usage.
- IDE vs Terminal: For coding, IDEs let you review changes easily. For operational work, the terminal is more powerful because it's not limited to code editing.
- Claude Code vs Windsurf/Cursor: Similar models under the hood, but Claude Code's terminal approach enables non-code workflows that IDE tools can't match.
- Getting started: Install Claude Code, create a project folder, add a CLAUDE.md file with context about your work, and start giving it tasks.
Here's the entire recording of the session.