TLDR AI 2026-02-11

Agents are smarter when they can search the web. But Google doesn't offer an API for web search results... so how are AI companies doing it?

Meet the worst-kept secret in the AI industry: the SerpApi web search API:

👉 Scrape Google and other search engines (including AI overviews, Maps, Local, Shopping, and Knowledge Graph)

👉 Integrate with a dead-simple GET request that any agent can make.

👉 Used by Perplexity, NVIDIA, Uber, Adobe, and even the United Nations.

Special offer for TLDR: Reach out via chat/email to claim 3 months at 50% discount (for new customers), or start with 250 free credits/month

xAI co-founder Tony Wu announced his exit from the company on Monday, then fellow co-founder and influential researcher Jimmy Ba announced his departure from the company the next day. Other co-founders, including Igor Babuschkin, Kyle Kosic, and Christian Szegedy, have also departed the company. Elon Musk launched xAI alongside 11 other people in 2023. The company was merged with Musk's aerospace company, SpaceX, earlier this month.

OpenAI will not ship its first hardware device to customers before the end of February 2027. The company has yet to create any packaging or marketing materials for the device. OpenAI will not use the name 'io' in connection with the naming, advertising, marketing, or sale of any of its AI-enabled hardware products. Few details are known about OpenAI's upcoming product, but it apparently won't be an in-ear device or a wearable.

Vercel described its AEO system for measuring how LLMs and coding agents discover and reference its content across development workflows. Unlike standard chat models, coding agents operate within full project environments, requiring infrastructure that simulates real IDE and terminal usage to evaluate search behavior and citation accuracy.

The US economy is buoyed by the AI boom, but nowhere near entirely driven by it. The tech industry's investments in AI have to pay off if they hope to accelerate American growth. Most AI companies still have massive spending capabilities from their other profitable digital ventures, and investors can be easily tapped for additional money at extremely high valuations. There are no signs of the American AI boom slowing down.

WarpGrep is a lightweight RL‑trained retrieval sub‑agent that finds the exact code files and line ranges developers need up to 5× faster than standard models by limiting search to four turns of parallel grep/read/glob calls. This approach keeps the main LLM's context clean, cuts noise, and halves token usage by specializing retrieval, not reasoning. In benchmarks on large repositories, WarpGrep doubled precision and maintained ~5‑second search times while reducing irrelevant context pollution.

Unwrap uses AI + NLP to bring all of your customer feedback into a single view.

✅ Trusted by Stripe, Perplexity, WHOOP, Clay, DoorDash, Southwest Airlines and others.

Learn more about how you can automate customer insights using AI.

OpenAI outlined a guide for agent skills, a system for packaging reusable “skills” as versioned bundles of instructions, scripts, and assets that agents can attach to execution environments. Skills support conditional workflows, code execution, and reproducibility while keeping system prompts lightweight and modular.

AIRS-Bench contains 20 machine learning research tasks defined by problem, dataset, and metric triplets with corresponding human SOTA targets. It measures whether autonomous agents can develop solutions that match or exceed published state-of-the-art results across diverse ML domains.

Kimi Agent Swarm is an AI CEO who helps find researchers, analysts, and fact-checkers without micromanagement. It builds an organizational structure with bosses, employees, and divisions of labor by itself. Agent Swarm excels where work can be parallelized, like broad research, batch downloads, multi-file processing, multi-angle analysis, and long-form writing. It creates the conditions for productive disagreement to avoid groupthink structurally. Agent Swarm is now available to Kimi Top Tier subscribers in early research preview.

Models are now in the range where they are starting to be plausibly dangerous. Evaluations do not consistently identify dangerous capabilities or propensities. Companies will likely not halt model releases without an extreme reason. No one is working to become better prepared.

Many GPT-4o users were in shock after OpenAI's announcement to retire the model on February 13. The model proved very sticky for users and helped drive OpenAI's fast consumer growth. However, it was criticized for being overly sycophantic and has been linked to cases of users developing psychotic delusions. The model's popularity and potential for harm both stem from its ability to build emotional connections with users. OpenAI is reportedly scrapping the model in part due to how difficult it is to contain its potential for harmful outcomes.

The release of reasoning models by OpenAI in 2024-2025 spurred anticipation for rapid AGI development.

Meta AI now allows Facebook users to animate profile pictures with preset animations like wave and heart.

As the first engineer on TLDR's new Applied AI team, you'll build AI agents and composable Claude Skills to let non-technical teammates create their own AI workflows. Learn more.

DialogLab is an open-source framework for authoring and simulating dynamic human-AI group conversations.

RynnBrain is an AI model that helps robots understand and interact with the physical world by identifying objects.

Runway raised $315M in a Series E round, hitting a $5.3B valuation, with backing from top investors like Nvidia and Adobe.