Coding agents for civilians | 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 4 Coding agents for civilians Watch first, then read. Same lesson, your pace. What you will learn – Installing and wiring OpenCode with free models – The supervision loop for agent work – What to delegate first (and what never) Your agent, your terminal OpenCode (opencode.ai) is an open-source coding agent that lives where Module 3 trained you: the terminal. It reads your project files, writes code, runs commands, reads its own errors and iterates, under your supervision. Because it’s provider-agnostic, your free-first stack plugs straight in : any OpenRouter :free model (Module 4 Chapter 6) or API keys from Kimi/Qwen/Claude/OpenAI. Install (one line per its docs, curl -based installer or npm i -g opencode-ai ), then from any project folder: cd projects/api-lab opencode It opens its terminal UI. Wire a model once via /models (pick an OpenRouter free model; paste key from .env habits). Then just talk: “Add a retry-with-backoff wrapper around ask.py’s API call, then run it once to prove it still works.” Watch it plan → edit → run → report. Every file change shows as a diff you approve or reject. The supervision loop Agents are interns with root access, supervise like it matters: Small asks first: one function, one fix, build trust evidence before big jobs Read every diff: approval buttons are not decoration; this is the permission prompt of Module 5 grown up Git before agents: initialise git init in any project an agent touches, instant undo for everything Budget by turns: decide upfront (“max ~15 tool calls”); stop and reassess when hit Off-peak for heavy runs: agent loops burn tokens; free models + quiet hours keep costs at ₹0 What to delegate first Good starter delegations: boilerplate writing, test generation, error explanation, refactoring small functions, README/docs drafting, dependency updates. Keep for yourself (for now): architecture decisions, anything touching .env /secrets/credentials, deletions beyond one folder, and final review of whatever ships. As evidence accumulates, widen the leash gradually, that’s Chapter 3’s promotion path applied to code. Try it yourself Install OpenCode, wire a free model, and git init inside projects/api-lab . Delegate three small tasks (e.g., add docstrings to ask.py; write a test that checks word_count from Module 5; refactor receiver.py’s signature check into a helper). Read each diff properly before approving. Log time-saved-vs-supervision-time in learn/opencode-v1.md . Key takeaways – OpenCode = terminal-native, provider-agnostic coding agent; free models drive it fully. – Supervise: small asks, read diffs, git first, budgeted turns, off-peak runs. – Delegate boilerplate/tests/explanations first; keep secrets, architecture and final review. Download the exercise sheet (PDF) Module workbook (PDF) ← Prev: Teaching it your world Module 10 index → Classroom / Module 10: Build Your Own Assistant / Chapter 4 Coding agents for civilians What you will learn – Installing and wiring OpenCode with free models – The supervision loop for agent work – What to delegate first (and what never) Your agent, your terminal OpenCode (opencode.ai) is an open-source coding agent that lives where Module 3 trained you: the terminal. It reads your project files, writes code, runs commands, reads its own errors and iterates, under your supervision. Because it’s provider-agnostic, your free-first stack plugs straight in : any OpenRouter :free model (Module 4 Chapter 6) or API keys from Kimi/Qwen/Claude/OpenAI. Install (one line per its docs, curl -based installer or npm i -g opencode-ai ), then from any project folder: cd projects/api-lab opencode It opens its terminal UI. Wire a model once via /models (pick an OpenRouter free model; paste key from .env habits). Then just talk: “Add a retry-
Coding agents for civilians
Written by
in