NewsFlash Articles Data Fundraising Skill&API

YC W26 Demo Day Deep Dive: The Entrepreneurial Truth of 200 Startups, Copilot is Dead, AI Agent Takes Over

Read this article in 34 Minutes

Data, Patterns, and Everything a Future Founder Needs to Know.

Original Title: What I Learnt From 199 Pitches at the YC W26 Demo Day
Original Author: Rathin Shah, Ex-founder of Spenny
Translated by: Deep Tide TechFlow

Deep Tide Summary: This is not a simple Demo Day observation report. After listening to 199 pitches on-site, the author used data and examples to reveal the underlying logic of current AI entrepreneurship: why 60% of companies are all in AI, why the copilot concept has nearly disappeared, and why the fastest to revenue are founders who “sell back to their ex-employers.”

More importantly, he pointed out the fatal risks behind seemingly popular tracks and the overlooked but potentially legendary blank spaces that everyone ignores.

I attended YC's 2026 Winter Demo Day. 199 companies. Here are all my observations: data, patterns, and everything you need to know if you are a future founder.

Key Lessons for Founders

About Market/Problem Statement

1. AI is not a category; it is infrastructure. 60% of the batch is AI-native. Another 26% are AI-empowered. Only 14% do not involve AI. The question is not “Are you using AI?” but “What can your AI do out of the box that the basic model cannot?”

2. Replacement, not augmentation. The core theme is the “AI employee,” not a copilot, not an assistant. The pitch is always “we end-to-end replace [an expensive human role],” with pricing being a fraction of that person's salary. Copilots are augmentations. Agents are actions. The industry has moved forward.

3. Find your field's “Claude Code.” Every profession has structured outputs that AI can now generate: contracts, CAD files, financial models, surgical plans, specifications. Find a profession with an hourly wage of $100-500+, tools with a history of 10-30 years, and clear validation steps. Broad areas include tax planning, civil engineering, management consulting, clinical trials, patent drafting, music production.

4. Consider a service model. Around 20% of the batch is establishing AI-native service companies (legal, recruitment, accounting, insurance), charging based on outcomes but enjoying software profit margins. They showcased the fastest revenue growth in the batch. The pattern is: start with services → gain revenue and data → deploy automation → evolve into a platform.

5. B2B Dominance. AI agents replacing B2B knowledge workers. 87% are B2B. Only 14 consumer-facing companies (about 7%). Current AI capabilities unlock perfect alignment with business workflows. It's a good deal, but the legendary companies in this batch are most likely outliers: a uranium exploration company, a lunar hotel, a robot cowboy, a parasitic drug company.

6. Build a Data Flywheel. Every customer interaction should make your product better. LegalOS trained on 12,000 visa applications → 100% approval rate. Perfectly improving with each hire. Without a data flywheel, you're just a packager.

7. Do not build a general AI packager. "AI for everything" loses to "AI replacing a specific $80,000/year job." Deep dive into an unsexy industry. The best opportunities are in industries you would never pitch at a cocktail party.

8. Consumer Absence is an Opportunity Signal. Zero to minimal education companies. Zero consumption social. Zero mental health/fitness. Zero govtech. Historically, the least funded categories produce the most outsized returns. Founders cracking native AI entertainment, social, or education will dominate the entire category.

9. Hardware is back. 18% of the batch has hardware components (robots, drones, wearables, space tech). This is a significant jump from recent batches. The entity product companies founded by SpaceX/Tesla alums are the most differentiated in the batch.

About Distribution Channels

10. Distribution channel is a prerequisite, not an afterthought. 60% of the top 15 growth companies acquire customers through founder networks or the YC network. If your first 20 customers needed to "figure out distribution channels," you've picked the wrong market.

11. Your former employer is your first market. Dominant GTM action (about 35% B2B): Founders spent years in the industry, left, then sold back to their network. Their rolodex is the distribution channel.

12. PE M&A channels are severely underrated. Ressl AI and Robby independently discovered PE-backed acquirers direly need profit-improvement tools. One PE deal = 50-200 nodes.

13. Choose a market where you already have a distribution network. Companies struggling with GTM are almost always those who build a product first and then ask, "How do we sell this?" The winners ask, "Who do I already have access to, and what do they urgently need?"

About the Team

14. Founder-Market Fit is the strongest predictor of revenue velocity. Founders who have truly done the work they are now automating can close deals in days. Others take months. Proximitty (<$700k ARR in under 3 weeks): CEO is a McKinsey banking risk consultant. Corvera ($33k MRR in 4 weeks): CEO ran a CPG brand.

15. Your co-founding relationship is your moat. 46% of batches are 2-person teams. Strongest teams have collaborated for years: former colleagues, classmates, siblings, repeat co-founders. If you haven’t launched something with a co-founder, you haven’t validated the most critical part of entrepreneurship.

