Daniel Kornev

There’s a lot of work underway that Pavan shared earlier today. Something that I know has been at the top of a lot of people’s minds is the feeling that feedback goes into a void, without real people to review and see it on the other side. To help with that, I’ll be taking on an expanded role as the exec sponsor of the Windows Insider Program to listen, engage, and help shape what’s ahead with the Windows community. Over the coming weeks, we’ll also introduce you to members of the product teams that will help you get answers on the topics you care about most. I’d love to hear from you. What would you like to see more of from us as we get started?

Macrohard or Digital Optimus is a joint xAI-Tesla project, coming as part of Tesla’s investment agreement with xAI. Grok is the master conductor/navigator with deep understanding of the world to direct digital Optimus, which is processing and actioning the past 5 secs of real-time computer screen video and keyboard/mouse actions. Grok is like a much more advanced and sophisticated version of turn-by-turn navigation software. You can think of it as Digital Optimus AI being System 1 (instinctive part of the mind) and Grok being System 2. (thinking part of the mind). This will run very competitively on the super low cost Tesla AI4 ($650) paired with relatively frugal use of the much more expensive xAI Nvidia hardware. And it will be the only real-time smart AI system. This is a big deal. In principle, it is capable of emulating the function of entire companies. That is why the program is called MACROHARD, a funny reference to Microsoft. No other company can yet do this.

Coding agents didn’t eliminate PMs, designers, or engineers. They eliminated implementation as the bottleneck. Now the constraint is taste, system thinking, and review capacity. Execution is cheap. Judgment is scarce.

The “Claude Marketplace” sounds like a procurement simplification tool. Enterprises can use existing Anthropic spend commitments to buy partner solutions. Anthropic just told you which AI applications it plans to build next and nobody is paying attention. Look at the launch partners. GitLab (code review). Harvey (legal). Lovable (app building). Replit (development). Rogo (finance). Snowflake (data). These are the six workflow categories where enterprises are already paying real money for Claude-powered tools. Anthropic is running at ~$19B in annualized revenue. 80% enterprise. Over 500 customers at $1M+ per year. Those committed spend pools are now flowing through a marketplace Anthropic controls. Which means Anthropic gets granular data on exactly which partner tools enterprises buy, how much they spend, which workflows drive the most usage, and where the willingness to pay is highest. This is the AWS Marketplace playbook. Amazon launched Marketplace to help enterprises consolidate cloud procurement. Then it watched which SaaS categories grew fastest. Then it built those products itself. Amazon RDS, Amazon Connect, AWS Lambda, all started as categories where third-party tools were thriving on AWS. Every partner joining the Claude Marketplace is handing Anthropic a roadmap. Harvey proves legal AI has enterprise willingness to pay at scale? Anthropic already has Claude for Financial Services and Claude for Life Sciences. You think Claude for Legal isn’t coming? The partners benefit in the short term. Fortune 10 access with pre-approved budgets is a cold-start solution most developer tools spend years trying to build. But the long game favors the platform. Meanwhile, every partner selling through Anthropic has switching costs compounding quarterly. Anthropic handles invoicing, procurement, distribution. The enterprise buyer consolidates AI spend under one commitment. Try moving that to OpenAI when your CFO just approved a $3M Anthropic commitment that covers six different tools. Six partners today. The real number to watch is which categories Anthropic enters directly within 18 months. The marketplace is the map. Anthropic is reading it.

I have been testing GPT-5.4 Pro extensively (while it was on a flight and after the release). For my high-complexity, math-heavy tasks it is a step backward. My best guess is, the agentic harness pollutes the context: it keeps looking for skills, surprised the folder is empty.

The key to better agent memory is to preserve causal dependencies.

introducing polymarket cli - the fastest way for ai agents to access prediction markets built with rust. your agent can query markets, place trades, and pull data - all from the terminal fast, lightweight, no overhead

Robotics just proved it can scale like language models. SONIC trained a 42 million parameter model on 100 million frames of human motion and achieved 100% success transferring to real robots with zero fine-tuning. The breakthrough isn't the robot doing backflips. It's that someone finally found the "next token prediction" equivalent for physical movement. For years, training robots meant hand-crafting reward functions for every single skill. Want your robot to walk? Design rewards for balance, foot placement, energy efficiency. Want it to dance? Start over with entirely new rewards. This approach hits a wall because humans can't manually specify every nuance of natural movement. SONIC replaces this with motion tracking: the robot learns by watching 700 hours of motion capture data and trying to mimic it, frame by frame. The data itself becomes the reward function. Scale the data, scale the model, scale the compute, and performance improves predictably. Just like GPT. This unlocks something robotics has never had: a universal control interface. One policy handles: 1. VR teleoperation using head and hand tracking 2. Live webcam feeds converted to robot motion in real-time 3. Text commands like "walk sideways" or "dance like a monkey" 4. Music audio where the robot matches tempo and rhythm 5. Vision-language models for autonomous tasks (95% success rate) All inputs get encoded into the same token space, then decoded into motor commands. No retraining. No reward engineering. No manual retargeting between human and robot skeletons. If this holds, robotics just closed a 5-year gap with AI. Language models scaled by finding one task (predict the next word) that generalizes to everything. Vision models did the same with image classification. Robotics now has motion tracking. Expect the next wave of humanoid companies to train on billions of frames, not millions.