Agents

GPT-5.5 lands with stronger coding, longer-horizon reasoning, and noticeably better agentic behavior — rolling out to Plus, Pro, Business and Enterprise this week. Framed as a step toward the ChatGPT super app strategy.

At Google Cloud Next, Google unveiled two new TPU generations — 8t for training and 8i for inference — and opened Gemini Enterprise Agent Platform with Model Garden access to 200+ models. The clearest direct shot at Nvidia's AI-compute moat Google has taken yet.

Microsoft's Agent Framework hits 1.0 for .NET and Python, with observability, durable state, and MCP tool support built in. Quietly the most credible production-grade orchestrator outside the big labs.

A practitioner's end-to-end tutorial on MCP — the three-layer agent stack, progressive discovery, programmatic tool calling, and hands-on labs from beginner to advanced. Useful if you want to go deeper than the official docs.

The Honeycomb team shares their second round of hard-won patterns for running MCP and canvas-based agent workflows in real observability work. Rare glimpse of the integration grief no vendor doc admits to.

A solo developer's honest sketch of how to actually learn and ship AI in 2026 — the rare roadmap that admits which pieces are still in flux. Worth it for the "what I would skip" list.

Swyx's take on what "slop" actually means for AI engineering — and why the next year of agent work will be decided by who can scale quality-control without drowning in noise. The kind of piece that quietly becomes a reference in team Slacks.

Microsoft shipped in-app agents for Copilot that plug directly into SAP, ServiceNow, Salesforce, and Workday. Users can trigger multi-system workflows from inside Teams or Outlook without switching contexts, the clearest signal yet of Microsoft's "agents wherever work happens" strategy.

A detailed practitioner argument for designing "harnesses"—structured planning environments that make coding agents think before they type. Covers symbol analysis, repository-impact mapping, and scoped permissions with concrete patterns. One of the clearest articulations of how teams are adapting their tooling to agentic workflows.

Five production patterns distilled from Google's internal Agent Bake-Off: treat agents as microservices, specialize sub-agents, use supervisor loops, instrument everything, and constrain the tool surface. Practical rather than promotional.

SAP's month-long coding challenge walks practitioners through building CrewAI agents with real orchestration, RAG, and deployment. Good structured path if you want hands-on time with an agent framework this month.

A candid practitioner walkthrough of building an agent from zero in one sitting—what broke, what worked, and the four-component mental model (LLM, memory, tools, runtime) that finally made it click.

Hard numbers on the agent-driven shift: deployments initiated by coding agents grew 10x in six months and now account for roughly 30% of Vercel's platform traffic. Includes the architectural tweaks the team made to absorb it.

Simon Willison's practitioner-grade survey of where we are: why November 2025 was the real agentic-coding inflection, the three patterns he uses daily, and what the "dark factories" of automated content mean for product builders. Honest, specific, and useful.

End-to-end walkthrough for building an autonomous AI agent from scratch in 2026. Covers the four essential components—LLM, memory, tools, and runtime—with code examples and deployment patterns.