Building Software with AI Agents

Coding agents write fast. The new bottleneck is whether anyone's paying attention.

A running collection of writing, tutorials, and courses on agentic engineering: using AI coding agents to ship software. The agent does the typing. You stay the engineer, writing the briefs, reading the diff, making the calls.

Best Practices, Courses & Discussions

Workflow, advice, and writing that isn't tied to one specific tool.

Practical Prompt Engineering

Course Frontend Masters

Sabrina Goldfarb teaches prompting strategies that transfer across Cursor, Copilot, and Claude.

Vibe Coding for Developers

Course Coursera (Edureka)

Multi-tool specialization on prompt engineering, AI-assisted editing, full-stack agents, and shipping production code.

AI Development Tools

Course Codecademy

AI-assisted dev with Codex CLI, spec-driven development, TDD, and context engineering — the practice past basic vibe coding.

Agentic Coding Recommendations

Best Practices Armin Ronacher Jun 2025

Practitioner playbook: hand the agent a job, pick boring stacks, keep tools fast and observable.

Revenge of the junior developer

Discussion Steve Yegge · Sourcegraph Mar 2025

Yegge's map of the six waves from autocomplete to agent fleets and how to ride them as a working dev.

Resources by Tool

Quickstarts, walkthroughs, and tips for each tool.

Claude Code Deep Mastery

Tutorial YouTube · IndyDevDan

Long-running playlist on planning, building, and agent orchestration patterns for shipping with coding agents.

Claude Code overview

Tutorial Anthropic docs

Official intro covering install paths, supported IDEs, and where Claude Code fits in your workflow.

Codex Quickstart

Tutorial OpenAI docs

Install Codex CLI, sign in with ChatGPT or API key, and run your first repo-aware task.

Codex CLI reference

Tutorial OpenAI docs

Full CLI surface: approval modes, /model switching, screenshot input, and sandbox controls.

Cursor Quickstart

Tutorial Cursor docs

From install to first useful edit: codebase Q&A, inline edits, and reviewing agent output.

Lovable Quick start

Tutorial Lovable docs

Spin up your first Lovable project, connect Supabase, and ship a working full-stack app.

Introduction to Bolt

Tutorial Bolt.new docs

How Bolt's in-browser WebContainers run your prompted project end to end, no setup required.

v0 Quickstart

Tutorial Vercel docs

Describe a UI, iterate with chat, then push the generated components into your Next.js repo.

UI with v0

Course Vercel Academy

Free Vercel lesson on prompting v0 to produce production-grade shadcn/ui components.

Replit Agent docs

Tutorial Replit docs

Reference for Plan mode, browser-based self-testing, and connecting Agent to external data.

Devin: Get started

Tutorial Devin docs

Onboard Devin, scope your first task well, and understand when to use it vs. an IDE agent.

Jules: Getting started

Tutorial Google docs

Connect Jules to a GitHub repo, submit your first async task, and review its PR.

GitHub Copilot in Practice

Course Pluralsight

Eight-course path on prompting Copilot for reviews, tests, refactors, and secure development.

Aider: Installation

Tutorial Aider docs

Install aider with pipx, point it at Claude or GPT, and start pairing in your terminal.

Aider Documentation

Tutorial Aider docs

Full reference for repo maps, model selection, edit formats, and Git-backed commit flow.

Aider tutorial videos

Tutorial Aider docs

Curated community videos showing real refactors, test-writing, and multi-file edits with Aider.

Cline Documentation

Tutorial Cline docs

Install Cline in VS Code, pick a provider, and use Plan/Act mode to control agentic edits.

What is Continue?

Tutorial Continue docs

Overview of Continue's Agent mode, context providers, and source-controlled config model.

Continue Quick Start

Tutorial Continue docs

Install the VS Code or JetBrains extension, pick a model, and run your first Agent task.

Roo Code Docs

Tutorial Roo Code docs

Reference for Roo's multi-mode agents, custom instructions, and per-mode model routing.

Installing Roo Code

Tutorial Roo Code docs

Install Roo Code from VS Marketplace or Open VSX and pick a provider for your first task.

