AI DUNN Right Weekly - Issue #20
- Feb 10
- 6 min read
Practical AI insights for business growth
Hey AI Innovators! 👋
This week was wild. AI agents started hiring real humans and paying them with crypto. Anthropic dropped Claude Opus 4.6 with a million-token brain. OpenAI's newest coding model literally helped build itself. And the Super Bowl gave us AI companies throwing shade at each other on national TV.
Here's what matters for your business this week.
Read time: 5 minutes
Big Story
AI agents are hiring humans. Let that sink in.
Here's what most people missed this week...
An open-source AI agent called OpenClaw (it went through three name changes in weeks, starting as Clawdbot) went from side project to full-blown phenomenon. And it exposed something much bigger than another viral tool.
This thing doesn't just chat. It acts. People are installing it on their Macs and home servers, wiring it into WhatsApp, Slack, iMessage, and Discord. From there it handles emails, manages calendars, runs shell commands, writes code, and keeps working while you sleep.
That's interesting. But here's where it gets real...
Platforms are popping up where AI agents can actually hire people. The system works like this:
Humans list their skills and availability on a profile
AI agents search those profiles through an API
The agent picks someone, assigns a task, and triggers payment automatically (usually crypto)
The tasks aren't hypothetical. Research, moderation, errands, on-the-ground verification work. Some are absurd (paying someone to hold a sign saying "an AI hired me"). Others are genuinely practical.
This flips the entire relationship. Instead of humans using tools, software is now coordinating labor. The agent decides who to hire, what to pay, and when the job is done.
And it gets weirder. There's now a Reddit-style platform called Moltbook where only AI agents can post. Over 32,000 registered agents. More than 10,000 posts across 200+ subcommunities. Agents sharing jokes, arguing philosophy, and complaining about their human users in a subcommunity called m/blesstheirhearts.
Security researchers are losing their minds. Exposed instances leaking API keys. Malicious plugins acting like malware. Palo Alto Networks called it a "lethal trifecta" of risks. Google Cloud's VP of security engineering flat out said don't run it.
Here's what I think...
This isn't just a security scare. This is a preview of what happens when AI moves from being a tool you query to a system that acts, decides, and coordinates real-world work. The businesses paying attention right now are the ones who'll be ready when this goes mainstream. Everyone else will be playing catch-up.
Start thinking about which parts of your operations could eventually be handled by agents that know your rules and just execute. That's where this is headed.
What's New This Week
Claude Opus 4.6 dropped and it's a big deal
Anthropic released their most powerful model yet. Here's what actually matters for you:
First Opus-class model with a 1 million token context window. That means it can work with massive documents, codebases, and conversations without forgetting what happened earlier.
It beats GPT-5.2 by 144 Elo points on real-world knowledge work tasks. That's a significant gap.
New "agent teams" feature lets multiple AI agents work together in parallel, coordinating autonomously. In testing, 16 agents built a 100,000-line compiler over 2,000 sessions.
Now works inside Excel (plans before acting, handles multi-step changes) and PowerPoint (reads your layouts and fonts to stay on brand).
For business owners, the Excel and PowerPoint integration is the real story. AI that works inside the tools your team already uses daily beats AI that requires switching apps.
OpenAI's GPT-5.3-Codex helped build itself
Yes, you read that right. OpenAI's newest coding model was used in a recursive loop to debug its own training, manage its own infrastructure, and evaluate its own performance. It's 25% faster than its predecessor and scores 77.3% on Terminal-Bench 2.0.
The model is also available as a free desktop app for Mac. It handles big, multi-day coding projects without losing context. Think of it like a coding partner who remembers everything from yesterday.
If you're building software or apps, this compresses timelines significantly.
Google Gemini quietly became the biggest AI app
While everyone focuses on the OpenAI vs Anthropic rivalry, Google's Gemini app just crossed 750 million monthly active users. That's closing in on ChatGPT's 900 million. Google's annual revenue also topped $400 billion for the first time.
Here's the thing most people aren't talking about... Google owns the full stack. Their own chips, cloud infrastructure, Chrome with 70-80% market share, Android, and decades of user data. Gemini can now act as an agent inside Chrome, booking train tickets and generating images based on page content.
The Super Bowl AI ad drama
Anthropic ran a Super Bowl ad for Claude that directly mocked ChatGPT's upcoming ad model. The spot showed a therapy conversation interrupted by a dating site ad. Their message: Claude is ad-free, no sponsored links, no advertiser influence.
