My First YC Demo Day in 2025: Chaos, Genius, and Why This Batch Feels Like the Future

I've been knee-deep in the startup world for years—founding companies, advising founders, crunching data science models to spot patterns before they become trends. But nothing quite prepared me for my first Y Combinator Demo Day in 2025. Held at their sleek new offices in San Francisco's Dogpatch neighborhood, it was a masterclass in production value: seamless tech, high energy, and an atmosphere that screamed "this is where the future gets built." If you're wondering why YC continues to dominate, let me break it down from an insider's view—complete with the chaos, the connections, and the signals that have me more excited than I've been in years.

The Setup: More Than Just Pitches

Picture this: a full day of back-to-back presentations, each founder getting exactly one minute and one slide to pitch their "life's" work. Breaks were strategically timed for networking, with food trucks parked for breakfast and lunch—think gourmet tacos and craft coffee fuelling deal talks under the California sun. The energy was electric, a mix of nervous excitement from founders and calculated intensity from investors. YC's Dogpatch HQ felt like a tech temple: modern, spacious, and designed for serendipitous collisions. No wonder they pulled off a production this polished; it wasn't just an event, it was an ecosystem accelerator on steroids.

Shoutout to the unsung heroes—the technical support team behind the Demo Day investor portal. When a few of us hit demo day account access glitches (hey, even the best tech has hiccups), they resolved them in real-time, mid-event. That's the kind of operational excellence that keeps the machine running smoothly.

The Founders: Youth, Tech, and Raw Ambition

What struck me most were the founders themselves. This batch skewed young—seriously young. I'd estimate 15-20% were college dropouts or fresh grads who'd already racked up impressive feats since high school: building apps that scaled to millions, hacking together AI prototypes in dorm rooms, publishing AI research papers or launching side hustles that caught YC's eye. At most, I spotted three with MBAs; the rest were deeply technical, often with backgrounds in engineering, CS, or data science. These aren't polished executives—they're builders who code first and pitch second.

Take the companies from the S25 graduates list I reviewed: teams like BootLoop (firmware in minutes via AI) or Janet AI (an AI-native Jira alternative) exemplify this. Founders aren't waiting for permission; they're leveraging AI to solve real problems in dev tools and beyond. It's refreshing—no corporate fluff, just raw innovation from people who grew up with GitHub as their playground.

One subtle tell? Look at their company URLs. Domains like phases.ai, getlilac.com, or bootloop.ai aren't flashy .coms or clever wordplays—they're straightforward, often incorporating "AI" directly. It screams product focus over marketing polish. These founders prioritize building something that works over snagging the perfect brand name or premium URL. In an era where AI lets you prototype in days, that no-nonsense approach is a competitive edge.

The Room: Celebrities, Athletes, and Serious Money

The investor crowd was a who's who of tech and beyond. I spotted comedian Hannibal Buress scanning pitches with a notebook in hand, and Ty Montgomery—the former NFL football player and Stanford alum—deep in conversation about deals. It wasn't just VCs; it was a melting pot of cultural influencers, athletes, and operators who see startups as the next big bet. That diversity amps up the energy—suddenly, a quick chat over lunch could lead to a celebrity endorsement or a strategic partnership.

For me, the real magic was in the outreach. Out of the 150+ pitches, I reached out to 20-30% to express interest in chatting further. That's by far my highest engagement rate in years. Why? The format forces clarity: one minute, one slide strips away the noise, letting the core idea shine. Many pitches hooked me instantly—they're not just building products; they're redefining categories with AI at the core.

Why YC Is Skyrocketing—and What It Means for the Rest of Us

Personally, I think YC is about to pull even further ahead. Their network is unmatched: a global web of alumni, investors, and experts that feels like they're living in the future. What I do with machine learning—spotting patterns in markets, predicting startup successes—they're embodying it daily through relentless iteration and founder support. This batch's AI dominance (90% of pitches were AI-centric, per reports) isn't hype; it's a signal. We're seeing agentic systems, voice AI, and edge computing solve real problems in healthcare, fintech, and dev tools.