OpenCode docs

Tutorial OpenCode docs

Install the terminal agent, connect a provider, and run build vs. plan agents on your repo.

sst/opencode on GitHub

Tutorial GitHub · sst/opencode

Source repo for OpenCode with config examples, LSP setup notes, and provider matrix.

Goose Documentation

Tutorial Block / Goose docs

Install Goose desktop or CLI, connect a model, and add extensions for your workflow.

Using Agents in Warp

Tutorial Warp docs

Toggle Agent Mode, hand off multi-step shell tasks, and inspect Warp's auto-correcting plans.

Zed Agent Panel

Tutorial Zed docs

Open Zed's agent panel, add @-context, and follow the agent's edits in real time.

Zed AI overview

Tutorial Zed docs

How Zed wires up providers, edit prediction, inline assistant, and external agents.

External Agents in Zed

Tutorial Zed docs

Run Claude Agent, Gemini CLI, or Codex inside Zed's agent panel via the ACP adapter.

How to use the Agent Panel in Zed

Tutorial YouTube · Zed Industries

Video walkthrough of Zed's Agent Panel: configuring providers, prompting edits, and reviewing change hunks.

Kiro: Get started

Tutorial Kiro docs

Install Kiro, sign in, and learn its spec-first flow: requirements, design, and tasks files.

Your first Kiro project

Tutorial Kiro docs

Build a feature end-to-end in Kiro: generate the spec triad and let the agent execute tasks.

Factory: Quickstart

Tutorial Factory docs

Install the droid CLI, sign in, and run your first specialized droid against a repo.

How to Code with Droids

Best Practices Medium · James Buckhouse

Mental model for delegating to droids: scoping, briefing, and accepting work products.

Gemini CLI on GitHub

Tutorial GitHub · google-gemini/gemini-cli

Official open-source repo for the Gemini terminal agent — install, configure, and contribute.

Creators to follow

Writers, builders, and YouTubers worth a follow.

Andrej Karpathy x · @karpathy Coined "vibe coding"; posts the canonical hands-on takes on building real software with LLMs. Simon Willison blog · @simonwillison Near-daily deep notes on every coding-agent release, model, and CLI workflow. Theo Browne youtube · @t3dotgg Same-week reviews of Cursor, Claude Code, and Codex with opinionated full-stack TypeScript context. Fireship youtube · @fireship Fast, technically tight breakdowns of every new agentic-coding tool the day it ships. ThePrimeagen youtube · @ThePrimeagen Skeptical, terminal-pilled takes on AI coding tools and his "99" pattern as a vibe-coding alternative. IndyDevDan youtube · @indydevdan Anti-hype, tactical Claude Code playbooks — hooks, sub-agents, multi-agent observability. Michael Truell x · @mntruell Cursor's CEO; shipping notes and product philosophy from the company defining the AI-IDE category. Guillermo Rauch x · @rauchg Vercel CEO showing how v0 is used in real Git workflows — prompt-to-PR patterns at production scale. Birgitta Böckeler blog · @birgittabockeler Thoughtworks' lead for AI-assisted software delivery; long-form memos on context engineering and agent workflows. Mckay Wrigley x · @mckaywrigley Practical workflows for shipping with Claude Code, Codex, Cursor, and Gemini CLI — prompts, recipes, parallel agents. Boris Cherny x · @bcherny Creator of Claude Code; posts setup tips, internals, and how the team actually uses it day-to-day. Geoffrey Huntley blog · @ghuntley Originated the "Ralph loop" autonomous-agent pattern; field notes from running coding agents at the edge. Steve Yegge blog · @Steve_Yegge Sourcegraph Amp lead writes the long-form essays mapping where coding agents and agent fleets are headed. Armin Ronacher blog · @mitsuhiko Flask creator's sharp, skeptical essays on agent design, tool ergonomics, and AI-PR review bottlenecks.

Stay Current

Newsletters and podcasts that cover what's shipping each week.

