TLDR AI 2026-01-22

That review step is where bugs slip through, context gets lost, and teams slow down.

Greptile reviews each PR with full repo context and learns your team's conventions over time from comments, reactions, and what gets merged. It flags issues and suggests fixes that match your team, not generic best practices.

✅ Now integrated with Claude Code: install via /plugin.

✅ Free for open source.

✅ Trusted by engineering teams at NVIDIA, Scale AI, and Brex.

See why devs love Greptile

Anthropic has published a new constitution for Claude. The document is a detailed description of Anthropic's vision for Claude's values and behavior that explains the context in which Claude operates and the kind of entity the company would like Claude to be. The constitution is a crucial part of the model training process, and its content directly shapes Claude's behavior. This post describes what's included in the new constitution and some of the considerations that informed Anthropic's approach.

OpenAI's CEO, Sam Altman, has been meeting with investors in the Middle East to line up funding for a new investment round that could total at least $50 billion. He recently visited the region, where he spoke with some of the leading state-backed funds in Abu Dhabi. The talks are early, and the amount could change. OpenAI has also recently held talks with Amazon to raise at least $10 billion.

Meta's new artificial intelligence lab has delivered its first high-profile models internally. The company's CTO, Andrew Bosworth, says the models show a lot of promise. There is still a tremendous amount of work required in the post-training phase before the models can be delivered in a way that's usable internally and by consumers. Meta is starting to see favorable returns from its big gambits in 2025. The next two years will be important for bringing consumer products to market.

Pass@k is the probability that at least one of k different attempts will succeed. It is one of the most common metrics used for agents. The problem with the metric is that it's exponentially forgiving, and humans interacting with agents aren't nearly that forgiving. Pass@k should be a metric that's rarely used, and carefully justified every time it is used.

Novita AI introduced performance optimizations for GLM4-MoE models using SGLang, achieving faster Time-to-First-Token and better token generation speed under agentic coding workloads.

This post contains a curated list of news, tutorials, tips, and articles on Claude Code. It covers recent upgrades, tools that complement Claude Code, and more. The post provides advice on how to skill up with Claude Code as well as predictions on where the technology is headed.

AI is everywhere, but something is holding orgs back from scaling and governing them in prod. Over 1,000 senior tech and business leaders spoke to Camunda about their challenges. In this report, you'll see how teams are managing risks and improving orchestration to deliver reliable AI agents. Get the report

When the Model Context Protocol (MCP) exploded a year ago, everyone rushed to build MCP servers. A year later, most MCP servers disappoint. While developers blame the protocol, enterprise adoption tells a different story. Companies are deploying MCP servers, and integrations are live. This post breaks down why MCP servers fail, best practices for building ones that work, and how Skills and MCP complement each other.

This implementation introduces token-wise branch-and-merge reasoning to enable more expressive multi-path computation while keeping token representations compact.

Devin Review is a code review tool that uses AI and UX to scale human understanding of complex code diffs. It is currently free and works on any public or private GitHub PR. The tool helps in every step of the PR process. It allows developers to chat about changes without leaving the review.

Notion is expanding its Custom Agent platform to include Slack integration and plans for Calendar and Mail connectors for automation. Integrations with third-party services like Cursor, Linear, and Ramp are in development, enhancing project management and finance tasks. New sections like Feed, Library, and Workers will enable developers to create custom integrations, while AI features like an AI co-editor enhance productivity.

Apple plans to transform Siri into a chatbot, similar to ChatGPT, with the expected launch integrated into iOS 27. The revamped Siri, codenamed "Campos," will support both voice and text inputs, marking a strategic shift due to competitive pressure. Apple has chosen Google's Gemini as its AI partner after evaluating options like OpenAI and Anthropic.

Is agentic AI your next step? Beware of getting bogged down with custom integrations. Algolia's white paper breaks down how to build AI agents on a deadline. Read it now

Aliisa Rosenthal, OpenAI's former sales leader, emphasizes context engineering as a key advantage, particularly in enterprise applications, and advocates for developing lighter, cost-effective models.

Two of the fifteen external contributors who helped review Claude's new constitution are Catholic members of the clergy.

Apple's strategy focuses on its durable ecosystem, prioritizing hardware sales over AI innovations, despite the industry's AI investment focus.

AgentOS provides a production-ready runtime for multi-agent AI systems, solving the infrastructure challenges that prevent most projects from shipping.