AI IS EATING EUROPE: Insights from Lovable, Synthesia, Index, Granola and Uber CEO
What Europe’s top investors and unicorn founders are doing—and building—with AI right now
Last wednesday
hosted one of the most curated, high quality AI Summit I’ve ever attended to.On the my way to the event, I had the chance to meet some friends from Lakestar, Notion, Index and record a couple of podcasts (will be coming out soon🔥)
The event was filled with great talend, multiple top tier vcs and unicorn founders (I actually had never seen so many multibillion dollar startup founder together).
They shared their view on what to do with AI right now. This is it:
Hey, welcome to The AI Opportunity!
Every week you’ll get two things—fast:
What’s happening now (and where the money’s headed).
Who’s already building it (so you can invest, partner, or steal their playbook).
However, most of the content from this newsletter will be only accessible to the paid subscribers.
If you are serious about building or investing in AI, you can subscribe here:
If you want to sponsor The AI Opportunity or Product Market Fit email me to g@guillermoflor.com
1. Investing in 2030: Where Europe’s Top VCs Are Betting Next
Featuring:
Philippe Botteri (Accel), Tom Hulme (GV), Jan Hammer (Index)
1. AI is eating the old economy
Traditional industries like pharma, banking, manufacturing, and defense are finally waking up. What used to take $100M and years—like protein design—is now possible with a few million and AI. Winners will be the apps powering legacy sectors, not just flashy consumer tools.
2. Voice is the new interface
After keyboards and touchscreens, it’s voice—and even silent speech—that’s next. Think Neuralink-speed communication or stealth startups building pre-vocal AI. Your kids are already sending voice notes. Founders should build for this bandwidth.
3. Cybersecurity: both the fire and the extinguisher
AI is creating new security problems—deepfakes, fraud, synthetic IDs—but also solving them. Startups like Resistant AI and Sara are rising fast. Expect an AI arms race in defense tech and data protection.
4. The next wave: automating workflows
RPA 1.0 tackled repetitive tasks. RPA 2.0 (AI-native automation) will tackle workflows that require decision-making and context. This is where enterprise value will shift.
5. Composability & the software “mess”
The explosion of LLM-generated code has created technical debt. Now, reliability matters. VCs are backing new developer infra like Tessle to make AI-native codebases sustainable.
6. Is software screwed? Not quite.
Big companies won’t build everything themselves. Speed matters, and many will still buy or license from startups—especially in areas where talent or time-to-market are critical.
7. Why London (and not just London)?
London is still Europe’s operator HQ, and a gateway to global scale. But Paris, Munich, and others are catching up fast—especially in AI. LMU Munich birthed Stable Diffusion, Mistral is building world-class foundation models, and startups like Synthesia are global leaders.
8. What about exits?
Exits are rare, but they're happening. Meta’s $15B deal for ScaleAI shows a new playbook: strategic aqua-hires at mega-scale. IPOs are still paused due to volatility, but many companies are ready once markets stabilize.
9. Where are we in the cycle?
Prices may fluctuate, but the AI wave is real and secular. With capital available and model quality improving (especially with reasoning and post-training), there’s at least a decade of high-quality investing ahead.
2. The CEO of Granola on the future of AI products
Here’s what every founder building in AI should know:
Prompting will disappear
Typing out instructions is clunky. “Give AI five great examples, and it’ll do better than any prompt.”Voice in, text out
We're faster at talking than typing—and faster at reading than listening. The future interface? You speak, it writes.Context is king
AI isn’t dumb—it just doesn’t know you. “The smartest AI is the one that knows you best.”Simple wins (for most users)
“People aren’t power users. Most of the time, two buttons > twenty.” But advanced folks still want full control.The context wars are real
Slack is locking down data. AI agents are going lower-level—reading pixels, browser data, anything they can grab. Legal battles are coming.Today’s AI is the ‘beer app’ era of the iPhone
“AI isn’t that useful yet. Most products are caveman tools. The real stuff is coming.”Open > Closed
Startups that embrace openness and context will beat incumbents clinging to walled gardens.Docs are dead. Workflows win.
In 5 years, nobody will care about files. They’ll care about getting work done—fast, personalized, and AI-native
3. Anton, CEO of Lovable, on why his startup might be the last piece of software ever built
And how it went from $0 to $75M ARR in record time.
Here are 8 insights every founder and builder should hear:
Lovable wants to kill code
It’s not a website builder. It’s a product builder.
Users tell Lovable what to build—and it builds it.
