When people argue about "agents," half the confusion is that they're picturing different things. One imagines a scary autonomous system running their business unsupervised; another means a prompt that reruns every morning. Both are right, because "agent" describes a spectrum of autonomy, not a single tool. Once you can see the four rungs, the choice stops being intimidating and starts being obvious.
Verified July 2026. The tools below are the fastest-moving thing in this whole course — names, features, and prices shift constantly. The four rungs won't. When you go to pick a tool, glance at its current site to confirm the specifics.
The four rungs of autonomy
Rung 1 — Scheduled AI tasks
The gentlest step up from a chatbot: the same prompt runs automatically on a schedule and reports back. You write it once — "every morning, summarize my unread email and give me three priorities" — and it simply happens. Both major assistants offer this now: ChatGPT has Scheduled Tasks and Claude has Scheduled Tasks in Cowork (which arrived in 2026). It's "agent-ish" because it acts without you being there, but the job is fixed. This is where almost everyone should start.
Rung 2 — AI-with-your-files sessions
You point the AI at a folder or a set of documents and give it a real multi-step job: "read these five contracts and build me a comparison table," "reorganize this messy folder and tell me what you moved." It reads, works, and reports — a genuine agent loop, but you kick it off and watch it. This is the sweet spot for one-off jobs that are too fiddly for a single prompt.
Rung 3 — Automation platforms with AI steps
Tools like Zapier and Make connect your apps: a trigger fires, an AI step does some thinking, and the result lands somewhere — a new lead emails you a drafted reply, a form response gets summarized into a spreadsheet. In 2026 both went further than fixed "if-this-then-that" recipes: Zapier Agents let you hand a goal plus a set of connected apps and let the agent choose the actions; Make's Maia assistant builds and runs scenarios from plain-English descriptions. Powerful for wiring tools together — but it's the rung with the most setup, so it's not where a beginner starts.
Rung 4 — Full autonomous sessions
The top rung is a general-purpose agent you give a goal and largely let work: Claude's Cowork and Claude Code are the well-known examples — you describe an outcome, and it plans, uses tools, checks itself, and delivers, checking in when it needs a decision. ChatGPT's Agent mode (which drives its own browser and virtual computer) sits near here too. Remarkable, and genuinely useful for bigger jobs — but the most autonomy means the most need for the manager habits in Lesson 5. Don't start here; graduate here.
July 9, 2026: the top rung got a new name on the OpenAI side — ChatGPT Work, a desktop agent with Codex built in that supersedes Agent mode as OpenAI’s flagship agent (and renamed the old desktop app “ChatGPT Classic”). The ladder itself is unchanged — ChatGPT Work sits at rungs 3–4, and the manager habits in Lesson 5 matter for it just as much. Full comparison: ChatGPT Work vs Claude Cowork.
Which rung, and what it costs
Verified July 2026 — treat these as ballparks, not quotes, and check each tool's pricing page before paying.
| Rung | Example tools | Rough cost | Best for |
|---|---|---|---|
| 1 · Scheduled tasks | ChatGPT Scheduled Tasks · Claude Cowork Scheduled Tasks | Free tiers exist; full features on the ~$20/mo plans (ChatGPT Plus, Claude Pro) | Your first agent. A recurring brief or summary that just shows up. |
| 2 · Files sessions | Claude (Projects / Cowork) · ChatGPT with file access | Works on free tiers; heavier use wants a ~$20/mo plan | One-off multi-step jobs on your own documents. |
| 3 · Automation platforms | Zapier · Make | Free starter tiers; paid plans scale with usage (Make tends cheaper at volume) | Connecting several apps so a result moves between them automatically. |
| 4 · Autonomous sessions | Claude Cowork / Claude Code · ChatGPT Agent mode | On the ~$20/mo plans; heavy users go higher | Bigger, open-ended jobs — once you've built the review habit. |
Stand on Rung 1. Pick one recurring task, write it as a scheduled prompt in whichever assistant you already pay for (or the free tier), and let it run for a few mornings. You'll learn more about managing an agent from one real scheduled task than from a month of reading about autonomous ones — and you can climb the rungs later, once the habit is yours.
Notice what didn't make the table: anything requiring code, a credit card for an enterprise plan, or a "book a demo" button. Everything you need to start is a rung-one feature inside tools you likely already have open.
- Open the AI assistant you use most and find its scheduled-tasks feature (in ChatGPT, the task/clock option; in Claude, Cowork's scheduled tasks).
- Don't build anything yet — just confirm it's there and note where it lives.
- Hold the recurring task you wrote down in Lesson 1 next to it. That's the rung, and that's the tool. Lesson 3 shows you how to brief it.