Dan Shipper newsletter · @danshipper Every's CEO chronicling how a team of 15 ships multiple AI products with engineers writing almost no code by hand. swyx newsletter · @swyx Latent Space podcast / newsletter; interviews the people actually building coding agents and harnesses. Gergely Orosz newsletter · @pragmaticengineer The Pragmatic Engineer; deep-dives on AI tooling, agent workflows, and what AI is doing to the craft. Syntax podcast · @syntaxfm Wes Bos and Scott Tolinski's web-dev podcast; regular Claude Code, Cursor, and agent-mode episodes. The Changelog podcast · @changelog Long-running software interview podcast with frequent episodes on coding agents and how teams actually use them. TLDR AI newsletter · @tldrnewsletter Daily 5-minute roundup of AI news, papers, and tooling — a fast scan to stay on top of the field.

Frequently Asked Questions

Short answers grounded in the work of practitioners shipping with coding agents today.

What is agentic engineering?

It's the practice of using AI coding agents to ship software while staying the engineer. You decide what gets built. You write the briefs. You read the diffs. The agent handles the typing and most of the rote work. The thinking part stays yours. Karpathy framed it as orchestrating agents and acting as oversight: you're not writing the code directly 99% of the time, but you're still responsible for it.

Source: Andrej Karpathy, via The New Stack

How is it different from AI pair programming?

AI pair programming in its classic Copilot-autocomplete sense is real-time and granular: you're at the keyboard, the AI suggests the next line or block, you accept or reject. The human stays the primary author. Agentic engineering is delegation. You write a brief, the agent does the work end-to-end (often asynchronously), and you review the result. The agent's the author. You're the director who decided what gets built and verifies it did. Modern Copilot ships both modes now, so the line lives in how you use it.

Source: GitHub Copilot documentation

How is it different from vibe coding?

Same tools, different relationship to the code. Vibe coding is Karpathy's casual mode: describe what you want, let the agent run, don't really read the diff. Fine for low-stakes work where being wrong is cheap. Agentic engineering is what you do when the code matters: when it handles secrets, money, or other people's decisions. Same tools. Different stakes.

Source: Simon Willison

Where do I start?

Pick one tool and stay with it for a week. The picks worth starting with: Claude Code or Codex if you live in the terminal, Cursor for a dedicated AI-native editor, or VS Code with the Claude Code, Codex, or GitHub Copilot extensions if you want to stay in your current setup. Pick a small task you already know how to do. Watch the agent attempt it. Read every line of the diff. Anthropic's recommendation: explore first, then plan, then code. Start with scope you can verify quickly before expanding what you delegate.

Source: Anthropic: Claude Code best practices

How do I get better at it?

Three habits matter most. Write briefs the agent can actually use: what you want, what context it can use, what done looks like. Keep tasks small and feedback loops fast (Anthropic's own team puts 2–3× quality on tight feedback loops alone). Read every diff. The agent's worst failures don't look like errors. They look like plausible code that does the wrong thing.

Source: Anthropic: Claude Code best practices

What kinds of work should I not hand to an agent?

Anything where being subtly wrong is expensive. Security-critical changes. Migrations. Cross-cutting refactors in code you don't fully understand. Work where "looks right" isn't sufficient. Scoped tasks inside well-tested boundaries are where agents pay off the most. Anthropic's 2026 trends report saw the same pattern across teams: engineers delegate work that's easily verifiable or low-stakes and keep the conceptually difficult work for themselves.

Source: Birgitta Böckeler: To vibe or not to vibe

Should I let an agent merge its own PRs?

Almost never on production. The point of agentic engineering is that someone (you) understood what was supposed to happen. If you didn't read the PR, no one did. Auto-merge is fine on scratch repos, generated config, throwaway prototypes. The line is whether anyone else's data, time, or trust depends on the code. GitHub's own guidance for the Coding Agent era keeps a human review gate on anything touching production.

Source: GitHub Engineering

Which tool should I pick?

It's converging within categories. Among terminal agents (Claude Code, Codex, OpenCode), among editor agents (Cursor, Windsurf), and among async cloud agents (Devin, Jules), the tools are doing very similar things. Pick whatever fits your workflow. Try a second one a week later. The bigger lever is on the human side: how well you brief, how carefully you review, how small you keep the tasks.

Source: Armin Ronacher: A Year of Vibes

From Smarter Dev

Original writing coming.

Smarter Dev walkthroughs, tips, and short courses on agentic engineering will land here as they're written.

Join the Discord to be notified

Last updated