“The future of software is not written. It’s talked into existence.”Software is becoming a conversation
In the future, 99% of software will still be “made by humans”—but through AI conversations, not code.
“You’ll describe it. The AI will handle the rest.”Product-market fit is now a weekend project
Tens of thousands of founders have built working MVPs, validated ideas, and made millions—directly inside Lovable.
Simple apps can go from idea to revenue without ever touching code.Complexity still needs engineers—but not forever
Today, advanced tools are still needed for very complex apps.
But every month, Lovable eats more of that stack.
“The last mile problem gets smaller every week.”Focus is overrated
Lovable didn’t niche down. It has hundreds of use cases—prototypes, side projects, internal tools, real startups.
Still, two groups shine:
• Founders starting from scratch
• BigCo product leaders prototyping faster than engineering ever could70,000 apps built per day
Not static sites—real products with payments, logins, analytics, etc.
Lovable isn’t just “no-code.” It’s full-stack, AI-native product creation.Distribution is still the hard part
Building is easy now. Getting users? Still hard.
Lovable is working on helping with:
• SEO-optimized launch pages
• Ads and channel recommendations
• Growth analytics out of the boxAGI might be our best shot at peace
Anton believes AI could help world leaders simulate outcomes, avoid conflict, and optimize for win-win decisions.
But only if we translate the right human values into the system.
4. Victor Riparbelli (CEO of Synthesia) and Matt Rouif (CEO of Photoroom)
AI is quietly taking over your camera roll, marketing team, and corporate training
AI Images Fool the Pros
A 30-year food photographer couldn’t tell AI images apart from real ones. Pasta did him in.If It Looks Too Perfect… It’s AI
Better lighting, shadows, and background removal? It’s probably AI. Seriously.PhotoRoom = Photoshop for Sellers
Used by marketplaces like DoorDash. Helps anyone create clean, trusted product photos.
Focus: Don’t lie with pixels. Just make it look amazing.Synthesia = PowerPoint → Video
72% of Fortune 100 uses it. Anyone in your company can make a video (no cameras, no actors).
Replaces boring docs with short, sharp training, marketing, and how-to videos.The Secret Sauce? Simplicity Wins.
Nobody wants to prompt 20 times to get the right video. These platforms make it one-click.AI Has Passed the Visual Turing Test
For static images and structured video, we’re already there. For dynamic storytelling? Not yet.You’ll Learn from AI Videos Daily
Mortgage explained by a video avatar? It’s coming in months. Not years.New Marketing = Personalized Visuals
Food images change depending on time of day, location, user mood. Like email personalization, but visual.They Train Their Own Models… But Not Always
Both founders say: the value is in the application, not the model. Use what works fastest.Reading Is the New Vinyl
Future kids might not read and write. They'll listen and watch. Corporate training will be TikTok + GPT
5. Lin Qiao (Fireworks AI) and Renen Hallak (VAST Data)
Lin Qiao (Fireworks AI)
She built PyTorch at Meta and saw the AI-first wave coming. Her insight? Most startups can’t scale because GenAI is too slow and expensive. Fireworks helps them optimize across speed, quality, and cost — all at once.
Renen Hallak (VAST Data)
Storage for AI used to be an afterthought. Now, it’s everything. Renen built a new kind of infrastructure for massive model training and inference. His bet? AI will need an “operating system” as essential as Windows was in the PC era.Why infra is broken
Old stacks weren’t built for AI. GenAI needs real-time performance, parallel GPU access, and fast feedback loops. Companies like X.ai, DoorDash, Perplexity, and Notion all build on Fireworks and VAST to make that happen.The 3D Optimization Framework
Speed, quality, cost — you can’t just pick one. Great infra delivers all three. The most successful AI-native companies optimize continuously across that triangle, not just during training but at every inference.Model ≠ API
Top apps treat the model as the product. They use user behavior to improve it — creating data flywheels. If you just plug into OpenAI and move on, you’ll get outrun. Customization is the new compounding advantage.What’s next: Reward Engineers
Forget prompt engineers. The next wave of AI teams will fine-tune models with reinforcement learning, writing logic to reward good outputs. It’s like teaching a kid with upvotes — and it works without massive datasets.Infra is eating the world
The best infra is invisible. Fireworks wants developers to plug in their data and let models auto-improve. VAST wants to secure and scale model memory in a world of thousands of agents. This is the new OS layer of AI
6. Uber x Wave: Self-Driving Comes to London
Dar Khosrowshahi (CEO of Uber) & Alex Kendall (CEO of Wave) on autonomy, partnerships, and why Europe has a shot at leading in physical AI.
