// blog
AI in production, no smoke.
Weekly trends on agents, models, and automation — analyzed from the trenches, focused on what serves your company.
Subscribe via RSS//
Weekly AI Roundup (November 16, 2026): Gemini Decision 1, self-improving agents, and the open-weight wave
Gemini Decision 1 scores 1,000 options in one pass, Salesforce ships agents that improve their own models, and open weights surge with ERNIE 5.5 and Praxis-1.
//
Weekly AI Roundup (November 10, 2026): Claude Fable 5.2 Takes the Top, TPUs 8 Arrive, and Agents Go Corporate
What moved AI this week: Claude Fable 5.2 and Mythos 5.2, Gemini 4 Carbon open with /swarm, TPU 8 "for the agentic era," IBM+Microsoft Imagine Enterprise, and CrowdStrike\u2019s kill switch.
//
MCP and A2A: The Protocols That Make AI Agents Speak the Same Language (2026 Guide)
Model Context Protocol and Agent-to-Agent became the de facto standard: Adobe, Oracle, and Box already use them. What they are and why they matter for your company.
//
GPT-6.1 Terra and Gemini 4 Carbon: The Week That Moved the AI Floor (Late October 2026)
OpenAI launched GPT-6.1 Terra ($1/M input, 1.05M context) and Google answered with Gemini 4 Carbon for cyber defense. A summary for decision-makers.
//
Enterprise AI Agents in 2026: Why Only 11% Reach Production (and How to Be in That Group)
Deloitte: 38% of organizations are piloting AI agents but only 11% run them in production. Gartner expects 40% cancellations. How to be the exception.
//
Gemini 4 Argon: 1 Million Tokens of Context and First Place in Benchmarks, Explained for Companies
Google launched Gemini 4 Argon with 1M context and GPQA Diamond leadership (96.4%). What changes for RAG, document analysis, and agents in production.
//
LLM Pricing in 2026: Claude Haiku 5.5 at $0.10 per Million Tokens Changes the Math for Companies
Anthropic launched Claude Haiku 5.5 at $0.10/M input tokens and Mistral Large 4 with an open license. What this price war means for automating your business with AI.
//
Introducing ZRN Dev
Why we founded an engineering studio and what "software that survives production" really means.
//
In praise of boring engineering
The most valuable systems are the ones nobody notices. This is how we design for boredom.
//
LLMs in production, without fear
Validated outputs, continuous evaluation, and cost control: how we integrate language models into real systems.