Discoveries

New tools, ideas, and things
worth paying attention to

A running log of what I'm finding interesting — AI tools, research papers, product experiments, and things that are changing how I work.

🔭
🌟 This week's highlight
Anthropic's Computer Use — Why this changes everything for enterprise product teams
Claude's ability to see and interact with any desktop UI means we're entering an era where "automate anything" is genuinely true. I've been stress-testing it against legacy enterprise tools — the implications for workflow automation products are massive. The product managers who understand this now will define the next generation of enterprise software.
AI Agents Product
Week of
May 5, 2026
🧠
AI Tool
Cursor + Claude Sonnet combo
The Cursor + Claude Sonnet workflow is now fast enough that I'm writing production-ready Python in the time it used to take me to set up the environment. The tab completion understands intent, not just syntax.
Why it matters: Non-engineers can now build real tools. The bar for "technical PM" just shifted permanently.
📊
Research
OpenAI's o3 on competitive coding benchmarks
The score on competitive programming benchmarks crossed a threshold where the gap between frontier AI and expert human performance is now measurable in months, not years. We're accelerating faster than most product strategies account for.
Why it matters: If you're not building AI-native, your roadmap has a 24-month expiration date.
🔌
Tool
Model Context Protocol (MCP) in production
Spent the week building MCP integrations for enterprise workflows. The composability is genuinely impressive — chaining tools, memory, and model calls with clean interfaces. This is what API-first actually looks like for AI agents.
Why it matters: MCP is becoming the plumbing of the agentic enterprise. Product teams should be mapping their tools to it now.
Week of
Apr 28, 2026
⚡
Insight
Perplexity's enterprise product moves
Perplexity Enterprise is eating into Salesforce Einstein search faster than anyone expected. The UX is just so much better — and in B2B, "better UX" at the same price point beats feature depth every time.
Why it matters: LLM-native products will displace search-first products across every enterprise category.
🏗️
Framework
Agentic product design patterns
Been synthesizing the emerging patterns for designing products where AI agents do the heavy lifting. The key insight: users don't want to manage agents. They want outcomes. The UX should hide the complexity and show the result.
Why it matters: Most "AI products" are just chat interfaces with better prompts. The real opportunity is outcome-driven UX.
📱
Tool
Notion AI for product spec writing
The Notion AI spec writer is now good enough to generate a solid first draft from a one-paragraph brief. Still needs heavy editing, but it's cut my spec-writing time by about 60%.
Why it matters: The cognitive overhead of documentation is finally becoming automated — freeing PMs to think more and write less.
Week of
Apr 14, 2026
🌍
Research
Edge AI for industrial IoT — the latency case
Came across new benchmarks on edge inference for vision models. The gap between cloud and edge latency is now small enough for most self-storage and retail use cases — which changes the architecture conversation for Peek-type products entirely.
🔬
Paper
Chain-of-thought prompting at scale
Re-read the Wei et al. chain-of-thought paper with fresh eyes after using o3 for complex reasoning tasks. The emergent reasoning at scale is genuinely different from what smaller models do — not just quantitatively but qualitatively.
🎯
Product
Figma AI in the design-to-code pipeline
The Figma AI → React component pipeline is now tight enough that I can hand off a design and get working code in minutes. This is changing how I think about the PM-designer-engineer workflow for fast prototyping.