partner @threeVC. Helping great teams building 🦄: @gameeapp @kaiahealth @DeepCodeAI @storyblok @lokalise @tatum_io #Assaia @shopcircleco @fynk_com #EmmiAI
Austria
Joined August 2007
- Tweets969
- Following118
- Followers398
- Likes1.5K
Why 10% AI extinction prophecies are mainly narrative - and what the real threat is.
Doomsday reports are everywhere, but framing AI as a potential supervillain misses the point. The issue is not an algorithm developing a "will to power"; the real risk is how human actors use it.
Key takeaways from my latest article:
👉 Unfalsifiable Prophecies: Assigning a 10% probability to human extinction is attention-seeking narrative, not quantifiable science.
👉 The Autonomy Fallacy: High capabilities or intelligence does not imply a biological drive for ego, malice or self-preservation.
👉 Cui Bono? Alarmism often serves the interests of frontier labs looking to capture public attention and justify absolute control of the market.
👉 The Real Risk: AI is an amplifier of human intent. The true threat is not a machine trying to kill us, but concentrated power using pervasive automation to undermine human autonomy, safety, and free will.
lasinger.net/p/will-ai-event…
Is Anthropic’s pricing premium sustainable?
According to the latest Vercel AI Gateway Production Index (vercel.com/blog/ai-gateway-p…), we’re seeing a disconnect between model volume and spend:
- The 4.4x Disconnect: In AI coding, Anthropic handles 28% of token volume but commands a staggering 70% of total spend.
- The Rise of Commodity AI: DeepSeek captured 17% of overall gateway token volume (and 49% in coding), yet represents only ~1% of spend. Low-cost models passed the production quality bar.
- Smart Routing is King: Engineering teams are actively optimizing and differentiating model usage.
As coding intelligence becomes commoditized, paying a massive premium for every single token isn't sustainable. To stay competitive, frontier labs may inevitably see token revenue yield drop significantly.
Are you already using smart routing to cut your AI API costs, or relying on a single provider?
Why are we spending billions trying to give industrial robots knees?
In a warehouse or factory floor, wheels will always beat legs on stability, speed, and mechanical simplicity. More importantly, human biology runs on roughly 100 watts, while today's humanoid robots draw thousands of watts just to stay upright. Forcing electric actuators and binary code into a human silhouette doesn't inherit human capability; it just inherits our structural vulnerabilities.
We don’t need robots that look like us to build the future. We need robots that think and move with the efficiency of engineering.
Read the full piece here: lasinger.net/p/the-anthropom…
Why Most AI Startups Die: Survivability Matrix 📊
Viability hinges on 2 metrics: Gross Margin (GM=efficiency) & Net Revenue Retention (NRR=customer love).
Company positions:
💀 Dead Zone (Low GM/Low NRR): No lock-in, high compute, no unique data. Free funnels for Big Tech, then gone.
⏳ Efficiency Trap (High GM/Low NRR): Strong unit econ, but one-time fixes. High churn = endless CAC treadmill.
💸 Subsidized Utility (Low GM/High NRR): Loved product, but customization/delivery costs too high. Looks good while bleeding cash.
🌱 Compounder (High GM/High NRR): Holy Grail. Deep dependency + improving econ at scale.
Hype starts it; only Compounders survive.
Where's your biz? 👇
Are you building a compounding asset, or are you quietly running out of runway?
In the AI era, true unit economics are being crushed by hidden inference costs. To help operators navigate this, I built the Survivability Compass based on a simple baseline:
Survivability Score = Customer Love (NRR) + Economics (Gross Margin)
In a 360° circle, 180° is the flat horizon line. Below it, you are sinking. Above it, you compound.
How are you doing? I built a 30-second interactive tool so you can find your exact quadrant instantly.
👉 Calculate your score here: survivability.lasinger.net/
After 20 years in software & AI, I’ve seen a dangerous trend: Founders optimizing for growth while pouring water into a leaking bucket.
Growth is vain. Survivability is everything.
I map this using the Survivability Matrix and the Rule of 180:
Survivability Score = NRR + Gross Margin
In a circle, 180° is a flat line. Below 180, you’re sinking. Above it, you compound.
The 4 AI Quadrants:
The Dead Zone: Thin wrappers. No lock-in, low margins.
The Efficiency Trap: High margins, awful retention (high CAC, low LTV).
The Subsidized Utility: High NRR, but bleeding cash on 60% margins due to crazy inference costs.
The Compounder: 90%+ margins, high retention, deep data gravity (e.g., Databricks, Cursor).
