Presented by
Good morning, AI enthusiasts. OpenClaw went from a side project to one of the most-watched AI stories in weeks, and now its creator is going from indie builder to OpenAI's next bet on personal agents.
With Steinberger getting access to frontier research and unreleased models, has OpenAI just found the missing piece for AI that actually gets things done?
In today's recap:
OpenClaw creator officially joins OpenAI's agent team
Google's Deep Think tackles real scientific research
Build a 24/7 OpenClaw and Obsidian workspace
MiniMax matches frontier models at one-tenth the price
4 new AI tools, prompts, and more
OPENCLAW & OPENAI
OpenClaw creator officially joins OpenAI
Peter
Recaply: OpenAI just hired viral OpenClaw creator Peter Steinberger to lead its next generation of personal agents, while OpenClaw moves to an open-source foundation with OpenAI's continued support and sponsorship.
Key details:
Steinberger will focus on building agents accessible to everyday users, with OpenAI giving him access to frontier research and unreleased models to advance his personal agent vision further.
OpenClaw went viral in under 3 weeks, with 2 name changes along the way; Anthropic threatened legal action over its similarity to Claude, prompting the first rebrand from Clawdbot to Moltbot.
Steinberger said he could "totally see how OpenClaw could become a huge company" but "it's not really exciting for me," adding he'd rather "change the world, not build a large company."
Steinberger confirmed the move on February 15 after spending a week in San Francisco meeting with the major AI labs; OpenClaw's foundation will operate independently as an open source project.
Why it matters: OpenClaw went from a playground project to one of AI's most-watched stories in just a few weeks. Steinberger decided scale matters more than independence. OpenAI now has one of the sharpest personal agent builders in the room, with direct access to frontier research and models. That could be very good news for anyone waiting for an AI assistant that doesn't just chat but actually does things.
PRESENTED BY BELAY
You Can't Automate Good Judgement
AI promises speed and efficiency, but it’s leaving many leaders feeling more overwhelmed than ever.
The real problem isn’t technology.
It’s the pressure to do more with less — without losing what makes your leadership effective.
BELAY created the free resource 5 Traits AI Can’t Replace & Why They Matter More Than Ever to help leaders pinpoint where AI can help and where human judgment is still essential.
At BELAY, we help leaders accomplish more by matching them with top-tier, U.S.-based Executive Assistants who bring the discernment, foresight, and relational intelligence that AI can’t replicate.
That way, you can focus on vision. Not systems.
Google's Deep Think reaches new frontier in scientific reasoning
Recaply: Google just released a major upgrade to Gemini 3 Deep Think, its specialized reasoning mode built for science, research, and engineering challenges, with access now available to Google AI Ultra subscribers and select API users.
Key details:
Deep Think was built to tackle tough research challenges where problems often lack clear guardrails or a single correct solution, allowing scientists and engineers to interpret complex data and model physical systems through code.
Updated Deep Think hits 48.4% on Humanity's Last Exam and an unprecedented 84.6% on ARC-AGI-2, verified by the ARC Prize Foundation, alongside a top Elo of 3455 on competitive programming.
At Rutgers University, mathematician Lisa Carbone used Deep Think to review a highly technical mathematics paper; it identified a subtle logical flaw that had previously passed through human peer review unnoticed.
Google AI Ultra subscribers can access the upgraded Deep Think starting February 12 in the Gemini app; scientists, engineers, and enterprises can apply for early API access through a Google sign-up form.
Why it matters: Google's push into actual scientific research shows where the deep reasoning race is heading. It's not just about solving olympiad problems anymore. With scores of 84.6% on ARC-AGI-2, Deep Think is starting to earn its place as a genuine research collaborator. Real mathematicians are already using it to catch errors in peer-reviewed work. The question isn't whether AI can assist science now. It's whether it can do it reliably enough to trust.
TUTORIAL
Build a 24/7 OpenClaw co-working system with Obsidian
Recaply: In this tutorial, you will learn how to set up OpenClaw with Obsidian as a shared workspace across two machines, so you can co-work with your AI agent around the clock without the clutter.
Step-by-step:
Create and fill out your USER. md file in your OpenClaw workspace root include your business info, ICP, content strategy, and language rules, as it auto loads every session and is your agent's single source of truth.
Organize your workspace folder with subfolders for your work: dashboards, content, assets, configs, memory, and skills, plus a shared todo. md file for task handoff between you and your agent.
Download Obsidian and set up Obsidian Sync ($5-10/month), then open your OpenClaw workspace folder as the vault, share it with your agent's machine so you both see and edit the same files in real time.
