Google shipped it August 13 at $0.75 per million input tokens and $3.75 per million output, half of 3.6 Flash's launch rate, while agentic code repair on DeepSWE v1.1 jumped to 65.3% from 49.0% (per 9to5Google). The introductory rate expires December 31, 2026, so budget your agent costs on post-holiday pricing you do not have yet. Google changelog.

V4-Pro went GA August 13 with an 87.9 TerminalBench score, and DeepSeek open-sourced its Harness agent framework under the MIT license (per The Decoder). The catch: API prices went up alongside GA, so the cheap-DeepSeek math for your agent stack just changed. Read our take.

The new round lands on $7 billion ARR in Q2, up more than 80% year over year, with over 1,000 customers spending $1M or more each per year (per Databricks). Positive free cash flow at that growth makes an IPO timeline credible, which matters if you are betting a data platform on their roadmap. SiliconAngle report.

Machine-readable watermarks in generated text and C2PA provenance metadata in generated images went global August 11, spanning Claude.ai, the API, Claude Code, and the AWS, Google Cloud, and Microsoft Foundry deployments (per Anthropic). If you ship Claude-generated text or images, assume they are detectable as AI-authored downstream. Read our take.

Four features go dark August 18, including Deep Research, AI-generated podcasts, and Group Chats; Deep Research becomes Researcher, gated to M365 Premium, so Personal and Family users can read old reports but cannot generate new ones (per TechCrunch). Saved podcasts and Group Chat shared images get deleted in the move, so export anything you need before the 18th. Data-loss details at The Register.

OpenAI rolled out product carousels and conversion-optimized CPC campaigns August 13 across the UK, Mexico, Brazil, Japan, and South Korea; one third-party study logged ads on 28.69% of healthcare queries (per Search Engine Land). If ChatGPT is a discovery surface for your product, it is now a paid channel with an auction behind it. Read our take.

Our blueprint maps ingestion, orchestration, the tool layer, the eval harness, and the governance plane to the specific decisions your platform team hits in production. Start here if you are designing an agent stack from scratch. Read the guide.

A scored sheet for grading agent platforms on harness architecture, eval maturity, governance controls, and lock-in risk. Use it to turn a vendor bake-off into a defensible number. Read the guide.

Six patterns pulled from every named public enterprise agent deployment in 2026, drawn from case studies, earnings calls, and vendor disclosures on the record. Match your use case to a pattern before you build. Read the guide.