Agentic AI, LLM engineering, and the models behind modern automation — multi-agent systems, LLM evaluation and comparisons, RAG, fine-tuning, AI infrastructure, security, and production AI engineering.
From the blog
RNNoise compiled to WASM and loaded into an AudioWorklet processes 48 kHz audio at ~13 ms latency, runs noise-gate hysteresis, and converts Float32 to Int16 PCM — all on a non-DOM audio thread.
Claude 1M, Gemini 2M — but 1M tokens of garbage is still garbage. We unpack the 2026 long-context prompt anatomy, the MRCR recall numbers (Claude 78.3% vs Gemini 26.3%), the prefill latency tax, and the structure CallSphere uses to keep agent recall above 90% past 500k tokens.
Infrastructure-level look at Computer Use 2.0 virtualized desktops, including agent infrastructure, deployment topology, region availability, and cost considerations.
How to remember a user across days, sessions, and devices without leaking context between accounts. The three-layer pattern that works at production scale safely.
LangGraph 1.0's interrupt() primitive makes human-in-the-loop a one-liner. Approval queues, edit-and-resume, and dynamic break points in production.
Pydantic AI's April release tightens the typed-agent loop and adds structured tool definitions. Why type-safe agents reduce production bugs and speed iteration.
Snowflake's MCP server exposes Cortex, semantic models, and SQL to agents cleanly. How to build a data agent without giving it raw warehouse access or query budget risk.
Honest production comparison of OpenAI AgentKit 1.0 versus CrewAI for multi-agent orchestration — DX, observability, scaling, and cost.
AppOmni rolled out an AI agent permissions module in April 2026, mapping OAuth scopes for Slack, Salesforce, and ServiceNow agents to a least-privilege baseline.
