Sunday, April 19, 2026
Code Mode and MCP are killing the context bottleneck
April 19 · 3 videos
Sunil Pai says JSON tool calling is dead at scale.
Cloudflare's 2,600 endpoints now use Code Mode.
Token usage dropped by 99.9 percent.
MCP hit 110 million monthly downloads.
David Soria Parra calls it the connective tissue for 2026.
Product managers are facing a judgment day.
Nikhyl Singhal warns that information movers are dinosaurs.
“The information mover is essentially going to become a dinosaur.”
Code Mode: Let the Code do the Talking - Sunil Pai, Cloudflare
Sunil Pai · AI Engineer · 19 min
Watch on YouTube →Sunil Pai introduces Code Mode at Cloudflare. It replaces massive JSON tool definitions with a search and execute pattern for AI agents.
- Representing Cloudflare's 2,600 endpoints as tools would cost 1.5 million tokens.
- Code Mode reduces this token overhead by 99.9 percent using a V8 isolate sandbox.
- Agents now inhabit the state machine rather than just calling external programs.
- The Harness architecture provides a secure, capability-based environment for long-running workflows.
- Generative UI allows for interfaces built dynamically for specific user tasks.
- Developers must optimize documentation for agents who dream in types and syntax errors.
The Future of MCP — David Soria Parra, Anthropic
David Soria Parra · AI Engineer · 18 min
Watch on YouTube →David Soria Parra from Anthropic outlines the roadmap for the Model Context Protocol. MCP is evolving from local experiments to an enterprise connectivity standard.
- MCP adoption is growing twice as fast as React did in its early days.
- The protocol currently sees 110 million monthly downloads across its ecosystem.
- The 2026 connectivity stack includes Skills, MCP, and Computer Use.
- Progressive Discovery allows agents to load tools on-demand to prevent context bloat.
- New specifications for June 2026 include stateless transport and server discovery via well-known URLs.
- Enterprise features like centralized authorization and cross-app identity are the next priority.
Why half of product managers are in trouble | Nikhyl Singhal (Meta, Google)
Nikhyl Singhal · Lenny's Podcast · 95 min
Watch on YouTube →Nikhyl Singhal discusses the existential crisis facing product managers on Lenny's Podcast. AI is automating the coordination tasks that previously defined the role.
- Product management is shifting from a coordination role to a raw judgment and execution role.
- Companies may shed 30,000 staff and rehire only 8,000 in AI-first positions.
- The obsolescence mindset requires PMs to use software to eliminate their own administrative overhead.
- The equal disappointment algorithm describes the struggle to balance industry pace with personal life.
- Career planning should focus on the skip move: the job after the next one.
- Modern PMs must overcome the psychological threshold of reinvention to survive.
References
PeopleSunil Pai (https://sunilpai.dev/) · David Soria Parra (https://x.com/dsp_) · Nikhyl Singhal (https://x.com/nikhyl) · Lenny Rachitsky (https://twitter.com/lennysan) · Kenton Varda · Matt Carey · Amol Avasare · Sam Altman
ToolsCode Mode · Model Context Protocol (MCP) · V8 isolates · The Harness · Generative UI · React · REST