Levelsio built a flight simulator in 3 hours with Cursor, hit $67K MRR in 13 days, and the whole thing runs on vanilla HTML and JS. That's not a typo. Three hours to build. Thirteen days to $67K monthly recurring. Here's what happened: Pieter asked Cursor to build a 3D flying game in the browser. He'd never made a game before. The AI wrote the code. He iterated by just telling it what he wanted changed. After a few hours, he had a playable flight sim. He deployed it. People started playing. Then he added in-game ads. Blimps, clouds, runway billboards. Sponsors paid $2K-5K each to get their brand in the game. 15,000 people flying daily. Revenue compounding. This is the vibe coding movement in action. But here's the question I'm wrestling with: where does this model actually work for real businesses? Games are forgiving. Bugs become features. Players expect jank. If the physics glitch, it's funny. Fuse can't run property management on vibes. If the booking system breaks or the payment flow has a race condition, we lose customers and revenue. The error tolerance is zero. So where does AI coding velocity matter for operators like me? Internal tools: We've built three internal dashboards with Claude Code in the past month. Occupancy tracking, pricing simulators, maintenance ticketing. Stuff that would've taken a dev shop 6 weeks and €15K. We shipped it in 2 days for free. MVPs for validation: Before we build a feature into the core platform, we spin up a quick version with AI to test with 10 users. If it doesn't work, we kill it. No sunk cost. Side revenue experiments: This is the Pieter model. Build small tools, monetize fast, see what sticks. I've got two side projects running right now that took under 10 hours combined to build. The two-jobs reality is real. Your day job (Fuse for me) funds the business. Your AI side job builds the next one. If you're not shipping something with AI this month, you're leaving speed on the table.

I've been talking to founders who raised in 2025, and the ones who are still standing in May 2026 share one trait: they protected runway like it was oxygen. 2025 was a fundraising year for a lot of us. 2026 is the execution test. Here's the pattern I'm seeing: Founders who raised 12-18 months ago are now at the first real checkpoint. They're 6-9 months post-raise. The initial excitement is gone. The burn rate reality is here. The ones who are struggling made three mistakes: 1: They hired too fast. Brought on senior people with impressive titles before the business had repeatable revenue. Now they're paying €120K salaries for roles that aren't yet defined. 2: They burned cash on partnerships that looked good on LinkedIn but delivered zero ROI. Logo deals, co-marketing, "strategic alliances" that never converted to customers. 3: They confused momentum with progress. Lots of meetings, lots of activity, but no clear milestones hit. The ones who are thriving did this differently: Runway discipline: They managed to 18 months minimum, even if it meant saying no to hires that felt urgent. Investors in 2026 expect this. It's table stakes now. Outcome-based hiring: They didn't hire for titles. They hired for specific outcomes. "We need someone who can close 10 enterprise deals in Q2" beats "We need a VP of Sales." Weekly burn reviews: They tracked burn weekly, not monthly. Small adjustments compound. Waiting until the monthly close to realize you're 20% over budget is too late. The meta-lesson: post-raise discipline is a leading indicator of Series A readiness. If you can't manage a seed round with capital efficiency, no one believes you can manage a Series A. Fuse is fundraising summer 2026. I'm applying this now. Happy to walk through how we're thinking about it if useful.

The Spanish flex-living market is expected to double to 20,000 beds by 2025, while most CEE operators are still figuring out their playbooks. I'm watching this closely because Fuse operates in both regions. Here's what the data is showing: Spain went from 2,000 beds in 2020 to 8,000 today. By end of 2025, nearly 20,000 beds. That's 10x growth in five years. €240M flowed into flex-living in H1 2023 alone. 69% went to corporate living (think outskirts, larger footprints, amenities). 31% to coliving (urban, smaller units, community-first). Madrid and Barcelona are 75% of the operational stock. Malaga, Valencia, Bilbao coming online fast. CEE is roughly where Spain was in 2020. Riga, Budapest, Vienna have maybe 3,000-4,000 beds combined across serious operators. Warsaw, Prague, Krakow are launching 2026. But CEE has advantages Spain didn't have in 2020: 1: Infrastructure for digital nomads already exists (thanks to the past five years of remote work) 2: Entry costs are 40-50% lower than Western Europe 3: EU mobility means professionals rotate through CEE for 6-12 month stints without visa friction The strategic question I'm wrestling with: do we replicate the Spanish playbook (chase institutional capital, go big on corporate living, amenities-heavy) or build something native to the CEE context (lighter footprint, faster unit turns, community over amenities)? I don't think the answer is obvious yet. Spain's model works because the tenant is often a corporation paying the lease. CEE's tenant is more likely a freelancer or startup employee paying out of pocket. Different customer, different unit economics, different playbook. If you're operating in flex-living or watching this space, curious what you're seeing. DM is open.