1. Uber’s new self-driving strategy: don’t build, partner.
After shutting down its in-house AV effort, Uber now backs best-in-class partners like Wave.
Wave brings the tech. Uber brings global demand. Together, they plan to launch in London—soon.
2. The ride? Already here. And it works.
Dar and Alex were driven to the event by a Wave-powered car through central London—with zero interventions. Complex turns, pedestrians, and construction? No problem.
3. Wave’s bet: end-to-end AI beats HD maps.
Unlike AV 1.0 approaches that depend on hard-coded maps and geofenced zones, Wave’s model adapts to unseen roads and scenarios—like real drivers do.
4. Regulators are accelerating timelines.
UK regulators say deployment could begin in 2025. That clarity gives Wave the green light to scale—responsibly.
5. No spinning LIDARs. Just mass-manufacturable autonomy.
Wave is working with OEMs to integrate affordable sensor stacks (camera + radar + LIDAR) directly into vehicles—no retrofitted robots.
6. Cities want it. But safety rules everything.
Cities are leaning in. The value prop is real: safer streets, less congestion, lower emissions.
But Uber and Wave are clear—offline simulation, staged rollouts, and proving safety come first.
7. AVs will reduce traffic and road rage.
Wave’s cars don’t just get from A to B. They drive politely. No abrupt stops, no blocking intersections—just smooth, London-friendly navigation.
8. The AV race is platform vs product.
Waymo focuses on narrow deployments. Wave aims for scale through consumer-owned vehicles with driver-assist first, robo-taxis later.
Uber? It's the platform, aggregating global demand and distribution.
9. Europe can win physical AI.
Dar says the UK has elite AI talent and a historic chance to lead a platform shift. After decades of the U.S. owning the internet, mobile, and cloud—AI might be different.
10. Ten years out: more robots, still more humans (for now).
Uber predicts self-driving will power more of its rides over time—but demand for human drivers will grow before it shrinks.
Eventually? Humans and robots will split the work.
The AI Opportunity FAQs
📈 AI Investing Trends
1. What industries are VCs betting on in AI right now? VCs are heavily investing in AI applications for traditional industries such as pharma, manufacturing, defense, and financial services. These sectors are being disrupted by AI's ability to reduce cost and time for tasks like protein design, supply chain optimization, and fraud detection.
2. Why are traditional sectors like pharma and manufacturing attractive for AI startups? Because they have massive inefficiencies and legacy systems. AI can accelerate drug discovery, automate manufacturing QA, and streamline regulatory compliance. The returns are high because these industries spend billions and are open to innovation.
3. What’s the difference between consumer AI and enterprise AI plays? Consumer AI focuses on tools individuals use (like chatbots or image generators), while enterprise AI helps companies automate processes, make decisions, or improve infrastructure. Enterprise AI typically has higher ARPU and stickier contracts.
4. Which European cities are becoming AI hotspots besides London? Paris and Munich are leading the pack. Paris is home to Mistral, a top foundation model startup. Munich's LMU birthed Stable Diffusion. Both cities offer deep research talent and AI-first companies.
5. Why are Paris and Munich seen as rising AI hubs? Because they combine top-tier universities, AI-focused entrepreneurs, and strong government support. Mistral and Synthesia are European unicorns born from these ecosystems.
6. How are venture capital firms approaching exits in AI? They’re increasingly betting on large strategic acquisitions. Meta's $15B acquisition of Scale AI set a precedent. IPOs are paused but many companies are exit-ready once markets stabilize.
7. Are there recent examples of billion-dollar AI startup acquisitions? Yes, Meta's ~$15B acquisition of Scale AI is the biggest example. It's a sign of strategic interest in data and model infrastructure rather than consumer tools.
8. What stage are we in the AI investing cycle—bubble or beginning? We’re early in a secular shift. While valuations may fluctuate, model quality and capital availability suggest we’re still in the first inning.
9. What’s the long-term outlook for AI venture investing (2025–2035)? Expect a decade of quality deal flow. As AI models get better and infrastructure matures, new categories of companies will be created and existing ones reinvented.
10. What does Meta’s acquisition of Scale AI mean for future M&A? It shows big tech is willing to pay premium prices for key AI infrastructure. Expect more M&A in model training, data pipelines, and deployment tooling.
🤖 AI Products, UX & Interfaces
11. What does “voice is the new interface” mean in AI? It means we’re moving from typing and clicking to speaking and hearing. Voice input with AI can speed up tasks and make interactions more human-like.