Three quick provocations for anyone in the startup game:

Embrace the Youth Wave: If 20% of top founders are fresh out of school, rethink your hiring. Technical depth plus unjaded ambition is the new superpower.

Network Like It's Lunch: Events like this prove serendipity scales. Skip the formal meetings; real deals happen over food trucks.

Bet on Clarity: In a world of infinite prototypes, the winners distill complexity into one slide. That's the AI-era moat.

Demo Day Summer 2025 wasn't just an event—it was a glimpse into tomorrow. If this is YC's trajectory, count me in for the ride. 

Quick Summary:

Executive Overview:

🤖 AI Dominance: Unprecedented Concentration

Key Finding: 92.2% of companies are AI-focused
  • 127 companies (83.0%) explicitly in "Machine Learning / AI" category
  • 141 companies (92.2%) mention AI in descriptions or categories
  • This is the highest AI concentration in YC history
Top AI Use Cases:
  1. AI Agents - 30 companies (21.3%) - Autonomous software workers
  1. AI Automation - 12 companies (8.5%) - Workflow automation
  1. Voice AI - 9 companies (6.4%) - Voice assistants
  1. AI Analytics - 9 companies (6.4%) - Data analysis tools

🏢 Business Model Analysis

B2B Enterprise Focus: 74.5%
  • 114 companies target business customers
  • 84 companies (54.9%) are SaaS businesses
  • 150 companies (98.0%) target global US/World markets

🎯 Key Trends Identified

1. The AI Agent Revolution
  • 30 companies building autonomous AI workers
  • Replacing human tasks across industries
  • From customer service to software development
2. Developer Tools Renaissance
  • 33 companies (21.6%) building dev tools
  • "Cursor for X" pattern emerging
  • AI-powered coding assistance everywhere
3. Vertical AI Solutions
  • Healthcare: AI EMRs, clinical trials, pharmacy
  • Finance: AI accounting, billing, analysis
  • Real Estate: AI property management
  • Manufacturing: AI operations
4. Enterprise Software Transformation
  • Traditional tools rebuilt AI-native
  • Focus on knowledge work automation
  • Workflow automation emphasis


Andrew Ng Just Solved the Wrong Problem (And Why That's Actually Perfect)

A fresh conversation on the No Priors podcast has Andrew Ng declaring product management the new bottleneck in AI startups. His core thesis: while engineering teams can now build prototypes "10x faster" thanks to AI-assisted coding, product teams are still operating with pre-AI workflows, creating a dangerous mismatch that's "really painful" for founders.

The surface-level take: Andrew is simply highlighting an operational inefficiency that smart teams will optimize away.

The deeper read: He's accidentally identified the next phase of startup competitive advantage in the AI era—and most people are going to get it completely wrong.

1. The Speed Trap Everyone's Walking Into

Andrew's math is seductive but incomplete. Yes, prototypes that once required "six engineers three months" can now be weekend projects. But the obsession with prototype velocity misses a more fundamental shift: the economics of validation have changed, not just the mechanics of building.

When your prototype-to-feedback cycle compresses from weeks to days, you don't just need faster product decisions—you need fundamentally different validation strategies. The Valley's current playbook (build → measure → learn → iterate) assumes validation is the scarce resource. But what happens when building becomes nearly free and validation remains expensive?

Singapore's startup ecosystem offers a useful parallel. During our early 2010s acceleration phase, government grants and accelerator programs suddenly made seed capital abundant. Teams that optimized purely for fundraising speed got crushed by those who built systematic approaches to customer validation and market-product fit. Speed without direction is just expensive wandering.

2. The Product Manager Arms Race Nobody Asked For

Andrew floated a fascinating data point: some teams now propose PM-to-engineer ratios of 1:0.5—more product managers than engineers. This sounds like classic Silicon Valley logic: identify the bottleneck, throw resources at it, declare victory.