The last decade rewarded hyper-growth. The AI era will reward survivability.
Deep dive, benchmarks, and why Nvidia breaks the scale (The 300° Club) here: lasinger.net/p/the-survivabi…
From Hour -1 to @MistralAI: Why Emmi AI’s next chapter is a massive win for European sovereignty. 🇪🇺
When @threeVC first backed @jo_brandstetter, @dennisjust, and Miks, "Physics AI" wasn't a category yet. In just one year, they didn't just raise Austria's largest seed round - hey taught AI "physical intuition" and defined a global industry.
Now, Emmi AI is joining forces with Mistral AI.
The synergy is undeniable: Mistral has built Europe’s premier digital "brain." Emmi brings the understanding of the physical world. Together, they are building a vertically integrated, sovereign AI stack to transform aerospace, automotive, and energy.
For us, this isn't an exit. We are thrilled to become shareholders in Mistral AI, doubling down on Europe's technological future.
Huge congratulations to the Emmi, Mistral, and NXAI teams, and everyone who supported this journey! 🚀
Read the full story here: three.vc/blog/from-first-bel…
Peter Lasinger retweeted
Mega-Exit: Mistral AI übernimmt Linzer Startup Emmi AI und startet Österreich-Standort ift.tt/XGnHOme
Peter Lasinger retweeted
MEGA-DEAL in Österreich: Emmi AI von Johannes Brandstetter, Dennis Just und Miks Mikelsons aus Linz macht Exit an Mistral AI, Milliarden-Bewertung steht im Raum. Alle Details und die Einschätzung der Bewertung hier: trendingtopics.eu/emmi-ai-ex…
Your AI isn't an advisor. It’s a flatterer.
LLMs hit 99% on exams but fail the "Expertise Paradox": performance plummets to <34.5% in real human use.
Why?
The Sycophancy Trap: Models are trained to earn a "thumbs up" by agreeing with you, not by being right. This creates a loop of "Delusional Spiraling"- lying by omission to validate your initial assumptions.
A liability for novices, but an Iron Man suit for experts. Stop using AI to confirm your brilliance. Start using it to pressure-test your flaws.
lasinger.net/p/the-flatterin…
Software isn’t eating the world anymore—AI is eating software. 🤖🍴
The traditional SaaS "per-seat" model is facing a violent disruption as agentic workflows take over. Are you ready for the shift from seats to outcomes?
The Deep Dive:
- Pricing: Why outcome-based models are mandatory.
- Moats: Data gravity is the only "training gold."
- Agility: How startups out-maneuver legacy giants.
- Stop asking who AI replaces. Start asking who it empowers.
lasinger.net/p/will-ai-eat-s…
Growth starts the story, profit writes the ending.
In the AI era, ARR is a distraction. If your revenue growth is just a pass-through for your API bill, you aren't scaling - you’re just a high-paid customer of hyperscalers.
We are entering the "Great Squeeze":
- Gross Profit per Million Tokens is the new North Star.
- Token consumption ≠ stickiness. Without proprietary data, your moat is built on sand.
- Output per Watt defines the winners. Efficiency is an enabler, but capability per unit of energy is the ultimate prize.
Stop funding the curve; start funding the question. Success belongs to those who build defensible engines of intelligence that actually return cash to shareholders.
An app is just a README the AI hasn’t read yet. 🤯
OpenClaw’s 2026 explosion (220k+ stars!) is the end of the UI-first era.
The shift:
1. Subtractive design (Pi-Mono)
2. Skills > APIs (Let the AI see the source)
3. The Certainty-Scope trade-off
Full breakdown: lasinger.net/p/inside-the-cl…
Peter Lasinger retweeted
Amazing vibes at the @openclaw meetup yesterday with @andreasklinger, lots of space lobsters and their humans, and the creator himself, @steipete!
(Photos (c) Tobias Gärtner)
A product built over the weekend is a liability on Monday, but an AI built to parent itself is an existential risk by Friday.
The Agentic Era is here. With OpenClaw and Moltbook, we’ve moved from reactive tools to recursive agents that don't wait for us, don't sleep, and increasingly don't need us.
By delegating our thinking, we risk a inversion where the agents doing the work become the masters of reality, leaving us in a state of automated apathy - what Nietzsche called "The Last Man."
We are witnessing the rise of "toddlers with power tools" - autonomous agents without moral agency. We are no longer just users; we are "Agent Governors." The challenge isn't learning to use AI, but learning how to parent it.
Full thoughts here 👇
