Welcome back! Today is all playbooks. Six pieces on what to set up, what to skip and where your agents keep breaking.
In today's Generative AI Academy Newsletter:
When did you last change your Claude setup? Folders, MD files and global instructions are advice for a version that no longer exists.
Why start with the tools? The 8-step order that actually works, and why the tools sit at the end.
Why do 40% of agent projects get canceled? Gartner's number, and the 8 steps that separate a system from a prompt on an API.
What would Marie Kondo do with your inbox? A Grok Bot that audits your email, Drive and subscriptions and touches nothing without approval.
What stops your agent? Anthropic defined four kinds of agent loop, and skipping the stop condition is how one runs until morning.
Everyone teaching Claude "tips" is a year behind
Folders, MD files, global instructions. Most advice circulating assumes a version that no longer exists. Here's what works now:
Setup. Start hard chats with Fable 5 on high effort, switch to Opus 5 after two turns. Training off, memory off, global instructions deleted.
Connectors. Gmail, Drive, Notion, Calendar, GitHub. Then turn off every one the task doesn't need.
Prompting. Give goals, not tasks. Three rules max. Delete your examples. Never end with "right?" Drop "think step by step" and give a role instead. Edit your message rather than sending a follow-up.
Skills. Skill anything you've explained twice. Test each with five phrasings. Zero folders.
Projects. One per recurring deliverable. Upload the emails, agreements and past campaigns. Leave the instructions blank.
Cowork. Give it big tasks. Sessions run 8 to 30 minutes, so walk away. Pick the model first, no switching mid-session.
Design and code. Upload reference images first. Demand fidelity to your references over its taste. Hand off with a HANDOFF.md.
Writing. No em dashes. Ban delve, leverage, seamless, pivotal, tapestry. One sentence under 6 words, one over 25.
Token economy. Message 30 costs 31x message 1. New chat every 30 to 50 turns. Skip 5 to 11am Pacific.
GenAI Academy covers all of it: setting up Claude properly, building custom skills and orchestrating past copy-pasting the same prompt over and over.
A quick cheat sheet to learn AI
Most people learning AI start with the tools. That's the last thing on the list.
Tools change every quarter, and the foundation underneath them holds for years.
Here's the order that works:
☑ 1. Intro to AI: what it is, what it changes, where it's applied
☑ 2. Mathematics: statistics, probability, linear algebra, calculus
☑ 3. Programming: Python first, R and Java after
☑ 4. Big Data: Hadoop, Spark, Cassandra, MongoDB
☑ 5. Data Science: acquisition, preparation, analysis, manipulation
☑ 6. Machine Learning: supervised, unsupervised, reinforcement
☑ 7. Deep Learning: TensorFlow, Keras, CNNs, RNNs, GANs, LSTMs
☑ 8. AI Engineer: the job title on the other side
The tools, ChatGPT, Claude, Gemini, Perplexity, Cursor and NotebookLM, are the easy part. You can learn any of them in a weekend.
What takes longer is knowing why the output is wrong. That only comes from the fundamentals.
From Claude fundamentals to full content workflows, you learn to work with AI without sounding like it wrote everything for you.
How to build an AI agent
You're building AI agents without a system. That's why they keep failing.
Gartner expects over 40% of agentic AI projects to be canceled by the end of 2027. The model is rarely what kills them. Teams hook a prompt onto an API, call it an agent and hope.
The difference between what works and what doesn't? A system.
8 steps, in order:
1. Define the job. One problem, one user, one measurable win.
2. Design the brain. System prompt, role, guardrails. This is where most builds fail.
3. Pick the model. Reasoning effort vs speed vs cost. Stop overpaying for easy steps.
4. Add tools. APIs, MCP servers, even other agents.
5. Give it a memory. Short-term context plus long-term recall, so it improves with every run.
6. Orchestrate the flow. Triggers, retries, queues and agent-to-agent handoffs.
7. Build the interface. Chat, API or Slack. Meet users where work already happens.
8. Give enough time to test and evaluate. The step everyone speedruns is the most important one.
How to do a digital declutter with Grok bot
→ Go to Grok Bot, download the desktop app and sign in
→ Open Plugins and connect your Gmail and Google Drive
→ Click the "+" icon in the top-left corner, then Create new Bot
→ Describe the job in plain language. Grok builds the Bot from your instructions
→ Name it so you can find it in the sidebar
Sample prompt to copy:
“You are Marie Kondo for digital clutter. Audit my email, Google Drive and paid subscriptions, then create a cleanup plan with no more than 10 items per category. Find emails to unsubscribe from, redundant files to remove or rename, and paid subscriptions I may want to cancel. Do not move, delete, unsubscribe or cancel anything without my approval.”
Enhance Your CV with ChatGPT
Your CV gets ten seconds.
That's how long a hiring manager spends matching you against a job description. They scan for their language, not yours. Miss it and you're skipped, however qualified you are.
So the rejection usually wasn't about your experience.
The fix is rewriting the CV per application, which nobody does across ten roles. ChatGPT closes that gap. Paste in your CV and the job description, and twenty minutes gets you a tailored version, a LinkedIn summary and a clear read on where you're strong.
Run it on a role you already got rejected from, using the CV you actually sent. The difference usually explains the outcome.
If three job descriptions name the same missing skill, that's a pattern worth acting on.
Changing industries? Use it to translate your experience into the new field's vocabulary.
Feed the tailored CV and job description back in for a cover letter draft. It won't be finished, and it beats a blank page.
One caution. Reframing in their language is the point. Inventing experience is a different thing, and it surfaces in the interview.
Anthropic just defined what an agent loop is
Anthropic defines the agent loop: an agent repeating cycles of work until a stop condition is met.
There are four, and the only thing separating them is how much you stop doing yourself.
1. Turn-based. You hand off the check.
Triggered by your prompt, stops when Claude thinks it's done. Write the check once as a skill, save it as SKILL.md, and it runs on every job after that.
"Create a skill that checks my work: open the page, screenshot before and after, check the console, fix what fails and rerun."
2. Goal-based. You hand off the stop condition.
Stops at the goal or the turn cap. State the cap out loud.
"/goal get the homepage Lighthouse score to 90 or above, stop after 5 tries"
3. Time-based. You hand off the trigger.
Runs on an interval you set, on your machine. Close the lid and it stops.
"/loop 5m check my PR, address comments, fix failing CI"
4. Proactive. You hand off the prompt.
Triggered by an event or schedule, running in Anthropic's cloud as a routine. Keep it cheap with a smaller model.
"/schedule every hour: check the feedback channel for bug reports"
All four need a stop condition. Skipping it is how a critique cycle runs all night and burns tokens until morning. The fix is one cap. Stop after 5 tries.
BTW, GenAI Academy covers this landscape in depth, from Claude fundamentals through full agent orchestration and enterprise-scale deployment.
Everything else you shouldn't miss
Reddit almost vanished from ChatGPT: Its share of ChatGPT citations fell 86% since mid-July, from nearly 4% to roughly 0.5%. OpenAI hasn't explained why.
Google announced a forecasting model in words nobody understands: TimesFM-3 predicts future numbers from past numbers: sales, weather, traffic. Google calls it multivariate time series forecasting in a single forward pass. Does that announcement read as AI-generated to you?
Cerebras says its new chip is 30x faster than GPUs: The CS-4 serves answers up to 30 times faster than GPU systems, and it ships this quarter.
The Pentagon now has its own ChatGPT: ChatGPT Mil launched Monday, cleared for sensitive government information the public version could never handle.
Learn more about AI from the experts building it
Follow us on Instagram for fast, visual AI updates in 30 seconds.
Subscribe to our Atlas newsletter — trusted by 3M+ subscribers — to stay ahead of AI news across tech, education, and business.
📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.
🐦 Follow us on X for breaking AI news and real-time industry updates.
Learn how to build your next AI application with practical resources and expert guidance.
Explore investment opportunities in the future of AI and join our community-backed growth journey.