Except we've seen this movie before. Remember when "growth hacking" was the bottleneck? Or when "data science" was going to unlock everything? The pattern is predictable:

  • Phase 1: Shortage creates premium roles

  • Phase 2: Market floods with mediocre practitioners

  • Phase 3: Only exceptional talent differentiates

  • Phase 4: Back to fundamentals

The PM arms race will follow this exact trajectory. By 2027, every AI startup will have hired multiple "customer empathy experts" and "rapid product decision specialists." The real alpha will belong to teams that solved the underlying problem rather than optimizing for the symptom.

3. What Andrew Actually Discovered (But Didn't Say)

The most revealing quote wasn't about PM ratios or weekend prototypes. It was this: teams are "increasingly relying on gut" to make faster decisions.

That's not a bug. That's the feature.

In markets moving at AI speed, systematic data collection often arrives too late to matter. The teams winning in hyper-competitive super-app landscape aren't the ones with the best analytics—they're the ones whose founders have developed the most accurate intuitive models of their markets.

Andrew mentions this obliquely when he talks about PMs needing "deep customer empathy" and the ability to "synthesize lots of signals". But he's underselling the insight. What he's describing isn't product management—it's market sensing as a competitive moat.

Three Quick Provocations for AI-Era Founders

Here is how this trend actually plays out:

Customer Intimacy > Customer Research

The speed differential between building and validating creates systematic pressure to develop intuitive market models. Teams that can rapidly synthesize weak signals (user behavior, competitor moves, ecosystem shifts) will consistently out-maneuver those dependent on formal research cycles.

Tactical shift: Instead of hiring more PMs, invest in customer advisory relationships and systematic founder-market exposure. YCombinator's "get out of the building" remains true—but now it needs to happen at AI speed.

Prototype Portfolio Management > Single Product Iteration

When building costs approach zero, the optimal strategy shifts from perfecting one solution to systematically exploring solution spaces. This requires different metrics, different team structures, and different capital allocation frameworks.

Policy implication: Government innovation programs designed around linear "TRL progression" will systematically miss this shift. Singapore's NRF should consider evergreen exploration grants that reward systematic market sensing rather than just technical milestones.

Network Effects in Validation Speed

The teams with the best customer access networks will compound their advantage as build-validate cycles accelerate. This suggests ecosystem strategies matter more, not less, in the AI era.

Closing Thought

Andrew Ng identified a real problem. The issue isn't that product management needs to speed up to match engineering—it's that traditional product management becomes less relevant when the cost structure of innovation fundamentally changes.

The winning teams won't hire more PMs. They'll develop systematic approaches to market sensing that operate at AI speed. They'll treat customer intimacy as infrastructure, not a departmental function.

The real bottleneck isn't product management. It's developing market judgment that operates at the speed of artificial intelligence.

Why Venture Studios Rarely Deliver as Fund Returners – Lessons from the VC Trenches

Lately, I've been fielding questions about venture studios – those ambitious setups that aim to manufacture startups at scale. On paper, they look revolutionary: higher success rates, juicier IRRs, and a systematic approach to innovation. But as someone who's evaluated countless models for our portfolio, I can tell you the reality often falls short, especially when viewing them as an asset class for serious fund returns. Today, let's unpack the evidence why most don't work, and the rare formula that makes some succeed. This isn't armchair theory; it's grounded in data, case studies, and insights from my own deal flow.

The Seductive Promise – And Why It Often Crumbles

Venture studios burst onto the scene promising to fix VC's biggest pains: inconsistent dealflow, high failure rates, and inefficient capital deployment. The stats are eye-catching – studio-backed companies reach Series A 72% of the time versus 42% for traditional startups, with IRRs hitting 53% compared to 21.3% norms. They shave timelines too, getting to Series A in about 25 months instead of 56. I've seen pitches where studios position themselves as "startup factories," and honestly, who wouldn't want that in their portfolio?

But here's the rub: most flame out within 24 months. That's not random bad luck; it's baked into the model. In my experience reviewing studio proposals, the hype masks deep structural flaws that make them a risky bet for LPs seeking scalable, reliable returns. Let's break it down with the evidence.

