Week of May 17, 2026
This week's picks reveal a massive shift toward AI-as-a-service business models, from solo entrepreneurs charging $5K/month for AI agents to Notion's million custom agents spawning an entire developer platform. Meanwhile, OpenAI's new voice models are making real-time AI interactions mainstream, fundamentally changing how we'll interact with technology.
This Week's Top Videos
Reimagining the mouse pointer with AI
By Google DeepMind
This is how top AI companies spot bad sales hires
By 20VC
The $1M+ Solo AI Agent Business (Full Course)
By Greg Isenberg
Solo entrepreneurs are charging $5K/month for 'unlimited' AI agent services by selling outcomes, not infrastructure—positioning agents as digital employees rather than tech tools. The key insight: customers think they need 10+ agents but really only need 1-3, making 'unlimited' offers profitable while removing friction. This matters now because 99% of businesses are behind on AI implementation.
- Charge $5K/month for unlimited AI agents by betting customers will only use 1-3 agents despite thinking they need 10+
- Position AI agents as 'digital employees' not tech tools—customers should never hear words like 'tokens' or 'infrastructure'
- Go vertical and industry-specific to avoid being commoditized—don't just sell ChatGPT, sell industry-tailored solutions
- Deploy first agent within 48 hours and talk business outcomes (revenue generated) not time saved to close deals faster
- 99% of businesses are behind on AI—your basic Claude/Hermes setup skills are more valuable than you think
- Create abundance in your offer with unlimited usage, monitoring, and support to remove all customer friction
Build Hour: GPT-Realtime-2
By OpenAI
OpenAI dropped three new audio models including GPT Realtime-2 with 4x larger context (128k tokens ≈ 1 hour conversation) and parallel tool calling. The new voice agents can perform complex e-commerce searches, analyze product reviews, and control UIs in real-time with 200ms latency. This unlocks production-ready voice interfaces that could replace traditional app navigation.
- GPT Realtime-2 increased context window 4x to 128k tokens, roughly equivalent to an hour of conversation, enabling full session memory without truncation
- Parallel tool calling eliminates waterfall execution, allowing voice agents to perform multiple actions simultaneously instead of sequential operations
- Real-time Whisper model achieves 200ms latency with streaming capability, enabling instant voice-to-action workflows for production apps
- Voice agents can now control UI elements directly through tool use, demonstrated with e-commerce search that filters products and analyzes reviews in real-time
- Dynamic voice cloning supports multiple speakers with tone matching, allowing AI to distinguish between different people in conversations
- Controllable expressiveness lets developers program specific emotional responses like whispering, excitement, or jealousy for more natural interactions
Introducing Notion’s Developer Platform: Keynote (Ivan Zhao)
By Notion
Notion users have created over 1 million custom agents in just 3 months, and now Notion is launching a full Developer Platform with workers—TypeScript programs that sync any data, build custom tools, and orchestrate agents directly in Notion workspaces. This transforms Notion from a productivity tool into an AI-native development platform where teams of humans and agents collaborate seamlessly.
- Over 1 million custom agents were built on Notion in just 3 months after launch, showing explosive adoption of no-code agent building by non-technical users.
- Notion introduced 'workers'—TypeScript programs in managed sandboxes that handle syncing, automation, and agent orchestration with built-in deployment, security, and governance.
- The new Developer Platform enables syncing any data source (Postgres, Snowflake, internal APIs) into Notion databases to create a unified workspace for teams and agents.
- Notion made their Markdown API 91% more token efficient and added native agent language support, positioning themselves as an AI-first platform rather than just a productivity tool.
- The platform assumes workers run agent-written code that developers might not have time to read, with built-in tools for coding agents to test and validate their own work.
- Companies like Vercel and Rakuten are using custom agents for automated roadmap updates and overnight customer request processing, showing enterprise-level adoption.