agents → tools → loops → graphs → autonomous systems
Google just put all five in one free course. most people stop after the first one.
2 hours 17 minutes, taught by a Developer Relations Engineer at Google Cloud. no signup, no paywall.
39:45 - build your first agent
56:10 - connect MCP tools
1:14:34 - four loop patterns for running it
1:23:01 - turn those loops into a graph
2:05:00 - wire the whole thing into one autonomous system
the progression is the point. one agent is a demo, an agent with tools does work.
loops let it keep going. graphs let several of them coordinate.
only the last step runs without you in the room.
most people are still tuning the prompt inside step one.
worth more than most paid agent bootcamps.
bookmark and watch it today, then read the full graph engineering breakdown below
You're still on the hook for code your agent wrote. CodeRabbit stands between the agent and prod, catching the logic errors, security holes, and regressions before they merge.
Andrew Ng just dropped a 3-hour course on how to become an AI Engineer in 2026:
• 00:00 - How to build agentic AI systems
• 04:25 - Future of AI engineering
• 23:38 - AI Prompting full course
• 2:52:17 - Creating an app with AI in 30 minutes
This 3-hour watch could replace 10 AI engineering courses on the internet.
Watch it today, then read the 12- month path to becoming an AI Engineer in the article below.
99% of people using Claude are prompting it WRONG.
Anthropic just published how to actually prompt the Claude 5 family. Fable, Opus, and Sonnet each need a completely different approach.
Here's how to use Claude the right way:
Universal rules that apply to all 3:
→ Don't port over old prompts. Skills and system prompts built for older Claude models can actually make Claude 5 worse. Start fresh.
→ Match effort to the task. Low/medium for quick stuff, high as your default, xHigh+ for anything genuinely hard, Max only when quality is non-negotiable.
→ Say exactly what you mean. These models are literal now. They won't silently assume you meant something broader.
→ Effort level doesn't shorten responses. If you want brevity, ask for it directly.
Fable 5 (long, autonomous, expensive):
→ Tell it why, not just what. Give it the bigger picture: "I'm working on [task] for [who]. They need [outcome]. With that in mind: [request]."
→ Keep instructions short. Over-engineering the prompt makes a model built to find its own approach perform worse.
→ Set explicit checkpoints, or it'll set its own. "Pause only for destructive actions, real scope changes, or something only I can provide."
→ Give it a memory file. One lesson per note, updated not duplicated, wrong notes deleted.
Opus 5 (the daily all-rounder):
→ Stop telling it to "double-check" or "verify." It already does this. You're just burning tokens.
→ Ask for brevity explicitly, its defaults run long now.
→ Constrain scope clearly. It will expand a task on its own if you're vague.
→ For review work: ask for full coverage first, then filter. Telling it to "only flag serious issues" upfront makes it miss real problems.
Sonnet 5 (fast, cheap, the workhorse):
→ Raise the effort for anything complex. At low/medium it scopes work exactly to what was asked, sometimes too literally.
→ It calibrates its own response length based on task complexity, unlike Opus.
→ For frontend work: give it a concrete spec or design skill. Vague instructions just shift it to a different fixed style, not a better one.
→ It responds well to explicit tone instructions ("warm, collaborative, acknowledge my framing before answering").
I haven't seen anyone else talk about these principles - bookmark!