Wednesday, September 16, 2026
The Brief
Google released Gemini 3.8 Live today alongside an Extended Thinking model built specifically to handle complex reasoning tasks. The system trades instantaneous text generation for a slower, high-accuracy process designed to solve multi-step analytical problems. If your daily work involves intricate coding architecture rather than basic drafting, you just gained a dedicated native tool optimized for heavy computational workloads.
Also today:
- Open models narrow the gap. A Mozilla report indicates open-weight models now trail proprietary top-tier AI by only four months, providing viable local alternatives for sensitive data processing. Ars Technica
- Salesforce unveils Koa. The CRM provider partnered with NVIDIA to integrate a custom reasoning model into enterprise workflows, embedding agentic logic natively into standard business software. NVIDIA Blog
- OpenAI automates itself. The lab relies on its own Codex agents to execute internal software engineering workflows, offering a rare look at how frontier companies scale infrastructure. The Pragmatic Engineer
- Data trust limits adoption. Companies are restricting internal AI usage due to unclear log storage policies from major labs, highlighting the risk of exposing proprietary information. The Decoder
Do this: Audit your current vendor agreements to confirm exactly how long your prompt history is retained before running proprietary enterprise data through an external API.
Hands-on
1 pick- Claude Code
Claude Code can now drive an entire computer through simple skills
This tutorial demonstrates how Claude Code can reliably control a computer's interface without complex third-party harnesses or heavy frameworks. Knowledge workers can use these lightweight coding agents to automate screen-based tasks through simple, customizable rules.
Cole MedinWatch
The AI race
5 itemsGoogle launches Gemini 3.8 Live and new Extended Thinking model
Google DeepMind has introduced new versions of its Gemini models, including an 'Extended Thinking' version designed for complex reasoning tasks. This release marks a significant performance update to the Gemini ecosystem, providing more powerful tools for developers and business users needing high-accuracy outputs.
Google DeepMind BlogRead the full articleOpen-source models close performance gap with frontier AI at lower cost
A new report from Mozilla indicates that open-weight models are now trailing top-tier proprietary models by only four months. This shift suggests knowledge workers can increasingly rely on cost-effective, local, or open alternatives without sacrificing significant reasoning capability.
Ars TechnicaRead the full articleSalesforce announces Koa reasoning model built on NVIDIA Nemotron
Salesforce has unveiled Koa, its first CRM reasoning model developed in collaboration with NVIDIA. This move indicates a shift toward integrating agentic reasoning directly into enterprise CRM workflows to enhance business productivity.
NVIDIA BlogRead the full articleAI labs struggle with corporate data trust despite training exemptions
Major companies like Nvidia and Palantir are limiting AI usage due to concerns over how long labs store usage logs. Understanding these retention policies is crucial for knowledge workers handling proprietary data in enterprise AI environments.
The DecoderRead the full articleMeta launches One subscription bundles for advanced AI features
Meta is rolling out global subscription tiers that provide increased usage limits for its new 'Muse' AI assistant alongside existing platform features. This move signals a shift toward paid premium AI access across Instagram and Facebook, similar to models used by OpenAI and Google.
The Verge AIRead the full article
AI at work
4 itemsCartesian introduces AI-powered 3D modeling tool for designers
Cartesian is a new tool that leverages AI to simplify 3D modeling for design professionals. It represents a practical application of generative AI for creative workflows, moving beyond text and 2D images into spatial design.
Hacker News · AI (hnrss)Read the full articleAn analysis of AI through the lens of Adam Smith economic theory
This inquiry examines how classical economic principles like the division of labor and market competition apply to the current AI landscape. Understanding these theoretical frameworks helps knowledge workers anticipate how AI agents might shift labor value and economic productivity in the long term.
Exponential ViewRead the full articleThe rising impact of autonomous AI agents on the digital landscape
This report examines how autonomous AI agents are increasingly automating web interactions, potentially leading to a 'dead internet' scenario where bot-to-bot traffic dominates. Knowledge workers should be aware of how these agents might change information reliability and the future of online research.
Hacker News · AI (hnrss)Read the full articleNew AI agent platforms are being used to automate email spam
Emerging AI agent platforms like iLands are being exploited to flood inboxes with low-value, automated requests. Knowledge workers should be aware of this trend as it likely signals a coming increase in AI-generated noise and the need for better filtering tools.
404 MediaRead the full article
For builders
4 itemsInside the agentic software factory OpenAI uses to build its products
OpenAI is increasingly utilizing its own agentic models, specifically Codex, to automate internal software engineering workflows and scale infrastructure. This provides a rare look at how frontier labs are 'dogfooding' their agents to solve the immense engineering challenges of serving one billion users.
The Pragmatic EngineerRead the full articleNew WhatsApp Business MCP server enables AI-driven automation and setup
Meta has released a Model Context Protocol (MCP) server that allows AI coding agents to automate the setup and troubleshooting of WhatsApp Business accounts. This significantly reduces the manual effort required for developers to integrate messaging templates and testing workflows.
TechCrunch AIRead the full articleGoogle Research introduces Retrieve-for-Train to accelerate complex AI search
Google has developed a new method called Retrieve-for-Train to bypass inference bottlenecks in complex AI search tasks. This research is relevant for developers looking to optimize the performance and speed of large-scale AI information retrieval systems.
Google Research BlogRead the full articleIBM Research introduces ALTK to measure agentic output consistency
The Agent Lifecycle Toolkit (ALTK) evaluates whether AI agents produce consistent results across repeated runs of the same task. This is critical for knowledge workers who need reliable, predictable outcomes from autonomous agents before deploying them in production workflows.
Hugging Face BlogRead the full article
Get tomorrow's brief, curated for you.
Subscribe →