The OpenClaw-ification of AI

February 26, 2026 · Episode Links & Takeaways

HEADLINES

Anthropic vs. The Pentagon

The Pentagon has handed Anthropic an ultimatum in what could be the most crucial battle for AI safety to date. Anthropic has been insisting their technology should not power autonomous weapons or domestic surveillance; the Pentagon wants "all lawful use" with no tech company input on military operations. The conflict erupted after Claude was reportedly used during the raid that captured Venezuelan president Maduro. Defense Secretary Hegseth personally gave Dario Amodei a Friday deadline to agree or face being designated a "supply chain risk" — a label previously reserved for foreign companies like Huawei. He also threatened to invoke the Defense Production Act. Sources familiar with Anthropic say they have no plans to buckle. As of Wednesday, the Pentagon is taking preparatory steps toward the blacklist, reaching out to Boeing and Lockheed Martin about Claude dependencies. Lawfare Media nailed why this matters: "The terms governing how the military uses the most transformative technology of the century are being set through bilateral haggling between a defense secretary and a startup CEO, with no democratic input and no durable constraints."

Stargate Has Stalled

The Information reports that the entire Stargate project is stalling out — the joint venture still hasn't staffed up and there's no data center development under the Stargate banner. Sources say it began falling apart within weeks of the White House announcement as OpenAI, Oracle, and SoftBank couldn't agree on who would do what. OpenAI originally targeted 10 gigawatts in commitments by end of 2025; they currently have 6GW with Oracle and 2GW with SoftBank, with delays across the board. Sachin Katti, who leads compute scaling at OpenAI, pushed back, calling Stargate "the umbrella brand for our compute strategy." It's hard to know what to make of this one, but Wall Street will be watching closely.

NVIDIA Crushes Earnings

NVIDIA delivered another record quarter — revenue up 73% to $68.1B, beating expectations by 11 points, with data center business growing even faster at 75%. They guided even higher for next quarter at 77% growth, still assuming zero China revenue. Jensen Huang said "the demand for tokens in the world has gone completely exponential." NVIDIA also disclosed $3.5B in data center lease guarantees, 4x last quarter. On the software selloff, Huang argued "the markets got it wrong" — existing software firms will build agentic features on top of their products, not get replaced. Despite the blowout, the stock barely moved, further evidence that markets are completely unmoored right now.

MAIN STORY

The OpenClaw-ification of AI

Coming back from a few days offline, the biggest theme across the AI landscape wasn't a single product launch — it was a pattern. More and more products are starting to resemble OpenClaw in ways that are about something much bigger than just competing with a hot project. These aren't features to be copied. They are new fundamental primitives of the agentic era: interactive modalities that follow you wherever you are, persistent work that happens without prompting, and agent capabilities that can interact with all your personal context. They will become ubiquitous because they unlock things people very clearly want.

EVERYTHING IS CLAW

Claude Code Remote Control: "Go Take a Walk"
This was one of the first things that attracted me to OpenClaw — the ability to have an on-the-go coding agent. Remote Control brings that to the mainstream.
Anthropic launched the ability to kick off a Claude Code task in your terminal and pick it up from your phone. Platforms like Replit and Lovable have had mobile apps for months, but this is the first official phone-to-local-machine integration from any major AI lab. It hit like gangbusters — Claude Code PM Noah Zweben had to come back to apologize for capacity issues. The comparison to OpenClaw was immediate, but there are real differences: Claude Code is your work engine plugged into your specific dev environment; OpenClaw is a 24/7 butler managing your digital existence.

Claude Cowork: Scheduled Tasks
Scheduled tasks means Claude stopped being software you talk to and became software that works while you sleep. That's a category change, not a feature update.
One day after Remote Control, Anthropic announced scheduled tasks for Cowork — recurring tasks at specific times, plus plugins and connectors for Google Drive, Docusign, Slack and more. If you've used OpenClaw, you know the magic is largely in scheduled tasks and cron jobs. Simon Willison tested both features and noted they're still early — Remote Control is "a little janky" and scheduled tasks only run while your computer is awake with the app open. But the primitive is there.

Perplexity Computer
Interesting to see other companies figure out what a safer, more productized version of OpenClaw looks like.
Perplexity launched Computer — a general-purpose agent that orchestrates 19 models to run tasks in parallel. CEO Arvind Srinivas argued the differentiator over OpenClaw is access to multiple models rather than just Claude. The capabilities sit somewhere between Claude Cowork and OpenClaw, with a sandboxed environment for safety. An investment manager called Hampton used it to build a real-time NVIDIA analysis terminal using Perplexity Finance, calling it the first AI company to go head-to-head with the Bloomberg Terminal. Others were skeptical — Bloomberg terminals cost $25K/year because of proprietary data, not because of their notoriously terrible front end.

Notion Custom Agents and Airtable Super Agent
It would be a mistake to view all of these moves as companies trying to "catch up" with OpenClaw. They're all recognizing new primitives, not copying features.
Notion launched custom agents — "the AI team that never sleeps." Autonomous, multiplayer, model-agnostic, built for non-technical teams. Airtable launched Super Agent a couple weeks earlier, pushing in similar territory. Creator Andrew Warner called Notion's version "OpenClaw for regular people" — no Mac Mini needed, no CLI knowledge required. The con: you need a seat for everyone on the team.

The Bigger Point
OpenClaw was the starting gun for a totally new era of AI — not a hype phase confined to the junk heap of history.
Even if you think these productized versions make OpenClaw setup unnecessary, one of the best reasons to do the hard work of setting up OpenClaw now is that it gives you a better education in understanding these primitives than products where they're abstracted away. Most people won't take that time — but if you're subscribing to this newletter, you are not most people.