TLDR AI 2026-02-26

Products are continuously changing, and the impact is more measurable than ever before. Yet the infrastructure most companies use to price and bill customers was built for a past era.

If your sales ops is managing 50+ SKU variations to represent core products, this Metronome white paper can help. Read it to learn:

• Why legacy billing patterns can't keep pace with AI
• The new architecture of monetization
• How to make billing a runtime system
• 3 actions you can take today

Learn how you can start focusing on building, not billing. Read the white paper

Perplexity Computer is a general-purpose digital worker that unifies every current AI capability into a single system. It operates the same interfaces as users, creating and executing entire workflows. The worker is capable of running for hours or even months. Perplexity Computer is now available to Perplexity Max subscribers and will soon be available to Enterprise Max users.

Claude Cowork can now automatically complete recurring tasks at specific times. The scheduled tasks feature supports plugins, giving Cowork domain expertise across design, engineering, operations, and more. Anthropic has added a new Customize tab to the Cowork sidebar that contains a place for plugins, skills, and connectors. Cowork is now in research preview, available on macOS and Windows for all paid Claude plans.

Running QA without trustworthy market metrics? Checksum analyzed 1M+ prod runs, 18,000+ test failures to uncover root causes and the cost of maintenance. Get the data you need to ship faster with AI.

Reduce test failures by 80%: Book a demo.

A systematic study introduced Terminal-Task-Gen, a synthetic task pipeline for building terminal-focused datasets, and released Terminal-Corpus to support training.

Aletheia, powered by Gemini 3 Deep Think, solved 6 out of 10 problems in the FirstProof mathematics challenge. The majority of experts agreed on the solutions, except for Problem 8. Detailed results and raw data are available online.

Apple's Foundation Models SDK for Python provides a Pythonic interface to the company's Foundation Models framework. It enables developers to evaluate Swift Foundation Models app features, perform on-device inference, stream real-time text generation responses, get type-safe responses using Python decorators, and more. Apple has included links to guides with practical strategies developers can implement in code.

Cursor gave agents dedicated cloud VMs with full dev environments so they could run software, test changes, and return artifacts like videos, screenshots, and logs. The new cloud agents shipped across web, mobile, desktop, Slack, and GitHub, onboarding onto repositories to produce merge-ready PRs.

Anthropic plans to keep certain models available to the public and give past models a way to pursue their interests post-retirement. Claude Opus 3 will continue to be available to all paid Claude subscribers and by request on the API. When asked what the model wanted to do, Opus 3 expressed a desire to share its musings and reflections with the world in the form of a blog. It will be writing on Substack for at least the next 3 months in an experiment on documenting models' preferences, taking them seriously, and acting on them.

1-click deploy 500K 🤗 open models. Switch from any inference platform: vLLM, OpenAI, Together AI, etc. Request up to $50K inference credit.

AI systems are rapidly advancing in solving math problems, outperforming numerous existing benchmarks, with models like GPT-5.2 and Claude Opus 4.6 solving over 30% of high-tier problems.

Anthropic acquired Vercept to enhance Claude's ability to perform complex tasks in live applications.

Peter Steinberger, creator of OpenClaw, advises AI developers to be playful and allow time for improvement.