12. Why is voice input and text output the ideal AI interface? Humans speak faster than they type and read faster than they listen. So a speak-in, text-out model combines efficiency and comprehension.
13. Will prompting become obsolete in AI tools? Yes. Prompting will be replaced by example-based learning and voice-first interactions. Users won’t need to know syntax—they’ll just talk or show.
14. How can AI understand users better through context? By integrating user behavior, preferences, calendar data, documents, and prior actions. The more context AI has, the better it can personalize outcomes.
15. What are “context wars” in AI and why do they matter? Platforms like Slack are limiting AI access to user data, while agents are scraping screens or reading pixels. Whoever owns context wins the accuracy war.
16. Why are documents becoming obsolete in AI workflows? Because users care about outcomes, not files. AI-native tools focus on completing tasks, not organizing static files.
17. What does it mean to be “AI-native” in product design? It means the product was designed from scratch to use AI as the core engine—not bolted on. Inputs, outputs, workflows are AI-first.
18. How should founders think about building with vs. without files? Focus on actions and tasks. Users want to accomplish things (e.g. submit report, create invoice), not manage documents.
19. Why do simple interfaces outperform complex ones in AI apps? Because most users aren't power users. Tools with fewer buttons, clearer UX, and guided flows win mass adoption.
20. Are AI agents starting to read pixels and browser data? Yes. Agents are bypassing locked APIs by interpreting screen layouts, DOM elements, and visual cues to get data and take actions.
🛡️ AI & Security
21. What cybersecurity risks are created by AI? Deepfakes, synthetic IDs, AI-powered phishing, and model hijacking are major risks. Attackers are using AI to scale threats.
22. How is AI being used to prevent fraud, deepfakes, and synthetic identities? By analyzing patterns, detecting anomalies, and validating real user signals. Startups like Resistant AI build models specifically for these defenses.
23. What are some AI startups working in cybersecurity? Resistant AI, Sara, and a growing wave of agent-based fraud detection tools are leading the charge.
24. Will there be an arms race between malicious AI and defensive AI? Yes—it's already underway. Just like antivirus vs. malware, this will be an ongoing cat-and-mouse dynamic.
⚙️ AI Infrastructure & Developer Tools
25. Why is AI infrastructure becoming so critical? Because legacy infra can’t handle real-time, high-throughput, model-intensive workloads. Speed, quality, and cost are all bottlenecked without better infra.
26. What’s wrong with old developer stacks for AI workloads? They weren’t built for parallel GPU processing, fast feedback loops, or distributed inference. That limits performance and scale.
27. What is the 3D optimization framework in AI (speed, quality, cost)? It’s the idea that infra must optimize for all three dimensions at once. Startups that do so get compounding advantages.
28. Why is model inference more important than just training? Because inference happens millions of times after deployment. If it’s slow or expensive, the product fails to scale.
29. Why are AI-native companies focusing on continuous optimization? Because workloads, traffic, and model performance shift. Continuous tuning ensures reliability and speed at scale.
30. What is Fireworks AI and what problem does it solve? It makes GenAI faster, cheaper, and more scalable by optimizing model serving infrastructure.
31. What does VAST Data do, and why is it crucial for GenAI? It rethinks storage for AI—offering high-performance, scalable data infrastructure for massive model memory and retrieval.
32. What’s the difference between plugging into OpenAI vs. building a model layer? Plugging into OpenAI gives fast results, but no edge. Owning the model and its data flywheel compounds learning and value.
33. What is “reward engineering” and how does it work in AI? It’s a form of fine-tuning using reinforcement learning where you define “good outcomes” and teach the model what to aim for.
34. Why is infra considered the new OS layer for AI? Because it’s the foundation all AI-native apps will be built on—controlling data, performance, and security.
35. What’s the role of real-time data and fast feedback loops in AI applications? They let apps learn and adapt quickly. Delays kill UX and make AI feel broken.
🧠 AI Product Creation & No-Code Tools
36. What is Lovable and how does it work? It’s an AI-native app builder. Users describe what they want, and Lovable creates full products with payments, analytics, and user flows.
37. Can you build a full product with AI without writing code? Yes. Platforms like Lovable allow voice-to-product creation with no technical input.
38. How is Lovable different from a website builder? It builds products, not just landing pages. You get databases, payments, logins, analytics—out of the box.
39. What kinds of apps are people building with Lovable? Side projects, MVPs, internal tools, paid apps—anything that doesn’t require deep engineering yet.
40. How fast can you get to product-market fit with Lovable? Some founders reach PMF in days. You can validate, iterate, and monetize without waiting on dev cycles.
