Everything Is Code

01Peter Steinberger, OpenClaw & The Biggest Aqui-Hire of the Solopreneur Era

OpenAI acquired Peter Steinberger — the solo creator of OpenClaw, the open-source AI agent framework that took over developer Twitter. One in three startup acquisitions today is talent-driven. A company pays for a team, not a product. Ben did the math: if you acquire a 10-person company for $20M, that's $2M per engineer. When you factor in recruiting costs and the fact that these people have already proven they can ship, aqui-hiring is often the most rational move.

The Steinberger numbers? Nobody actually knows. Grok speculated $200–300M on the conservative end, up to $5–10B on the high end. He sold his previous company for around $110M. He was being courted by both Meta and OpenAI — Mark Zuckerberg is known for wild offers (reportedly $27B for Alexander Wang at Scale AI). Sam Altman ultimately won. Steinberger's condition: OpenClaw stays open source. That was non-negotiable. It's also one of the smartest marketing moves OpenAI has ever made — OpenClaw has been everywhere, and Sam effectively acquired a community.

Within 48 hours, Anthropic changed their ToS to block OpenClaw subscriptions. OpenClaw was driving enormous Claude usage — users burning hundreds to thousands of dollars a day in Anthropic API credits. Blocking that is either about safety concerns or competitive self-interest. Luca's read: "I can't understand what motivated them to be so aggressive in blocking access. OpenClaw is really driving up usage of these models."

The bigger point: OpenClaw gives you the ability to own your data, own your agent structures, and swap out the underlying model without starting from scratch. If you're storing your conversations, memory, and prompts inside a closed platform — you don't own any of it. OpenClaw is closer to web3's original promise of portable identity than any NFT project ever was.

02OpenAI's $100 Billion Funding Round — Bubble Territory or Inevitable?

OpenAI is finalizing a $100B fundraise. NVIDIA is allegedly in for $30B. Amazon reportedly $50B — despite being one of Anthropic's biggest investors. The spray-and-pray is real.

Ben's take: when everyone invests in everything, that's one of the clearest bubble indicators. But it's more nuanced. The DeepSeek question looms: remember when DeepSeek leveled the playing field? Ben thinks we're a few months away from another moment like that. Models are converging. Someone will build a Claude-level model at a fraction of the cost — and they'll probably do it by scraping the existing frontier models, just like DeepSeek did.

Where does the value actually accrue? The consensus: NVIDIA. They're the picks-and-shovels monopoly. Google TPUs are cheaper but less performant per watt — and in an energy-constrained environment, watt-for-watt performance is everything.

The circular investment structure: OpenAI invests in CoreWeave → CoreWeave leases Nvidia GPUs → Nvidia invests in OpenAI → OpenAI buys Nvidia GPUs. XAI is notably absent from this web. As Rik put it: if two dogs fight for a bone, the third one runs with it.

03The Non-Handshake That Launched a Thousand Tweets

Sam Altman and Dario Amodei stood next to each other at a PM photo op in India and very visibly did not shake hands. Our read: Sam instigated it. Luca disagreed — Dario looked frozen, internally furious but too professional to make a scene. Rik's theory: maybe Sam took the Super Bowl ads personally. Anthropic ran targeted Super Bowl ads that were pointed shots at OpenAI. Sam's angry tweet about them did more to spread those ads than the ads themselves ever would have.

The philosophy split:

04Data Centers in Space: Insane Idea or Inevitable?

Elon wants to launch data centers into space. Ben ran the math live using ChatGPT.

For an equivalent 10-megawatt data center: • On Earth: $60M–$325M all-in • In Space: $3.5B–$5.5B (just launch costs at $2,500/kg for ~1,000 tons)

That's roughly 20x more expensive. Seems insane — until you hear Luca's reframe: the real bottleneck isn't cost, it's energy. The US grid can't keep pace with AI compute demand. Nobody wants a massive data center near their city. If you move compute off-planet and use constant-angle solar (instead of Earth's 50% efficiency), the economics might improve over a 20–30 year horizon.

Rik's point: if Elon makes this work, businesses like CoreWeave — whose entire model is renting data center access — are structurally vulnerable. The big players (Meta, Amazon, OpenAI, Anthropic) are all racing to own infrastructure anyway. CoreWeave is one technology shift away from obsolescence.

05Claude vs ChatGPT: The Numbers Are Shifting

Ben pulled the search volume data live. Three months ago: ChatGPT at 70, Claude at 3. Now: ChatGPT at 87, Claude at 8. ChatGPT grew 15%. Claude grew 120%.

Engineers were the early adopters. They drove Claude's rise. But AI is eating software — the tools engineers adopt become the tools the next layer of builders adopts. Rik's housemate said something that stuck: everything is code. Images are code. Videos are code. Photo editing is code. Motion design is code. And Claude Code is the frontier model for all of it.

The token cost trend: the cost of a million tokens has dropped dramatically since GPT-3. Models have gotten exponentially better while getting cheaper. As Rik put it — AI is the worst it will ever be today.

Ben deleted ChatGPT from his devices. He's running everything on Claude.

06The AI Agency Opportunity — And Why Mark Cuban Is Right

Mark Cuban said it on video and we played the clip: who's going to help 30 million solopreneurs and small businesses implement AI to get a competitive advantage?

The answer isn't a software product. It's a person. An integrator. Someone who walks into a business, maps the processes, identifies what can be automated, and builds the custom AI layer on top of existing tools.

Rik is actively pursuing this — in talks with a short-form content agency about mapping their content creation, analytics, ad distribution, and UGC briefing workflows. The goal isn't to introduce new tools. It's to make existing tools work together better, with AI reducing the number of humans in the loop.

Ben's bold vision for what OpenClaw unlocks for outreach: connect Resend, use Brave Search to identify 500 target businesses, have Claude analyze each website, auto-generate a rebuilt version, and send a cold email saying "I rebuilt your site — click here to buy it for $500." That's not AI slop outreach. That's tangible value delivery upfront.

The real play: business process mapping as a service. Find the manual processes. Automate them. Save the client money. The tools to do this have never been more accessible.

07Operating Systems for Humans

The idea that will stick with us longest from this episode.

Rik articulated something genuinely new: OS yourself. Not just productize yourself (the 2024 meta). Build an operating system around your life. Plug in health data, project notes, client conversations, business metrics, Telegram messages. Create one central brain. Build it on something like OpenClaw so you own the data and can swap models as better ones emerge.

We've never had the ability to have this much structured, queryable knowledge about ourselves and our work. But it only works if the data architecture is right. Garbage in, garbage out — but if you structure it correctly, you have a compound advantage that grows with every week you use it.

Luca extended this to the corporate world: companies that have been through digitization have data. Lots of it. Most is siloed, unstructured, and not actionable. The businesses that build an internal AI layer on top of their existing data — without replacing their tools — will outcompete the ones that don't. The window is open right now. In two years, it'll be table stakes.

08Naval on Vibe Coding: "The Hottest New Programming Language Is English"

We're saving the full Naval episode for next week, but Rik flagged it as listen of the week. Naval's thesis: vibe coding is the new product management. English is the new programming language. There's no demand for average.

This is something the three of us have been saying for six months. Naval confirming it means it's crossed from builder Twitter consensus to mainstream framing. Which means the window of first-mover advantage is narrowing — not closed, but narrowing.

If you haven't started building with AI-native tools, the time to start is now. Not because it's fun (it is), but because the people who get their reps in now will have the intuition and pattern recognition that can't be shortcut when competition intensifies.

📊 The Weekly Scorecard