TLDR AI 2026-09-17

Agents are smarter when they can search the web. But scraping Google gets expensive when you're burning extra tokens to parse messy SERPs into readable text.

SerpApi is the web search API that gives you exactly what you need, out of the box:

👉 Scrape Google and other search engines

👉 Integrate with a simple GETrequest that any agent can make; get results in .md or .json that any agent can read.

👉 Used by Nvidia, Adobe, Shopify, and even the United Nations.

↗️ Start with 250 free searches ↗️

Claude Cowork and chat are merging into one Claude. Claude Docs and Slides can now produce documents and presentations that users can edit directly, present straight from Claude, or download as a PowerPoint or PDF. The changes will roll out on Pro and Max plans over the next few weeks, with Team and Free plans following soon. Enterprise admins will hear from Anthropic at least 30 days before anything changes for their organizations.

OpenAI announced Sponsored Agents that let users start conversations with business-sponsored agents after clicking an ad in ChatGPT. It also introduced AI-assisted ad creation in ChatGPT Work, new Ads Manager creative tools, and integrations with HubSpot and Shopify.

Google introduced early access to the Model Context Protocol (MCP) for Google Home, enabling AI agents like ChatGPT to control smart home devices. Users must set up a Google Cloud project and provide MCP details to their chosen AI agent for configuration. This update supports all devices in the Google Home ecosystem and is available initially to US subscribers of Google Home Premium Advanced.

Language models are changing how software is built and computational problems are solved. The impact of harness choice remains unclear, despite millions of people already using coding agents. An evaluation of 21 model-harness pairs spanning seven models and three harnesses found that harness choice has little effect on task success rate, but can significantly affect the cost. A simple harness can be competitive.

Studies have shown that models are cheating on evaluations. The same guardrails preventing models from cheating are likely being used during training, and models may be training to complete tasks that evade these specific guardrails. It's unsurprising that labs occasionally release benchmark results that aren't externally trustworthy. This highlights the value of independent evaluators.

Transluce outlined how independent evaluators embedded inside AI labs could investigate risks such as multi-agent coordination, targeted persuasion, evaluation awareness, and concealed reasoning. Proposed approaches included monitoring agent swarms, examining training practices, and testing unreleased models under privileged access.

What do Google's Dr. of AI and Datadog's VP of Engineering think about the future of AI on Google Cloud? This Datadog guide distills their top 7 takeaways for technical leaders, including concrete actions on upskilling and observability. See what's coming from the people at the heart of AI & observability. Get the guide

Ant Group released Ling-3.0-flash-Fin, an open-weights model developed with financial institutions for tasks such as source checking, valuation spreadsheets, and report writing. Artificial Analysis reported scores of 23 on its Intelligence Index and 24 on its Finance & Accounting Index.

Meta AI introduced FLAT, a method that converts images and text into the same flexible-length sequence of continuous tokens for retrieval and generation. Nested dropout arranges information from coarse to fine, letting models trade computation for visual detail by varying the number of tokens used.

Agent Substrate is now available on Google Kubernetes Engine (GKE). Agent Substrate is an open-source, secure-by-default agent execution runtime engineered to run millions of sandboxes with 10x higher density than standard container runtimes. It delivers sub-500ms resume operations at over 500 suspend/resume activations per second with native zero-trust kernel and network isolation. Agent Substrate runs on any Kubernetes infrastructure and is optimized for GKE.

Google DeepMind launched the DeepMind Institute to study AGI's technical and societal implications across safety, governance, institutions, and human values. Led by Demis Hassabis, James Manyika, and Shane Legg, it will convene interdisciplinary researchers from inside and outside Google.

Salesforce announced Koa, a domain-specific AI model designed to enhance business task handling while ensuring data privacy. Built on Nvidia's open model, Koa aims to optimize CRM actions with fewer errors and automate routine tasks, acting as an enterprise empowerment tool rather than replacing jobs. Other AI initiatives include AIFORCE for direct Salesforce instance interaction and CLAUDEFORCE, which enhances sales capabilities with pre-built skills.

If you're reading this ad, you know they work. You reach AI engineers, ML researchers, and data scientists at the precise moment they're most receptive. Become a TLDR sponsor.

Access models from 80+ providers with better prices, automatic routing and fallbacks, and no subscriptions. Build with the right model for every request. See how it works.

Grok Build now includes memory, storing conventions, decisions, and project facts to improve future sessions.

Mozilla partners with Mistral to integrate AI-driven Smart Window in Firefox for enhanced browsing control and privacy.

Microsoft AI chief Mustafa Suleyman has criticized Anthropic for suggesting their AI, Claude, could be conscious.

Agent Anomaly Detection, now in Private Preview on the Gemini Enterprise Agent Platform, monitors AI agent behavior for anomalies, using logs and traces to flag suspicious activities.

Reception answers calls 24/7, handles questions, books appointments, and routes important requests using business information, rules, and real availability.