The Structural Cracks That Sink Most Studios

From stakeholder wars to capital headaches, these issues aren't edge cases – they're systemic.

The Impossible Balancing Act of Four Stakeholders

Traditional VCs have it simple: LPs provide capital, GPs pick winners, founders execute. Studios? They must satisfy entrepreneurs (who want autonomy and upside), studio staff (needing comp and growth), follow-on investors (demanding clean terms), and LPs (chasing returns). It's a recipe for conflict. I've passed on studios where entrepreneurs felt like cogs in a machine, leading to talent flight and diluted innovation. Data shows this tension torpedoes governance and decision-making.

Resource Dilution: Great in Theory, Impossible at Scale

Studios tout "hands-on support" as their secret sauce, but the numbers don't add up. Launch one company a week with 100 staff? That's barely two full-timers per venture – hardly the deep operational help promised. Scale up, and quality tanks; stay small, and you can't deploy enough capital for fund-level returns. I've seen this play out in SEA studios that overextend, ending up as glorified consultancies rather than return generators.

Fundraising Nightmares and Wonky Capital Structures

Evidence from 2020-2024 reveals traditional VCs are 1.6x more likely to close funds than studios. Why? Complex valuations create the "valuation trap," higher fees scare LPs, and perceptions of "dead equity" deter later rounds. Add longer timelines and regulatory ambiguities (fund or operating company?), and you're burning cash before deploying it. In my due diligence, this often signals execution risk – markets shift, and the studio's left holding the bag.

Why Institutions Steer Clear

For big allocators like pensions or endowments, studios are a tough fit. Most cap at sub-$200M funds, too small for meaningful commitments, and their sector-specific focus creates concentration risks. Metrics suffer from survivorship bias too – we only hear about winners, not the ideation failures. Corporate studios? Even worse, often becoming "innovation theater" bogged down by parent company bureaucracy and mismatched incentives.

The Winning Formula: What It Takes to Make Studios Work

Not every studio crashes and burns. From my vantage point, the successes – think specialized operators in niches like fintech or healthtech – share a rigorous playbook. Here's the evidence-backed essentials, scored by their impact based on research and my observations:

  • Strategic Focus & Specialization (9.2/10): Ditch the generalist approach; vertical-agnostic studios succeed only 19% of the time. Winners leverage proprietary edges like data access or industry networks – crucial in emerging markets like SEA.

  • Operational Excellence & Proven Playbooks (9.0/10): Codified processes for everything from ideation to scaling. This includes stage-gates that validate ideas early, reducing waste and accelerating paths to revenue.

  • Significant Ownership Stakes (8.8/10): Holding 30-50% equity justifies the ops investment and captures value from solid exits, not just unicorns.

  • Experienced Team & Leadership (8.7/10): A mix of serial entrepreneurs, functional experts, and investors. Weak teams are a red flag in my evals – you need proven builders to navigate the chaos.

  • Quality Control Systems (8.6/10): Strict validation thresholds for market fit, tech viability, and business models. This kills duds fast and preserves capital.

  • Proper Governance & Alignment (8.5/10): Dual structures separating ops from investing, with clear equity rules to align all stakeholders.

  • Adequate Capitalization (8.3/10): Patient, reserved funding – balance sheet style, not skimping on ops budgets.

  • Market-First Approach (7.9/10): Prioritize customer validation over building; it's the best defense against vanity projects.

Implement all eight? You've got a shot at outsized returns. Skimp on any, and you're statistically doomed.

Wrapping Up: A VC's Practical Advice

Venture studios aren't the VC extinction event some predict – they're a niche tool, better for hands-on operators than broad institutional plays. The evidence shows their flaws often outweigh the upsides for fund returners, but in the right hands, they can be game-changers, especially in underserved regions.

If you're building or backing one, stress-test against these factors. 



