In partnership with

Good morning. It’s Monday, January 26th.

On this day in tech history: InIn 1972, Alain Colmerauer’s team created Prolog, a logic programming language that turned “programming” into constraint solving and unification. It later became a backbone of symbolic AI, expert systems, and the dream of declarative reasoning machines. If you’ve ever built a solver, wrote a type inference engine, or obsessed over Horn clauses—this is your lineage

In today’s email:

• Inside Clawdbot

• Anthropic to launch Security Center as Claude Code traffic spikes

• 5 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

Clawdbot: The open-source AI will full access to your computer

Peter Steinberger, the Austrian-born engineer who built and sold PSPDFKit (now Nutrient) for hundreds of millions, wasn’t done. After cashing out, he felt the void—and built Clawdbot. The open-source AI assistant exploded this past weekend, rocketing past 15–20k GitHub stars in days.

Clawdbot is an AI that runs locally, equips models like Claude with browser control, file ops, and shell commands, and proactively acts as your privacy-first digital employee across WhatsApp, Telegram, Discord, and Slack.

You install it locally (Mac, Windows, Linux, even Raspberry Pi). One curl line, a quick onboarding, and it owns your WhatsApp, Telegram, Signal, Discord, Slack, iMessage, Teams (pick your poison.) Behind the scenes: a single Gateway process handling channels via WebSocket, plus lightweight nodes for heavy lifting.

The magic is agency: Clawdbot controls Chrome to book flights, check you in, scrape sites, fill forms. It reads/writes files, runs shell commands, drafts and publishes blog posts unprompted. Persistent memory keeps context alive across days. Swap Claude, OpenAI, or local models at will. Add 50+ integrations (Gmail, GitHub, Spotify, Home Assistant) or teach it new skills—sometimes by asking it to write them.Real stories from the past week:

• One user snaps meal photos in Telegram; Clawdbot logs calories and macros automatically.

• Another wakes to lights gradually brightening, calendar already scanned, inbox at zero, all initiated by the bot.

• Developers revive dusty laptops as 24/7 agents; crypto traders automate on-chain monitoring and API calls.

• Someone let it monitor internal channels and auto-schedule social posts.

Critics shrug: “Just a wrapper with browser control.” True, but no one else shipped this level of proactive, cross-channel execution self-hosted and free. Setup snags exist (Claude token quirks, security trade-offs with system access, but sandboxing and local-first design keep most power users hooked.)

Clawdbot isn’t polite Siri waiting for your command. It’s a powerful automation tool with no guardrails. Try to run it yourself (with extreme caution, preferably on a new machine!)
→ clawd.bot
→ github.com/clawdbot/clawdbot

Anthropic to launch Security Center as Claude Code traffic spikes

Anthropic is preparing to roll out Security Center for Claude Code, a centralized dashboard that consolidates scans, flags issues across repositories, and lets developers run manual checks by repository or branch.

Formerly called AutoPatch, the feature integrates directly into Claude Code, simplifying security workflows for teams and individuals. Its release comes as OpenAI upgrades Codex and hits “Cybersecurity High.” No launch date has been announced.

Claude’s Excel integration, previously in beta, is now available to all Pro subscribers, embedding AI assistance directly into productivity workflows.

The Claude CoWork effect had interest in Claude Code surging. Similarweb reports strong growth in searches, website traffic, app downloads, and daily active users over the past month. Analysts link the spike to Security Center and expanded integrations, signaling rising adoption of Anthropic’s AI tools among developers managing complex, security-sensitive code. Read more.

• SimilarWeb: Gemini is eating market share, Grok is climbing, and ChatGPT is finally stabilizing

• xAI is opening the door to customized models in Grok with new ‘Dev Models’ section

• Meta blocks teens from AI characters during safety overhaul

• FBI bypasses laptop encryption with Microsoft’s recovery keys

• Apple releases toolkit exposing how unpredictable and fragile generative AI really is

• Clawdbot proves the future of personal AI isn’t in the cloud, it’s on your machine

• Multi-Head Latent Attention slashes memory and bandwidth costs while keeping AI models sharp

• AI adoption hits Japan’s creative workforce: 12% of artists lose significant earnings

• AI meets play: ex-Google founders launch an app that teaches kids by doing, not telling

• This shape-shifting graphene material may power next-gen soft robots

• Developers can now embed GitHub’s Copilot agent into any app with full planning and tool control

• A new study shows AI-exposed jobs were declining before ChatGPT ever launched

• LFM2.5 proves edge AI can think for itself, no cloud required

• Vibe coding may be behind a 60 percent spike in new iOS apps

• Watch: New footage shows China rehearsing robot war at scale

• China's rifle-wielding drone achieves flawless results in tests

• Microsoft-Tsinghua team trains 7B coding model that beats 14B rivals using only synthetic data 

• KPMG’s $450 million “pandemic flop” in Orlando is now a secret AI training powerhouse

• This founder cracked firefighting, now he’s creating an AI gold mine ​​

• How Americans are using AI at work, according to a new Gallup poll

• Watch: Robots are teaming with humans to triage the battlefield

• Origami-inspired ring lets users 'feel' virtual worlds

• Researchers use AI to reverse engineer molecules

• Conversational AI doesn’t understand users, 'Intent First' architecture does

📄 LLM-in-Sandbox Elicits General Agentic Intelligence

📄 Cosmos Policy: Fine-Tuning Video Models for Visuomotor Control and Planning  ​​

📄 VideoMaMa: Mask-Guided Video Matting via Generative Prior 

📄 Stable-DiffCoder: Pushing the Frontier of Code Diffusion Large Language Model

📄 EvoCUA: Evolving Computer Use Agents via Learning from Scalable Synthetic Experience

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on 𝕏!