Bots, assistants, agents | Master AI Automation in 4 hours Master AI Automation in 4 hours Course About Ayush Modules Sample chapter Toolbox The Microcap Minute Classroom / Module 10: Build Your Own Assistant / Chapter 1 Bots, assistants, agents Watch first, then read. Same lesson, your pace. What you will learn – The one axis that separates the three words – Where today’s tools sit on it – Why the distinction matters practically One axis: who decides next? Forget marketing. Sort every AI product by how much it decides without asking you : Chatbot Assistant Agent Decides Nothing; replies turn-by-turn Within your task Multi-step plans + actions Acts on world No (no tools) Maybe (tools you trigger) Yes, runs commands, browses, edits files Stops when Every message Task delivered Goal reached or budget hit Examples plain chat windows Custom GPTs, Claude Projects with files OpenCode, Claude Code, n8n flows A chatbot talks. An assistant is a chatbot with your context and maybe some buttons. An agent is an assistant handed hands and judgement: given “fix this bug”, it reads files, edits, runs tests, reads failures, retries. Module 5’s MCP round-trip was agent mechanics in miniature: model decides → tool acts → observe → repeat. Module 6’s pipelines were agents with the judgement removed (deterministic steps). This module assembles the full thing. Why the axis matters Two practical consequences: Autonomy trades convenience for risk. Agents that edit real files can edit the wrong ones; guardrails (Chapter 3) are how assistants earn promotion to agents. Most people’s “AI assistant” is actually a chatbot with homework. Loading instructions and files (next chapter) upgrades it more than switching models does. Try it yourself Classify five AI things you’ve touched this month (any apps/bots/features) on the table’s four rows. Then find one assistant-behaving-as-chatbot in your own life, same tool, but imagine it with your syllabus loaded and standing instructions, and write the two-minute upgrade plan in learn/agent-taxonomy.md . Key takeaways – Autonomy is the axis: chatbots reply, assistants contextualise, agents act. – MCP loops = agent mechanics; automations = agents minus judgement. – Upgrades come from context + guardrails before they come from models. Download the exercise sheet (PDF) Module workbook (PDF) ← Module 10 index Next: No-code first → Classroom / Module 10: Build Your Own Assistant / Chapter 1 Bots, assistants, agents What you will learn – The one axis that separates the three words – Where today’s tools sit on it – Why the distinction matters practically One axis: who decides next? Forget marketing. Sort every AI product by how much it decides without asking you : Chatbot Assistant Agent Decides Nothing; replies turn-by-turn Within your task Multi-step plans + actions Acts on world No (no tools) Maybe (tools you trigger) Yes, runs commands, browses, edits files Stops when Every message Task delivered Goal reached or budget hit Examples plain chat windows Custom GPTs, Claude Projects with files OpenCode, Claude Code, n8n flows A chatbot talks. An assistant is a chatbot with your context and maybe some buttons. An agent is an assistant handed hands and judgement: given “fix this bug”, it reads files, edits, runs tests, reads failures, retries. Module 5’s MCP round-trip was agent mechanics in miniature: model decides → tool acts → observe → repeat. Module 6’s pipelines were agents with the judgement removed (deterministic steps). This module assembles the full thing. Why the axis matters Two practical consequences: Autonomy trades convenience for risk. Agents that edit real files can edit the wrong ones; guardrails (Chapter 3) are how assistants earn promotion to agents. Most people’s “AI assistant” is actually a chatbot with homework. Loading instructions and files (next chapter) upgrades it more than switching models does. Try it yourself Classify five AI things you’ve touched this month (any apps/bots/features) on the tab
Bots, assistants, agents
Written by
in