Welcome back! Everything in this edition is about the part around the model. The features you haven't switched on, the harness nobody builds and the files Claude reads before it writes a word.
In today's Generative AI Academy Newsletter:
Claude does a lot more than answer questions: Which three features can you turn on before lunch?
Stop editing prompts, start building harnesses: Which of the six steps do most teams skip entirely?
Think in 7 layers before your AI agent acts: Why does a fuzzy business definition look exactly like a model error?
8 files your Claude setup is missing: Which file does the most work and gets skipped the most?
Claude does a lot more than answer questions
25 features in one cheat sheet, split by where you actually use them.
🟢 Claude Chat (claude.ai or desktop)
Extended Thinking, Memory, Web Search, Deep Research, File Upload, Artifacts, Charts, Projects, Skills, Connectors and Chrome.
Eleven features, and most people are using three of them.
🟠 Claude Cowork (Mac and Windows)
Real file system access, sub-agents running in parallel, plugins, scheduled tasks, Dispatch for sending work from your phone, Computer Use, plus Excel and PowerPoint add-ins.
🟡 Claude Code (terminal or VS Code)
CLAUDE.md, slash commands, Auto Mode, Code Skills, MCP servers and parallel agents.
Every row tells you what it does and exactly how to switch it on.
Settings paths, slash commands, keyboard shortcuts.
A few worth trying today:
→ Dispatch: Scan a QR code and send tasks from your phone while your desktop runs them.
→ Scheduled Tasks: Write the prompt once, pick a schedule, get the report without asking.
→ Auto Mode: Shift+Tab to cycle into it, and a safety classifier reviews each action.
Save it.
You don't need to pay to learn Claude
Seven Claude courses at GenAI Academy. All recorded, and all free.
Start with "What Can AI Actually Do For You?" if you're new, 30 minutes, no setup required.
Then the Claude Starter Course gets you running in 35.
Past that, pick by the problem you have. Hitting usage limits in Claude.ai, Cowork or Claude Code has its own course.
Building no-code agents has one.
So does rolling Claude out across an ops team in 30 days, from pilot to adoption data to full deployment.
The AI Portfolio Builder turns whatever you built into a case study you can show someone.
World Wide Vibes Hackathon is live too, $5,000 prize pool, beginner-ready, 100% online.
One login. Watch them in any order.
Stop editing prompts, start building harnesses
This is exactly what you are looking for.
A new 9-page paper on harness engineering contains the clearest 6-step playbook on the subject you'll see. Link in the comments.
The formula is simple: Agent = Model + Harness.
Prompt engineering shaped what you asked.
Context engineering shaped what the model saw.
This shapes everything around the model that decides whether the work holds up.
Here are the Six steps:
1- Add guides. AGENTS.md, rule files, constraint docs. Every line is a past failure turned into a permanent fix.
2- Add sensors. Linters, tests and validation scripts the agent runs on its own output before a human sees it.
3- Build the loop. Plan, execute, verify, fix. Bounded retries, budget caps, escalation when stuck.
4- Externalize memory. The model forgets every session. The harness remembers all of them.
5- Enforce permissions. Which tools, how many writes, what needs approval. The model does not enforce safety. The harness does.
6- Wire observability. Every tool call, cost and retry tracked, with trip wires when behavior drifts.
Takeaway: step 5 is the one most teams skip and the one that matters most. A limit written into a prompt is a request the model can reason past. A limit enforced by the harness is a refusal it cannot argue with.
Think in 7 layers before your AI agent acts
You now have to decide what it knows, what it can reach, how it recovers and when it has to stop.
That’s why it is useful to think about AI systems engineering across seven layers:
1. Foundation Model → Capability
Pick for the actual task. Know the strengths, the limits and the ways it fails.
2. Prompt Engineering → Instructions
Task, constraints, examples, expected output. Define success concretely enough to evaluate.
3. Context Engineering → Information
Assemble the evidence, memory and tool results the model needs for its next decision. Relevance and freshness both matter.
4. Harness Engineering → Runtime
The environment around the model. Tools, permissions, execution boundaries, state and logs.
5. Loop Engineering → Feedback
How it acts, checks the result and retries. Stopping conditions and an escalation path are part of this, not extras.
6. Graph Engineering → Coordination
Dependencies, branches and handoffs. What runs in parallel and what waits.
7. Ontology Engineering → Shared Meaning
The entities, relationships and business rules. What exactly counts as a customer, an approval or a completed order?
The layers overlap, and the hard failures happen where they meet.
A fuzzy business definition looks like a model error.
A tool timeout looks like a reasoning failure.
A missing stop condition turns a sensible retry into an expensive loop.
So before you rewrite the prompt or swap the model, find out which layer the failure actually lives in.
That diagnostic habit is becoming the job.
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.
8 files your Claude setup is missing
Set them once. Claude reads them before every task, forever.
1. about-me.md (who you are)
Role, company, what you're working on this quarter. Claude reads it before it writes a word.
2. voice-profile.md (your taste)
How you think and write, what you believe, what you love and what makes you cringe. This is the one that stops output sounding like a press release.
3. anti-ai-writing-style.md (your boundaries)
Everything Claude should never sound like. Ban "delve," "synergy" and "utilize" before they ever appear. This is the file people skip, and it does more work than the other seven.
4. The Cowork folder (four folders, that's it)
ABOUT ME for identity and rules. PROJECTS for briefs and drafts. TEMPLATES for finished work you want copied. CLAUDE OUTPUTS for where things land.
5. Global Instructions (stop retyping your rules)
Settings → Cowork → Edit Global Instructions. Tell it to always read ABOUT ME first. Never type it again.
6. The One Prompt (flips who asks the questions)
"I want to [TASK] so that [SUCCESS CRITERIA]." Read folder → ask questions → refine → execute. Claude interviews you instead of guessing.
7. Connectors (Claude inside your tools)
Slack, Google Drive, Notion, Gmail. Settings → Connectors → Browse → Add. No more pasting screenshots.
8. Plugins (skill packs you don't have to build)
Marketing, data, sales. Cowork → Customize → Browse Plugins → Install.
Swipe through all 8.
Everything else you shouldn't miss
Salesforce wants one dashboard for every agent: An Enterprise AI Harness built on six layers plus an AI Control Plane for managing agent identity, lifecycle, performance and cost. It covers third-party agents too, not just Salesforce's own.
Z.AI is raising $5 billion: The company formerly known as Zhipu is raising $2 billion in shares and $3 billion in convertible bonds for compute and its next generation of models.
China's spy agency has entered the AI debate: The Ministry of State Security issued its first public warning about generative AI, naming deepfakes, automated influence campaigns and sensitive data leaking through foreign AI tools. Until now this argument belonged to the tech regulators.
Reuters gave AI the edit suite: Reuters connected its MCP server to CuttingRoom's ShortCut platform, so journalists can search footage in plain language and have AI handle cuts, audio mixing, color correction and captions inside newsroom editorial rules.
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.
