Professional agentic AI, on the Claude platform.
Training for individual contributors and engineering teams.
Most people use AI assistance as a fancier autocomplete. Working with agents in production takes method: context, judgment and verification. That's what I teach.
✦ Claude Code
From zero to professional, one tip a day.
263
tips published.
Tomorrow, 264.
170 days in a row publishing.
Today: #263
By intent
Onboarding, mastery, autonomous...
By topic
Skills, MCP, hooks, models...
Latest
All content by date
My Claude Code workflow.
The skills, agents and instructions I use as a professional Software Engineer. It's called Craft, and it's a plugin you can install.
See the skillsTraining and consulting on the Claude platform.
AI First on the Claude platform
A working day with your team, online, on the Claude platform. More output without losing quality.
Agentic AI for companies
I assess how your team works and agentify the processes one or several agents can take on.
At FrontendLeap I teach you to build the professional Web with agentic AI and learning science.
FrontendLeapStop Claude Code asking for permission (without losing control)
Permission prompts in Claude Code don't go away with one setting. There are five levers (rules, modes, auto mode, the sandbox and bypass inside a container), and this guide puts them in the order that cuts the most prompts for the least risk.
Claude Code first steps: the fundamentals that don't expire
The Claude Code fundamentals that don't expire: install, sessions, context, quota, models, permissions. What you actually need to grasp on day 1.
AI: the joy of missing out on what doesn't matter
You can't keep up with AI, and you shouldn't try. The answer isn't resistance or passive acceptance: go all in, but deep on one thing instead of scattered.
Claude Code Agent View: managing parallel agents without leaving the terminal
I spent months managing many Claude Code agents by hand. Agent View now does it: my terminal-native workflow with Warp and Neovim, and the gaps I patch.
What is Claude Code: A Practitioner's Overview
What Claude Code actually is, explained by someone who uses it 12+ hours daily in production. No hype, real examples.
Claude Code stops before finishing the task: raise Fable 5.2's effort
Claude Code starts Fable 5.2 at medium, where more turns end with the work half done. Anthropic's system prompt fix is already built in: your part is raising effort for the long stretch and replying with the exact action.
Fable 5.2 on Claude Pro: included this week, then usage credits
Fable 5.2 on the Claude Code Pro plan sits outside your limit: it runs on usage credits. This week is the exception, and it pays to know where it ends and which sessions bill without asking.
Fable 5.2 in Claude Code: up to 30% faster for 20% less
Fable 5.2 lands in Claude Code faster, at a lower price per token and with medium as its default effort. Below v2.1.321 you are still on Fable 5.1, even though the picker says Fable.
Claude Code scheduled task memory between runs: give it a notes page
Claude Code scheduled tasks start every run from zero and repeat yesterday's mistake. A page they read at the start and append to at the end gives them the memory they lack.
Claude Code scheduled task succeeded but did nothing: open the run and read it
A scheduled task marked Succeeded only tells you the session ended without an error. A missing connector, a blocked request or a command auto mode stopped is inside the run, not on its row.
Hi, I'm Juan Andrés Núñez.
Senior Software Engineer specializing in agentic AI, professional educator and speaker. Over fifteen years building for the Web and teaching how to do it. Today I work with the Claude platform every day. Here I publish everything I learn.
The 51 essentials, as a guide.
One page per tip. Five chapters. What I actually use daily in production. No theory, no fluff.