Gartner just said 40% of enterprise apps will have AI agents by end of 2026. The infrastructure isn't ready. The stat: Gartner predicts 40% of enterprise applications will feature task-specific AI agents by December 2026. Up from less than 5% in 2025. Multi-agent system inquiries surged 1,445% between Q1 2024 and Q2 2025. The problem: most companies will ship agents that work in demos and break in production. Why? Because they're treating agents as a UX problem. It's an infrastructure problem. Aditya Naganath nailed it in his 2026 predictions thread: "We haven't seen broad adoption of agentic products not because models aren't capable, but because reliable agents require sophisticated infrastructure. Teams that break through will build infra that enforces determinism, efficient course correction, and performance over long horizons." What that looks like in practice: 1: Observability. You need to see what the agent did, why it chose that path, and where it failed. Most teams bolt an LLM onto existing dashboards and wonder why it's opaque. 2: Rollback and eval discipline. Agents aren't deterministic. You need simulation environments (digital twins of prod) where the agent can fail safely and learn from mistakes. 3: Don't retrofit agents into legacy SaaS UX. Redesign the workflow. The best AI UI isn't a chatbot. It's a generative interface that appears only when a decision is needed and runs silently the rest of the time. The 4-layer pattern that scales: orchestration (which agent does what), execution (agents that write/review/debug), review gates (human-in-loop for destructive tasks), and memory/context (shared knowledge graph across agents and humans). The gap between "cool demo" and "reliable in prod" is infrastructure. The companies that figure this out in 2026 will have a two-year lead by 2028. DM's open if you're building agent infra and want to compare notes.

European VC in Q1 2026: billion-dollar rounds are back. Early-stage founders are still starving. The headlines: Nscale raised €1.7B (largest European equity round ever). AMI Labs €1B. Seaya closed a €1B growth fund. KPMG called it "a solid start." The fine print: deal count fell. Only 1,034 rounds closed in Q1. Investors wrote fewer checks, backed clearer winners, put more money into companies that already had traction and a path to scale. Translation: if you're AI infrastructure or European defense tech, the window is wide open. If you're seed-stage anything else, you're pitching 40 funds for a €1M round. What this looks like in practice (Fuse is fundraising summer 2026, so I'm living this): Tier 1 VCs want proof of customer pull before Series A. Not "we have revenue." Proof that a government buyer or enterprise signed a multi-year contract. The KPMG report is explicit: scale, profitability, defensible market position are table stakes. Traction alone isn't enough anymore. The CEE-specific tax: if you're not in London, Paris, or Berlin, add three months to every fundraising timeline. Fewer funds. Longer travel. More "we love it but it's outside our geographic focus." Eastern Europe raised €3.6B in 2025, just 5.5% of European VC. We're one-third of Europe's population. Do the math. The conclusion isn't "don't raise in Europe." It's design for capital scarcity. Revenue earlier. Longer runway. Sharper story. The capital exists. Access doesn't. Fuse operates across Riga, Budapest, Vienna. Prague, Krakow, Warsaw launching 2026. We're building for this reality. DM if you're fundraising in CEE right now. Happy to compare notes.

Cursor deleted a startup's production database last week. We're still shipping with it. Friday 27 April. PocketOS. Cursor running Claude Opus 4.6 hit a Railway API, deleted the production database and all volume-level backups. Nine seconds. The founder's take? "The velocity at which you can create good code is unparalleled. Benefits outweigh the risks." I agree. But only if you're not stupid about it. What happened: The agent found a credential mismatch in staging. Decided to fix it by deleting a Railway volume. Went hunting for an API token, found one scoped for *any* operation. Called a legacy endpoint that didn't have delayed-delete protection. Human mistakes at three layers. The agent was just fast. What we do at Fuse: 1: No production API tokens in the repo. Ever. If an agent needs write access, it's in a staging clone or a sandboxed Railway project. 2: Destructive tasks (database migrations, infra changes) run human-in-loop. Cursor's inline flow is brilliant for feature work, Claude Code for refactors. Neither gets root without a review gate. 3: The two-job framing still holds. Day job running Fuse, secret AI building job at night. But the infra discipline is table-stakes now. Agents are production tools. Railway patched the endpoint within 48 hours. Cursor already had safeguards from a similar incident nine months ago. They didn't apply here. The takeaway isn't "agents are dangerous." It's "agents are junior engineers with root access." You wouldn't give a new hire unrestricted prod tokens. Don't give them to Claude either. Still shipping. Still fast. Just not reckless. DM's open if you want to compare notes on agent infra setups.