16. Domain expertise trumps pedigree. Most compelling founders have lived the problem: dentist building surgical AI, aircraft maintenance manager building mechanical tools, lobbyist building policy AI. "Ex-FAANG" is a hygiene factor, not a differentiator.

About Pitching

17. A strong closing line is crucial. When 199 companies pitch in one day, you need to be what they talk about over drinks. "The first AI Oscars will be born over Martinis." "You can book your stay at the Moon Hotel in 2032." Make your vision specific, falsifiable, quotable.

Avoiding Pitfalls

18. Avoid undifferentiated agent infrastructure. 8-10 companies are building agent monitoring/testing/compression. Core model providers will natively build these. If " [Existing DevOps Tool] but for AI agents" describes you, it's a danger zone.

19. Avoid AI-native services with no data moat. Fastest to revenue but lowest defensibility. Core tech replicable within weeks. Traditional companies adopt AI in 12-18 months. Without proprietary data or embedded distribution, the moat is thin.

20. Avoid commoditized workflow wrappers. AI does a narrowly defined task, and GPT-5 might do the same natively in 6 months.

On the Ground

199 pitches. Fresh startups out of the YC oven have a distinct smell. Excitement, high energy, never dull.

Some memorable moments:

A startup pitches the first hotel on the moon, with a White House invite and a $500 million Letter of Intent

A robot cowboy herds cattle with autonomous drones

An AI demo company live-generates its own demo deck during the demo

A company casually zooms in on satellite imagery to Tehran, Iran during a satellite image demo (room goes silent)

The Martini founder ends with, "Martini will win the first AI-produced movie Oscar!" – a line that either makes VCs roll their eyes or reach for their checkbooks

The hardware demo area is buzzing: robots, drones, microscopes with life sciences proteins, in-car radar. Tangible, touchable physical things. This isn’t just another batch of SaaS dashboards.

After listening to 199 pitches, you start to see patterns instead of individual companies. Here are my findings.

Macro Trends

Total Companies: 199

Business Model:

· B2B: 174 (87%)

· B2C: 14 (7%)

· B2B2C: 11 (6%)

Product Types:

· Pure Software: 163 (82%)

· Hardware + Software: 24 (12%)

· Pure Hardware: 12 (6%)

AI Classification:

· Native AI (AI is the product): 120 (60%)

· AI-Enhanced (existing workflow + AI): 52 (26%)

· Non-AI: 27 (14%)

Tractability:

· Estimated Median ARR: $50-100k

· Estimated Median Growth: 30-50% MoM

· Companies with ARR> $1M: ~5%

· No revenue: ~50%

Primary Industries: B2B Software (59%), Industrial (15%), Healthcare (10%), FinTech (8%), Consumer (4%).

Only 14 companies are consumer-facing, with YC officially categorizing only 7 as "Consumer." The rest are consumer products disguised as enterprise, categorized as B2B, Healthcare, or FinTech.

Ten Major Themes

1. AI Agent Replacing Entire Job Functions

Core theme.

Not a copilot, but a full replacement.

· Beacon Health replaces pre-authorization admin staff

· Perfectly end-to-end replaces recruiters

· Lance replaces front desks at 50+ Marriott/Hilton/Hyatt hotels

· Mendral (Docker co-founder) replaces DevOps engineers

· Canary replaces QA

The "copilot" framework decreased from about 4% of pitches in early 2025 to 1% in W26.

2. "Claude Code for X Field"

Claude Code and Cursor have proven that AI agentization is effective for code. W26 founders are applying the same paradigm to every profession with structured output:

· REV1 for Mechanical Engineers (3D to 2D drawings)

· Avoice for Architects (specifications, documents)

· Synthetic Sciences for Scientific Research

· Maywood for Investment Bankers

· Alt-X for Real Estate Underwriting (working directly in Excel)

· Cardboard for Video Editing

Mango Medical generates surgery plans in minutes instead of days

3. AI-Native Professional Services ("Service Business, Software Economics")

Not building tools for existing companies, but building AI companies to compete with them:

Four AI Law Firms (Arcline, General Legal, Vector Legal, LegalOS)

· AI Recruiting Agency (Perfectly)

· AI Accounting (Balance)

· AI Insurance Brokerage (Panta)

· AI Policy Consulting (Fed10, founded by three former lobbyists)

Panta explicitly states: "A software-economics-driven services business." Operating on outcome-based pricing, running on software profit margins, as AI does 80% of human work 20%. Arcline has 50+ startup clients. LegalOS boasts a 100% visa approval rate.