    MIT's AI Reality Check: Why 95% of Pilots Are Failing (And What It Means for Startups)

    Just back from a whirlwind of meetings in Singapore, and I couldn't ignore this MIT report that's been blowing up my feed. As someone who's invested in dozens of AI startups through Golden Gate Ventures, I've seen the hype cycle firsthand. But this study? It's a gut check for anyone betting big on AI. Let's break it down, startup-style—because if you're building or funding in this space, these insights could save you millions.

    The Headline That Spooked Everyone (And Why It Shouldn't)

    The MIT NANDA report dropped like a bomb: 95% of AI pilot projects fail to deliver any real financial uplift. Yeah, you read that right. They looked at 300 projects, chatted with 150 execs, and surveyed 350 employees. The result? Most AI initiatives are burning cash without moving the needle on profits.

    Investors freaked out—Nvidia, Microsoft, and others took a hit in the markets. But hold up: this isn't Altman saying public AI stocks are bubbly (though he did call out private startups). And it's not an indictment of the tech itself. As the report points out, the real issue is how companies are using AI, not the AI models themselves.

    Digging Into the Data: It's Not the Tech, It's You

    NANDA—short for Networked Agents and Decentralized AI—is an MIT project pushing for better AI architectures. Full disclosure: they might have skin in the game, promoting agentic systems as the fix. But their findings align with what I've seen in the field.

    Key takeaways from the report:

    • Failure Isn't About Capability: Execs blame weak models, but the data shows it's a "learning gap." Organizations don't know how to embed AI into workflows. Wharton prof Ethan Mollick nails it: stop forcing AI into old processes shaped by bureaucracy. Let it redefine how work gets done.

    • Startups vs. Corporates: New companies crush it because they lack entrenched systems. If you're a startup founder, this is your edge—build AI-native from day one.

    • Buy > Build: Vendor solutions succeed 67% of the time; internal builds? Only 33%. I've advised portfolio companies on this: unless you're in a hyper-regulated space, don't reinvent the wheel. Focus on your core IP.

    • Wrong Focus Areas: Too many pour money into marketing/sales AI. The real ROI? Back-end automation that cuts costs. Think ops efficiency over flashy demos.

    This echoes other studies—Capgemini saw 88% of pilots flop in 2023, S&P Global noted 42% abandoned this year. It's not new, but it's getting worse as hype outpaces execution.

    Lessons from the Trenches: What Winners Are Doing

    The pattern? Smart integration and realistic goals. Don't treat AI like a magic wand—it's a tool that needs the right setup.

    From the report and my experience:

    • Workflow Redesign is Key: Experiment relentlessly. One of our portfolio companies pivoted from generic chatbots to agentic systems that automate entire processes—ROI jumped 3x.

    • Data Privacy Isn't an Excuse: Regulated industries hide behind "build internal" for control, but vendors often handle this better. Pick partners wisely.

    • Measure What Matters: Track financial savings, not just "AI usage." The report slams vague metrics—get specific on P&L impact.

    Oh, and shoutout to Ethan Mollick again: AI shines when you let it bypass office politics. Startups, this is your superpower.

    The Bigger Picture: Bubble or Breakthrough?

    Look, investor panic is real—shares tanked on headlines alone. But this report isn't doom and gloom. It's a wake-up call that AI's impact is coming, just not how most expect. We're in the trough of disillusionment (Gartner hype cycle, anyone?), but the slope of enlightenment follows.

    For founders: Focus on agentic AI that scales autonomously. NANDA's pushing this, and it aligns with what DeepSeek's doing in China—efficient models that compete with OpenAI at a fraction of the cost.

    For investors: Don't bail yet. The trillions in data center spend Altman predicts? It's happening, but winners will be those solving real problems, not chasing buzz.

    Wrapping It Up: Your AI Playbook

    If you're building an AI startup, heed this: 95% failure rate is a feature, not a bug—it's your opportunity to be the 5%. Nail integration, buy smart, automate the boring stuff, and measure ruthlessly.

    The AI revolution isn't slowing—it's evolving. China restricting Nvidia sales? That's just accelerating local innovation. Google's Pixel AI features? Table stakes now.

    Stay sharp, folks. If you're pitching AI to VCs like me, show how you'll beat these odds.