MCP is going stateless: What the new spec means for AI agents
MCP's next spec revision goes stateless — New Relic breaks down what that shift means for agents built against today's session model.
AI Daily Recap
14 articles · 5 categories
The finishable daily brief
What happened in AI — Jul 31, 2026
Friday, Jul 31, 2026
14 articles · 5 categories
read top to bottom · then stop
In 30 seconds
- MCP's next spec revision goes stateless, and Dropbox is already using it to surface security context inside AI code reviews.
- DeepSeek's retrained V4-Flash now beats its own flagship V4-Pro on nine agent benchmarks.
- DeepSeek is building a 1-gigawatt AI data center in Inner Mongolia — hyperscaler-scale compute for a single lab.
- Kimi K3 hit 41% of global open-source downloads within two days of launch, overwhelming Moonshot's serving capacity.
- GPT-5.6 pricing dropped 20–80% versus GPT-5.4; Latent Space pegs the cost of equivalent intelligence down 13x in four months.
- OpenAI disrupted a Cambodia-based scam ring that used ChatGPT for investment, romance, and impersonation schemes.
Agent tooling kept moving from spec to production today: MCP's next revision goes stateless, Dropbox is already using it to pipe security context into AI code reviews, and LangChain built a benchmark to score code-review agents against real reviewer feedback instead of synthetic tests.
China's open-weight momentum is now as much an infrastructure story as a model one — DeepSeek is building a 1-gigawatt data center in Inner Mongolia, and Kimi K3 hit 41% of global open-source downloads within two days of launch, straining Moonshot's own serving capacity.
Agent Tooling & Protocols 4 items
MCP keeps hardening for production use: its next spec revision drops server-side state, and Dropbox is already using it to surface security context inside AI code reviews. Alongside that, Amazon Quick shipped a natural-language agent for cataloging data assets, and LangChain built ReviewBench to score code-review agents against real reviewer feedback.
Dropbox Integrates MCP and Dash to Close the Gap Between Security Design and Code Review
Dropbox wired MCP into its internal knowledge platform Dash so AI code-review tools can pull threat models and security requirements straight into the review.
Announcing the Agentic Catalog Experience in Amazon Quick
Amazon Quick's new Agentic Catalog Experience, now in preview, lets data curators describe what they need in natural language and auto-generates Datasets and Topics with inherited semantics.
Evaluating code review agents with ReviewBench
LangChain built ReviewBench, a benchmark that scores code-review agents against real PR feedback from trusted human reviewers instead of synthetic tests.
Developer Tools: AI-Native Workflows 2 items
Two solo-built Show HN launches point at the same instinct: replace legacy productivity formats with ones an AI can write directly, from a local-first knowledge OS to a Markdown-based slide format.
Show HN: Brainstorm – a local-first, AI-native OS for knowledge work
Show HN: Brainstorm is an open-source, local-first “AI-native OS” for personal knowledge management, shipped after three months of solo development.
Show HN: Slaide, open-source Markdown slides AI writes and PowerPoint opens
Show HN: Slaide is an open-source Markdown slide format that AI can write directly and PowerPoint can still open, built to kill manual deck-building.
Model Performance & Pricing 2 items
Model economics moved on two fronts today: DeepSeek's retrained smaller model is now beating its own flagship on agent benchmarks, and OpenAI's GPT-5.6 price cuts point to intelligence getting cheaper faster than headline model releases suggest.
DeepSeek Retrained V4-Flash Beats Its Flagship Pro on Nine Agent Benchmarks - Tech Times
DeepSeek's retrained V4-Flash — its smaller, faster tier — now beats flagship V4-Pro across nine separate agent benchmarks, per Tech Times.
[AINews] GPT 5.6 price cut by 20%-80%: Cost of GPT 5.4 Intelligence dropped 13x in 4 months due to GPT 5.6 recursive self-optimization
GPT-5.6 pricing dropped 20–80% versus GPT-5.4, and Latent Space pegs the cost of equivalent intelligence as down 13x in four months.
Open-Weight Labs: China's Infra & Momentum 3 items
China's open-weight push is now as much an infrastructure story as a model story: DeepSeek is building gigawatt-scale compute, and Kimi K3 is straining serving capacity two days after launch while analysts argue its real moat is the infra behind it, not the open weights.
DeepSeek to Build 1-Gigawatt AI Data Hub in Inner Mongolia - Briefs Finance
DeepSeek is building a 1-gigawatt AI data center in Inner Mongolia, Briefs Finance reports — compute capacity on par with major hyperscaler buildouts.
Kimi K3 Revealed Many Secrets, but Its Most Important Infra Is Hard to Copy: Moonshot AI Infrastructure Engineering as the Real Moat in Open-Source Model Economics - Pandaily
Pandaily argues Moonshot's real moat with Kimi K3 isn't the open model weights but the infrastructure engineering behind serving it, which is much harder to copy.
Overwhelmed Within Two Days of Launch: Moonshot AI Kimi K3 Open-Source Model Goes Global With 41% of World Open-Source Downloads and 4x Paid User Growth Overseas - Pandaily
Kimi K3 hit 41% of global open-source model downloads within two days of launch and grew paid overseas users 4x, per Pandaily, overwhelming Moonshot's serving capacity.
Security, Safety & Policy 3 items
Security framing shifted up a level today: Google Cloud's CISO office is pitching AI threat defense as a board-level agenda item, while OpenAI published both its EU AI Act compliance practices and a takedown of a Cambodia-based scam ring that abused ChatGPT.
Cloud CISO Perspectives: Why AI Threat Defense is the new boardroom baseline
Google Cloud's CISO office argues AI threat defense has become a board-level agenda item, not just a security team concern, in the second installment of its monthly Cloud CISO Perspectives series.
Advancing responsible AI across Europe
OpenAI published its safety, security, transparency, and provenance practices for Europe, framed explicitly around EU AI Act compliance as the law rolls out.
Disrupting a Criminal Scam Operation
OpenAI disrupted a Cambodia-based scam ring that was using ChatGPT to run investment, romance, gambling, and impersonation schemes at scale.
You are caught up for this edition