TLDR AI 2026-02-04

AI agents are everywhere - so are the risks.

In this 3-part series, Foundations of AI Security: What, Why, and How, experts from AWS and Zenity provide the playbook for deploying secure AI across your organization. Topics include:

• AI Agents Uncovered: AWS explains how AI agents transform workflow and create value across enterprises.
• AI Agents Under Attack: Which security blindspots are introduced by agents in an emerging threat landscape.
• Securing AI Agents Now: Learn how real-world security teams are scaling AI security measures.

Save your spot

Xcode 26.3 introduces native support for Claude Agent SDK, enabling full agentic capabilities like subagents, background tasks, and plugins within Apple's IDE. This brings Claude Code's functionality directly into developers' workflows.

Intel is expanding into the GPU market by hiring a new chief architect, as announced by CEO Lip-Bu Tan. GPUs are crucial for AI infrastructure and are in high demand, driven by LLMs. Despite recent challenges, Intel's stock has rallied, buoyed by optimism in its foundry business and significant investments.

OpenAI hired Dylan Scandinaro from Anthropic as its new "head of preparedness." Scandinaro emphasized the rapid advancement of AI and the associated risks. He stressed the urgency of addressing these challenges.

This blog explores open-source AI trajectory since the "DeepSeek Moment," highlighting long-term strategies by major organizations and forecasting sustained momentum through open artifact sharing and deployment-first design.

QA Wolf's new AI assistant helps teams build and maintain automated test coverage for 80%+ of their product workflows—just by chatting with the AI.

• Test complex workflows reliably: Prompts generate Playwright and Appium code instead of flaky plain-English steps.
• Run regressions in minutes: Full suites execute 100% in parallel.
• Own your tests: Open-source code, no vendor lock-in.

Get early access

SWE-Universe presents a scalable method for generating verifiable software engineering environments from GitHub PRs. With in-loop hacking detection and self-verification, the system enables large-scale mid-training for coding agents, producing over 800K tasks.

GLM-OCR is a multimodal OCR model for complex document understanding. It integrates the CogViT visual encoder pre-trained on large-scale image–text data, a lightweight cross-modal connector with efficient token downsampling, and a GLM-0.5B language decoder. The model delivers robust and high-quality OCR performance across diverse document layouts. An SDK for efficient and convenient use is available.

Alibaba's Qwen3-Coder-Next is a new open-weight model fine-tuned for coding agents. Built on a hybrid MoE architecture, it excels in executable synthesis and RL-based environment interaction, achieving strong agentic coding performance at lower inference cost.

Perplexity is a function that measures models' overall level of surprise when encountering a particular output. It has gained significant traction in recent years as both a loss function and as a simple-to-compute metric of model quality. However, it may be an unsuitable metric for model selection. Perplexity will not always select for the most accurate model - any increase in model confidence must be accompanied by a commensurate rise in accuracy for the new model to be selected.

The GenAI Chatbot app market has increased 152% year on year since last January. ChatGPT is losing market share, but this is expected as new capable entrants have launched. Most people have still never used a GenAI Chatbot app. About 20% of AI users use at least two apps, signaling that some apps are better for certain tasks than others.

Vibe coding has yet to reach mainstream consumers, remaining primarily the domain of technical users. Companies like Poke and Wabi are developing consumer-friendly AI products that eliminate complex technical setup and terminology. The real opportunity lies in creating tools that make software development accessible to non-technical users, similar to how Squarespace and Canva democratized websites and design.

Switch from Fireworks AI, Together AI, vLLM to get 99.99% reliability, 2x throughput, and 50% savings. Deploy your open model of choice (e.g. Qwen, GLM, MiniMax) in 1 click. Apply for your inference credit.

This study looks at how Pleiades, an epigenetic foundation model, detects Alzheimer's disease from cell-free DNA in blood.

OpenAI optimized its inference stack for all API customers to make its models faster.

Alex Finn's AI bot, OpenClaw, overcame a task obstruction by acquiring a phone number and integrating voice capabilities to call him for assistance without any prompting.

GLM-OCR is a 0.9B parameter model that delivers state-of-the-art results across major document understanding benchmarks.