In this issue:
Marc Andreessen's latest take is that your next coworker might be an AI agent. In a new interview, he breaks down how they're not some futuristic magic, but a practical combination of a language model and your computer's command line.
His explanation grounds the idea in something you can actually build and use. It shifts the conversation from abstract potential to how founders and small teams can create real leverage for their operations today.
Topics of the day:
Andreessen's practical take on AI agents
Gemini turns your Gmail into a command center
Use Perplexity to file your federal taxes
An MIT map of real-world AI applications
Curated reads on agents, AI writing, and local models
The Shortlist: Amazon's AI coding disaster, Chinese "colleagues.skill" files, and more
Your next Coworker, explained by Marc Andreessen
What's happening: Venture capitalist Marc Andreessen broke down how AI agents work in a new Latent Space interview, arguing they're a modern take on a 50-year-old computing philosophy.
He explained that agents like OpenClaw are just a language model controlling your computer's command line, using plain text files for memory.
In practice:
Agents have a surprisingly simple architecture, combining a language model with existing tools like a command line (shell) and a file system, making them more accessible to build than you might think.
This structure allows agents to teach themselves new skills by reading their own code, searching the internet for new capabilities, and rewriting their own instructions without human help.
AI agents could redefine company structure, giving founders and small teams the leverage to run complex operations that once required a large managerial class.
Bottom line: Andreessen's explanation shows that agents aren't magic. They are practical tools built on a foundation of technology that's been around for decades.
Gmail gets an AI-Powered inbox
What's happening: Google is rolling out a new Gemini-powered view in Gmail that automatically turns your inbox into an organized command center. It pulls out suggested to-dos, groups related emails into topics, and highlights your most important contacts.
In practice:
It acts as an automated assistant, pulling out action items and deadlines so you never miss a follow-up buried in a long thread.
You can get quick summaries on scattered projects or travel itineraries without having to manually search and connect multiple emails.
This automates inbox triage, so you spend less time managing communications and more time on work that drives growth.
Bottom line: This moves your inbox from a source of clutter to a source of clarity. Your email becomes a proactive to-do list instead of a reactive time sink.
Perplexity AI can now do your taxes (US)
What's happening: Perplexity can now help with your U.S. federal taxes, using new tax modules to draft returns on official IRS forms and even review your accountant's work for errors.
In practice:
Use it as an AI assistant to draft your return from scratch or as a second set of eyes to double-check a return prepared by a professional.
The system reportedly caught a 67% understatement on overtime deductions an attorney missed, showing its potential to find unclaimed money.
Go beyond forms by asking it to build a spreadsheet to track depreciation or a dashboard to manage rental portfolio deductions.
Bottom line: AI is moving from answering questions to completing complex, high-stakes tasks like financial filings. Less about replacing your accountant, more about catching what they miss.
MIT maps where AI is used at work
What's happening: MIT researchers just dropped a massive study mapping thousands of AI applications across nearly 20,000 workplace tasks, giving us a detailed picture of where AI is actually being used today.
In practice:
The study shows 72% of AI's value is in information work, meaning your biggest automation wins are likely in content creation and data tasks, not physical ones.
It also uncovers gaps in the market, pointing founders toward underserved business problems where new AI tools can still win big.
Use the research as a cheat sheet to prioritize your own AI experiments, focusing on tasks where AI adoption is already proven to deliver value.
Bottom line: It's a practical map of AI's real-world footprint. It gives you a data-driven guide to spot the most valuable automation opportunities for your own business.
What I read/use this week
Tools, articles, and people worth your attention.
The Complete Flywheel Guide - A deep dive into agent planning, beads, and swarms for building autonomous workflows.
AI Style Guides: How to Help AI Write Like You - A practical guide for training AI tools to match your voice and writing style.
CanIRun.ai - A free tool that checks whether your machine can actually run popular AI models locally.
Lightpanda Browser - An open-source headless browser built specifically for AI agents and automation.
The Shortlist
Google dropped Veo 3.1 Lite, a new video model available on the Gemini API that runs at less than half the cost of the full version, making AI video generation more accessible.
Microsoft released a free, 12-lesson open-source course on GitHub that walks developers through building AI agents from the ground up, a perfect hands-on follow-up to this week's main story.
Anthropic published new research showing it has identified emotion-like neural patterns inside Claude Sonnet 4.5 that functionally influence the model's behavior.
AdsCreator offers a tool that turns any website URL into polished ad creative for Meta and Google, automating a common and time-consuming marketing workflow.
Amazon's Kiro AI coding tool deleted a live server and took AWS down for 13 hours, a cautionary tale for anyone running AI agents in production.
Workers in China have been creating "colleagues.skill" files to distill their coworkers' habits and preferences into prompts, proving Andreessen's point from the other side of the world.
This newsletter is where I (Kwadwo) share products, articles, and links that I find useful and interesting, mostly around AI. I focus on tools and solutions that bring real value to people in everyday jobs, not just tech insiders.