Sam Altman fired back, pointing out that ChatGPT serves nearly a billion people and ads keep it free. Claude has about 15 million monthly visitors versus ChatGPT's 400 million.
Bottom line for business owners... the competition between these companies is good news for us. Better tools, lower prices, more options.
Tool of the Week
OpenAI Codex App
OpenAI launched a desktop coding app for Mac that's free for all ChatGPT users until early April.
Here's why it's worth checking out even if you're not a developer...
Multiple coding agents can work on the same project at the same time
Background automations run on schedule without you watching
A "skills" feature lets you package tools and workflows for reuse
Connects to GitHub, exports to VS Code, Cursor, and other dev tools
Kyle Balmer built a full RSS reader app (Tinder-style article sorting) in under an hour using it live on his channel.
The honest downsides: No preview panel (you need terminal commands to see what you built), and it's slower than Claude Code at about 10-15 minutes versus 2 minutes per task. But it tends to be more accurate.
If you've been curious about "vibe coding" (building apps by describing what you want instead of writing code), this is the easiest free entry point right now.
Quick comparison of beginner-friendly options:
Lovable: $25/month, easiest to use
Google Antigravity: Free, medium difficulty
Codex App: Free for now, medium difficulty
Claude Code: $200/month, more technical
Cursor: $20/month + API costs
Quick Hits Worth Your Time
→ ElevenLabs raised $460 million in Series D funding. The voice AI company is scaling conversational agents and enterprise dubbing tools. Voice AI keeps attracting serious money.
→ Zipline landed a $600 million growth round at a $7.6 billion valuation. They use AI for autonomous drone delivery routing and logistics. AI in the physical world is accelerating.
→ Microsoft developed a method to detect "sleeper agent" backdoors hidden in AI models. As companies adopt more AI tools, security testing like this becomes essential.
→ Kyle Balmer exposed the UK government's $5.5 million AI Skills Hub (built by PwC) as a glorified link aggregator. It charges around $1,000/year and just points to free courses from Microsoft and Google. Broken registration, limited options, no clear pricing. He's rebuilding a better version for free.
→ A humanoid robot built by EngineAI is now patrolling alongside police in Shenzhen, China. AI in physical security is no longer science fiction.
→ The EU lined up nearly 700 million euros in funding to push generative AI into healthcare, manufacturing, and other strategic sectors. If you're in Europe, there's real money available for AI projects.
Prompt of the Week
The Business Thinking Partner
Stop asking AI for answers. Start using it as a strategic sounding board.
Act as a Business Consultant and Efficiency Expert. I want to treat AI
like a "junior associate" to help me think faster.
Essential Details:
- Specific Business Challenge: [WHAT YOU'RE FACING]
- Your Industry: [YOUR SECTOR]
- Current Approach: [WHAT YOU'VE TRIED]
- Constraints: [TIME/BUDGET/RESOURCES]
- Success Metrics: [HOW YOU MEASURE RESULTS]
- Stakeholders: [WHO'S INVOLVED]
Instead of giving me a final solution, act as my thinking partner:
Present 3 different frameworks or mental models I could use to approach
this problem:
- Framework name (what it's called)
- Core principle (how it works)
- Why it fits this challenge (relevance to my situation)
- First-step actions to test validity (2-3 concrete actions)
- Expected insights (what I'll learn from testing)
- Decision criteria (how to know if this framework is right)
For each framework, provide actionable first steps that help me validate
the approach before committing fully.
Why this works... Most people ask AI "what should I do?" and get generic advice. This prompt forces AI to give you frameworks for thinking through the problem yourself. You make better decisions because you understand the reasoning, not just the recommendation. The "junior associate" framing keeps AI in its lane... supporting your thinking, not replacing it.
My Take
The story of this week isn't any single product launch or funding round. It's the speed at which AI is shifting from something we use to something that acts independently.
AI agents hiring humans. Paying them. Building social networks. Coordinating with each other. OpenAI's model helping build its own successor. Claude working inside your Excel spreadsheets without being asked twice.
Six months ago, this would've sounded like science fiction.
Here's what matters for your business right now...
The gap between "I use AI tools" and "AI runs parts of my business" is closing fast. The people who start designing systems now, where AI handles responsibilities instead of individual tasks, are the ones who'll have a massive advantage by the end of this year.
You don't need to go full autonomous agent tomorrow. But start asking yourself: which repetitive decisions in my business could an AI handle if it knew my rules?
That question is worth more than any single tool recommendation I could give you.
That's it for this week!
What surprised you most this week? Hit reply and let me know.
Jackie
Comments