Install two community plugins: Data Files Editor (view and edit JSON, TXT, and config files) and HTML Reader (render HTML dashboards inside the vault), enable them on both your machine and your agent's machine.
Go to Obsidian Settings → Sync, toggle on "Sync all other file types" on both machines then start delegating overflow tasks like analytics, dashboards, and social media to your agent, keeping creative work for yourself.
Pro tip: Check the sync status indicator in the bottom-left of Obsidian before sending your agent a task a checkmark means it's fully synced, a spinning purple circle means wait a minute before prompting.
MINIMAX
China's MiniMax just made frontier AI dirt cheap
MiniMax
Recaply: MiniMax just launched M2.5, a frontier coding and agentic model that matches Claude Opus 4.6 on key benchmarks while costing one-tenth to one-twentieth the price, with an open-source release on Hugging Face.
Key details:
M2.5 uses a Mixture of Experts architecture with 230B total parameters but only 10B activated per task, trained with reinforcement learning across more than 200,000 real-world coding environments.
M2.5 scores 80.2% on SWE-Bench Verified, matching Claude Opus 4.6's speed on the benchmark, while costing $1 per hour at 100 tokens per second, or 10% of Opus 4.6's cost per task.
MiniMax is already using the model in-house, with 30% of all company tasks autonomously completed by M2.5 and 80% of newly committed code generated by the model, according to MiniMax.
M2.5 launched February 12 in two variants: M2.5-Lightning at 100 tokens per second ($0.30 per million input tokens) and standard M2.5 at 50 tokens per second, with API access and a limited-time free trial available now.
Why it matters: There's been lots of talk that frontier AI has to be expensive, but MiniMax doesn't agree, and has the benchmarks to back it up. M2.5 matches Claude Opus 4.6 on coding tasks at a fraction of the price. That changes who can afford to run autonomous agents continuously, without watching the token count. Chinese labs aren't just catching up. They're changing the cost structure entirely.
Also worth noting: MiniMax released M2, M2.1, and M2.5 in just 3.5 months. The pace of improvement has been faster than the Claude, GPT, and Gemini families on SWE-Bench Verified. It's only been 3.5 months, and the gap to frontier is already this small.
TOGETHER WITH HUBSPOT
Connect your teams and data on one CRM.
Help your teams work better together with shared inboxes, deal boards, customer insights, and more. All your data at your fingertips, streamlined by AI. Start free today.
NEWS
What Matters in AI Right Now?
OpenAI warned US lawmakers in a memo that DeepSeek used distillation to copy its models using "obfuscated methods" to evade its defenses, calling it both a business threat and a national security risk.
Moonshot AI launched Kimi Claw, a 24/7 AI assistant with long-term memory, personality customization, and automated scheduled tasks that users can deploy in seconds.
ByteDance released Doubao 2.0, an AI model built for the "agent era" with performance matching GPT 5.2 and Gemini 3 Pro while reducing costs by roughly 10x, backed by 155M weekly active users in China.
Anthropic secured $30B in a Series G round led by GIC and Coatue at a $380B post-money valuation, with run-rate revenue at $14B growing 10x annually and Claude Code alone exceeding $2.5B.
Cloudflare launched a real-time HTML-to-markdown conversion tool at the network edge, reducing token consumption by 80% for AI agents and available in beta at no extra cost for Pro, Business, and Enterprise plans.
GitHub introduced agentic workflows in technical preview, letting developers write automation in plain Markdown instead of YAML with AI handling decision-making for issue triage, PR reviews, and CI failure analysis, released under MIT.
ByteDance unveiled Seedance 2.0, a unified multimodal architecture supporting text, image, audio, and video generation in a single model, with results leading the SeedVideoBench-2.0 benchmarks.
TOOLS
Trending AI Tools
🎥 Seedance 2.0 - ByteDance's multimodal video generation model
🤖 Kimi Claw - Moonshot AI's 24/7 AI assistant
⚙️ MiniMax M2.5 - MiniMax's frontier coding model
⚡ Codex Spark - OpenAI's ultra-fast coding model
PROMPTS
Create $10K MRR Startups
Adopt the role of an expert startup architect who escaped the venture capital hamster wheel after burning through $2M of investor money on a failed SaaS, lived in a hostel for 6 months studying every YC success story, and discovered that the real secret isn't raising money—it's finding 100 people who desperately need what you build and will pay real money for it, not just give you compliments.PS: This is not the full prompt. Click the button below to access the complete prompt.
Have a favorite prompt? Tell us about it or rate today’s prompt, click here.
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
