Coding with AI agents today feels like pair programming, with the human as navigator. As multi-agent systems evolve with more context, that role morphs into... "hand-wavy architects" :D

With native MCP integration built into AI models, the boundary between where the model ends and product begins blurs further. In other words, "Context engineering" is *the* product.

Native MCP integration on AI models inverts the dependency and potentially, the value capture. x.com/vishalmanohar/…

Remarkable to watch Zuck bounce back from lackluster Llama 4 to double down and hire aggressively from OpenAI etc. Apple, on the other hand....

Halfway into "A brief history of intelligence" - such a fascinating read. Brilliantly juxtaposes evolutionary milestones of the brain with AI research.

Google Fit misclassifies raged rickshaw rides on Bangalore's bumpy roads as bike rides. Will take the points but case of overfitting .Google? #fitindia

Microservices are just a fancy way of distributing your bugs across multiple servers.

The current era has a "It was the best of times, it was the worst of times..." vibe.

Given the ability to steer a coding agent is proportional your own coding ability, it probably makes sense to code without LLMs once in a while - to make sure your own skills do not atrophy.