🤖 Hermes Agent launches Computer Use

Good Morning! Here's what's happening:

Nous Research expanded Hermes Agent with full computer use support on Windows and Linux via trycua, adding to its existing macOS support and making it the first open-source agent with computer use across all three major operating systems.

Computer use on macOS was already significant. Bringing it to Windows and Linux means the full Hermes toolkit is now available regardless of what operating system you use. Most tools with computer use capabilities are expensive and closed. Hermes is free and open source, and it just became the most capable open-source agent you can run locally.

xAI introduced /goal in Grok Build, a new command that lets you give Grok a single goal and have multiple rounds of subagents work through it, implement it, and verify it automatically from start to finish.

Most AI tools hand you a result and stop. /goal keeps going. You give it a target, and it figures out the steps, runs them with multiple subagents, and verifies each one before moving forward. For anyone using Grok to work through complex tasks, this removes the need to babysit the conversation.

ElevenLabs introduced Ads Engine inside ElevenCreative, a tool that connects directly to your Google, Meta, and LinkedIn ad accounts, localizes your existing ads across 50+ languages, and pushes the finished creatives back to your ad platform automatically.

Localizing ad campaigns used to mean hiring voice talent in multiple countries, working with translation agencies, and managing uploads across platforms one by one. Ads Engine collapses that entire process into one workflow. For any creator or business running ads across multiple markets, this removes something that used to take days.

The system that turns consumption into actual knowledge.

Last semester a student came to my office hours confident.

He'd been following AI for months. Newsletters, YouTube, podcasts. He knew every model release, every funding round, every viral demo. He could name the companies, the founders, the benchmarks.

I asked him one question: explain how attention works in a transformer.

He froze.

Not because he was unprepared. Because he'd been consuming AI content for six months without actually learning any of it. The information had passed through him and left nothing behind.

I see this constantly. And if I'm honest, I catch myself doing it too. Opening a newsletter, reading it end to end, feeling informed, and three days later having nothing to show for it except a vague memory that something interesting happened.

AI has made this worse, not better. It's never been easier to get a summary of anything. Which means it's never been easier to feel like you understand something without doing the cognitive work that actually creates understanding.

This guide is the system that fixes it.

The real problem with AI and learning.

Here's the counterintuitive part.

Active recall produces 50-150% better long-term retention than passive methods. Re-reading and highlighting are rated low utility by cognitive science research. They don't create the processing that leads to retention.

Reading a summary Claude wrote doesn't count as active recall. Watching a YouTube explanation doesn't either. Skimming a newsletter and nodding along definitely doesn't.

Real learning happens when your brain has to retrieve information from memory. Not recognize it on a page, not skim past it in a feed, but actually pull it out of nothing when someone asks you a direct question.

The rule behind the whole system: if you can't close every tab and explain something out loud from memory, you haven't learned it. You've encountered it.

Step 1: Map before you read.

Before touching a single article or paper, spend three minutes building the skeleton.

New information needs structure to attach to. Without a map, everything feels equally important, nothing sticks, and you finish reading with a vague sense of understanding that evaporates within 48 hours.

Become a paying subscriber to get access to all premium content.

✓ Full archive of premium guides with ready-to-use prompts

✓ Structured AI courses (step-by-step, start-to-finish)

✓ Every upcoming premium tutorial

OpenAI launched Daybreak, a set of security tools that includes a Codex Security plugin that finds, validates, and fixes vulnerabilities right inside Codex, a full release of GPT-5.5-Cyber for security defenders, a Cyber Partner Program for security companies building on top of OpenAI's capabilities, and a Patch the Planet initiative working with open source maintainers to secure critical projects.

Google made the Interactions API generally available as its primary interface for Gemini models and agents. Built for stateful agentic workflows, it brings Managed Agents, background execution, expanded tool support, and multimodal generation into one unified interface.

NotebookLM launched fully customizable flashcards. Edit questions, tweak answers, and add brand new cards to any set. Share the finished deck with classmates, colleagues, or anyone who needs to study the same material.

🤖 Grok: xAI's platform now includes /goal in Grok Build. Give it a single goal and multiple rounds of subagents implement and verify it automatically.

🎙️ ElevenLabs: Ads Engine in ElevenCreative connects to your ad accounts, localizes campaigns across 50+ languages, and pushes finished creatives back automatically.

💻 Hermes Agent: Free, open-source AI agent now with computer use on macOS, Windows, and Linux. Background subagents, iMessage support, and a full desktop app.

🎬 Google Vids: Design custom AI presenters that speak 24 languages.

Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!

See you tomorrow :)

- Dr. Alvaro Cintas