Safetymogging and marginmaxxing the accelecels

1. Key Themes

AI safety rhetoric and lab financials are two sides of the same self-interested coin

Rechtman argues that Anthropic's public safety advocacy and its recently leaked financials are not separate stories but the same strategic play, because the company's core assets depreciate too fast for it to unilaterally slow down.

"So if you believe that Dario and the Anthropic leadership genuinely does believe in AI Safety as a worthy cause, it is impossible for them to act on it unilaterally because of how quickly their core assets (models) depreciate. If everyone doesn't slow down at once, Ant dies. Conveniently for them if everyone (foreign and domestic) were to slow down, they would win!"

"Adjusted" profitability metrics from AI labs should be treated with extreme skepticism

Anthropic's claimed profitability excludes the very costs that define its business model, drawing a direct—and unflattering—comparison to WeWork's infamous metric.

"Anthropic leaked some pseudo-financials that claimed they're profitable and high gross margin… if you exclude stock comp, training, and revenue share - so you know all their costs…" "Ant's margins are a sanely named but utterly insane concept."

Technology adoption fails for social/political reasons, not technical ones—and physical AI is next in line

Rechtman draws a historical pattern across energy, vaccines, and automation to argue that fear-driven underinvestment, not technical limitation, is the real risk to deploying robotics and physical AI.

"The future is not going to be wholly discontinuous with the past. And in the past when a whole class fails, it hasn't failed because of technology. It's failed because of very human problems that get in the way of successful deployment." "Datacenters and LLMs are on their way to becoming the most salient and negative issue areas since the Iraq War."

The investable opportunity in physical AI is in the "trust layer," not just core capability

Given the reputational and regulatory risks facing robotics/automation, Rechtman sees a market forming around de-risking deployment.

"It seems like there's a whole universe of opportunities specifically around safety, compliance, and insurance to try to smooth off some of the sharp edges." "Safety will both drive regulatory requirements that need to be solved/scaled through and create softer requirements on the part of buyers, especially as robots interact in mixed environments with humans."

2. Contrarian Perspectives

Made-up financial metrics aren't inherently bad—but Anthropic's specific metric is uniquely absurd, not just controversial like past examples. Rechtman pushes back on the reflexive dismissal of "adjusted" metrics by noting that EBITDA itself was once seen as illegitimate but proved useful; the issue with Anthropic isn't novelty of metric, but the fact that it strips out costs core to the business.

"A few decades ago EBITDA was considered a ridiculous made up metric (how could you exclude interest and taxes!) but the point was that it showed the debt load a business could support... In the case of Anthropic, it's not clear what the metric is supposed to tell us other than 'IPO incoming.'"

AI safety advocacy should not be dismissed as insincere, but it also shouldn't be trusted until costly action is taken. Rather than fully cynical or fully credulous, Rechtman stakes a middle position that's contrarian relative to both AI-doomer and accelerationist camps.

"I think it's worth believing the people worried about AI safety, or at least hearing them out. But until they do anything that runs counter to their self-interest it's hard to trust their proposed remedies."

AI-cheating detection is creating a new, dangerous binary that doesn't reflect reality. Against the assumption that better AI-detection tools (like Pangram) are an unambiguous good, Rechtman argues the current framework risks unfair "witch hunts" against legitimate human-AI collaboration.

"Right now we don't have good models between 'human' and 'slop' and I think it's fair to worry about witchhunts. 'One drop counts' doesn't strike me as the optimal outcome, nor does giving the AI total credit for execution of human ideas/thinking."

3. Companies Identified

Anthropic — AI lab (maker of Claude), used as the central case study for the article's argument about safety rhetoric and margin claims. "Anthropic leaked some pseudo-financials that claimed they're profitable and high gross margin… if you exclude stock comp, training, and revenue share."
WeWork — Failed co-working real estate company, referenced as the historical analogue for questionable adjusted metrics. "Many have likened it to WeWork's 'community-adjusted EBITDA' but it's actually even more ridiculous than that."
Pangram — AI-generated text detection company, cited as a best-in-class tool facing adoption/trust challenges. "Regular readers will know I'm a big Pangram fan but with time it's clear that we really don't have a good norms around AI writing."
Slow Ventures — Rechtman's own venture firm, mentioned for context on his investment focus. "I'm a partner at Slow Ventures, where I lead pre/seed rounds from a ≈$325M fund."
Sam's (Court Street) — Restaurant, mentioned as a personal recommendation, not business-relevant. "Really fun dinner and definitely worth a visit. The veal parm was a big hit."

4. People Identified

Dario Amodei (referred to as "Dario") — CEO of Anthropic. Cited for his public call to slow AI development, which the author frames as strategically self-serving. "Dario made the claim that we need to 'pace the frontier' and collectively slow down AI development through a series of controls and agreements."
Adam Neumann — Former WeWork CEO. Referenced as a comparison point to argue Anthropic's metric obfuscation is worse than WeWork's. "If Adam Neumann had been a normal guy and just called it 'same store gross profits' or something like that no one would have blinked."
Yoni Rechtman — Author, partner at Slow Ventures. Self-described investment thesis and focus areas provided directly. "I'm a generalist investor looking for weird takes on important stories: N-of-1 companies taking non-obvious approaches to markets that matter."

5. Operating Insights

Treat any "adjusted" profitability metric from a capital-intensive AI company as a narrative device, not a fact—ask what story it's designed to tell. Rechtman's framework: "New/made up metrics are fine... when you invent a metric you do so to highlight a part of that story that is otherwise un- or undertold." Founders and investors should reverse-engineer the intent behind unconventional metrics before accepting them.
For physical AI/robotics founders: build the safety/compliance/insurance narrative into the product from day one, not as an afterthought, since buyer trust and regulatory acceptance—not raw capability—will gate deployment. "It's incumbent on technology leaders to make an affirmative, positive and humanist case for physical AI, robotics, automation etc."
Use lightweight personal AI agents to manage attention/notifications during high-focus or off-grid periods—a validated, simple use case rather than a moonshot application. "I built an email bot that read my email for me and texted me about anything important... This was a great little personal agent use case for me."

6. Overlooked Insights

Training costs are best modeled as cost of revenue, not R&D—a subtle but important reframe for anyone valuing AI labs, since it implies gross margins (not just operating margins) are structurally worse than reported. "Training is a cost of revenue, not an R&D cost. Stop training, everyone churns and moves on to your competitor or a distilled Chinese version... we're rapidly approaching >$1B training runs."
Every stakeholder in the AI safety/open-source debate is operating from motivated reasoning, which should discount the weight given to any single side's public position. "Every argument from every side (safety, open source, etc.) just sounds like motivated reasoning from partisans with a ton of stake. Because it is!"