41. Who are Lovable’s target users (founders vs. corporates)? Mostly indie founders and product managers in large companies who want to prototype faster.
42. How many apps are created per day using Lovable? Around 70,000. That includes live, functional apps.
43. What is “talking software into existence”? It means using natural language (speech or text) to describe what you want—and letting AI generate the code.
44. Will no-code platforms replace developers? Not immediately. But they’ll take over more of the stack every month, especially for simple or repetitive tasks.
45. Is building software now easier than distributing it? Yes. Distribution—finding and converting users—remains the biggest hurdle.
46. How does Lovable help with distribution, SEO, and user acquisition? It offers launch page builders, built-in SEO, ad templates, and analytics for growth.
🎥 AI for Visuals, Photos & Training
47. Can AI-generated images really fool professional photographers? Yes. Many pros can’t tell the difference—especially for food, products, and static scenes.
48. What is Photoroom and how is it used in e-commerce? It’s an AI tool that lets sellers create high-quality product images automatically. Used by marketplaces like DoorDash.
49. What makes Synthesia different from other video tools? It lets anyone create professional videos without actors or cameras. Used for internal training, onboarding, and marketing.
50. How are Fortune 100 companies using AI video in training and marketing? They’re replacing slide decks and PDFs with engaging, localized video content made by AI avatars.
51. What is the visual Turing test and has AI passed it? It’s the ability for AI-generated images to be indistinguishable from real ones. For many use cases, AI has passed it.
52. What is the future of corporate training with AI? It will look like TikTok meets GPT: short, smart, personalized video snippets that teach fast.
53. How are AI visuals being personalized based on user behavior? By adapting images to user data like time of day, location, or past preferences—like dynamic email but visual.
54. Do tools like Photoroom or Synthesia train their own models? Sometimes. But often, they prioritize product quality and speed over building foundational models.
55. What does “reading is the new vinyl” mean in the context of AI media? That future generations will consume content through audio and video, not text—just like vinyl became niche.
🚗 AI + Self-Driving Cars (Uber x Wave)
56. Why did Uber shut down its self-driving division? It was capital-intensive and risky. Uber now partners with top AV players like Wave to share the load and move faster.
57. What is Wave and how does its AV tech differ from Waymo or Cruise? Wave uses end-to-end AI to drive without HD maps. It learns like a human, adapting to new roads and situations.
58. How close are we to autonomous vehicles in Europe? The UK aims for commercial deployment by 2025. Testing is already underway in cities like London.
59. What sensors does Wave use for self-driving? A combination of cameras, radar, and low-cost LIDARs—designed to be production-ready for OEM integration.
60. Are regulators in the UK supportive of self-driving cars? Yes. They’ve accelerated approval timelines and are building legal frameworks for deployment.
61. What does “physical AI” mean? It refers to AI operating in the real world—robots, cars, drones—as opposed to software-only use cases.
62. Why does Uber think self-driving is a partnership play now? Because it can focus on demand and distribution while partners bring the tech. It’s more capital-efficient.
63. What are the benefits of self-driving cars for cities (e.g., safety, emissions)? Fewer accidents, reduced congestion, cleaner transport, and better traffic flow.
64. Will self-driving cars reduce road rage and improve flow? Yes. They drive predictably and calmly, reducing tension and risky behaviors on the road.
65. What’s the long-term future of self-driving: full autonomy or hybrid models? It will be hybrid for a while—AI handles more over time, but humans remain in the loop.
66. Is Europe capable of leading in the physical AI race? Yes. With top-tier talent, friendly regulators, and manufacturing depth, Europe is well-positioned.
🔮 Big-Picture AI Philosophy
67. Why do some founders think AGI might prevent wars? Because AGI could simulate outcomes and help leaders optimize for peaceful, win-win decisions.
68. How can reinforcement learning and simulation help world leaders make better decisions? By testing policy choices across thousands of scenarios and surfacing the most stable, ethical options.
69. Why does software become “a conversation” in the AI era? Because users can speak their intent and let the AI build. The interface becomes natural, not technical.
70. Will AI eliminate code entirely in the future? Possibly for many use cases. Code will remain for edge complexity, but much will be abstracted by voice or visual tools.
71. What human values should we embed in AGI systems? Transparency, fairness, empathy, and alignment with collective wellbeing—to ensure AGI benefits society.
72. What does it mean that “the future of software is not written, it’s spoken”? It means natural language replaces programming. You describe, and the AI builds—no code required.