Bear Case: People in the loop cap profitability at 60-80%. Liability is real. Moat Issue: If core tech is "LLM + domain hints + manual review," what prevents replication? Emerging Answer: Start with service → release automation → evolve into a platform. Service is the wedge; software is the moat.

4. Infrastructure of the Agent Era

Every layer of the tech stack is being rebuilt for agents:

· Agentic Fabriq = "Okta for Agents"

· Sponge (three ex-Stripe crypto leads) = financial infrastructure for agents

· Moda/Sentrial = Datadog for agent reliability

· Salus = runtime guardrail

· 21st (1.4M developers) = React components for AI-driven UI

Zatanna is turning pre-LLM SaaS into a queryable database for agents

Risk: Foundational model providers are native-building these. About 30% competition overlap at this layer validates it's crowded.

5. Vertical AI in "Unsexy" Industries

Highest ROI in tech-overlooked sectors:

· Zymbly automates aircraft maintenance paperwork (5 mins of repair requires 45 mins of documentation)

· GrazeMate builds robotic cowboys, autonomous drone herders. You chuckle as they pitch. Sounds absurd until you figure out the founder grew up on a 6,000-head cattle ranch.

· OctaPulse: Computer vision for fish farming

· Squid: Addressing grid planning (a $760 billion annual inefficiency still managed through spreadsheets)

These founders have deep roots. The founder of Scout Out is a fourth-generation construction professional. The co-founders of LegalOS grew up in a family immigration law firm (each over 10,000 hours since the age of 12). The co-founder of Zymbly was a Virgin Atlantic aircraft maintenance manager. The best opportunities lie in industries you'd never pitch at a cocktail party.

6. Physical AI/Robotics Renaissance

18% of the batch has a hardware component:

· Remy AI and Servo7 are building warehouse robots that learn from human demonstrations (80% of warehouses are not automated)

· Origami Robotics is building robotic hands

· RoboDock saw a 60-day MVP deployment skyrocket and secured a $100,000 contract from Waymo

· Fort (three ex-Tesla engineers) is tracking strength training, something that Whoop/Oura still can't do

· Pocket has shipped over 30,000 units, with an annualized revenue of $27 million

The hardware demo area was the most vibrant part of the day.

7. Defense and National Security

Milliray (three Oxbridge PhDs) built drone-detecting radar for NATO (sold $470,000 within the batch)

Seeing Systems created AI-strike drones for the Royal Marines

DAIVIN! built tankless diving gear for U.S. Special Forces

Defense budgets are large, contracts are long, and reputations can carry over into commercial.

8. Data as a Moat

When everyone has the same base model, proprietary data is the primary defense:

· Shofo: World's largest indexed video library

· Human Archive: Dropped out of Stanford/Berkeley, moved to Asia, and collected data from thousands of families for human-shaped robots

· LegalOS: 12,000 Successful Visa Applications → 100% Approval Rate

Theme: Each customer interaction makes the product better. No data flywheel, you are the packager.

9. Deep Tech and Space

The boldest roadshow. GRU Space is building the first hotel on the Moon by 2032. As they roadshowed, rooms realigned: half thought they were crazy, half thought they might pull it off. $500 million LOIs, White House invite, 1B+ views. Beyond Reach Labs building an orbital solar array the size of a soccer field (power needs to increase 500x by 2030). Terranox using AI to discover uranium deposits (single discovery = $200-700 million).

Ditto Biosciences might have the most creative take: parasites evolved proteins that control the human immune system over millions of years. Ditto uses AI to identify and design autologous immunotherapies. Evolution solved the problem; they just read the answer.

10. AI-Native Research and Science

Talking Computers deploying AI scientist fleets (ARR over $1 million)

Aemon (twin brothers, published at ICLR/EMNLP before age 20) sets sub-$10 computation world records on NP-hard math problems, outperforming Google DeepMind

Ndea, co-founded by Mike Knoop of Zapier and François Chollet, creator of Keras, explicitly building AGI that can innovate

Founders: A Pattern of 429

Demographics:

· Approximately 60% Immigrant/International

· 86% Male, 14% Female

· Top Schools: Berkeley (approx. 45), Stanford (approx. 35), MIT (approx. 20), Waterloo (approx. 15)

· 55% CS Major; 45% Non-CS

Background:

· Approximately 30% ex-Big Tech

· Approximately 25% Previous Entrepreneurship

· Approximately 12% Former Finance/Trading (Citadel, Jane Street, Jump)

· SpaceX alone has about 12 founders, the vast majority building hardware and aerospace

Team:

46% are 2-person teams, 15% solo founders

Most common archetype: a technical duo with different expertise (about 35%), not the classic "hacker + hustler"

19% of companies have at least one founder with a Ph.D.

How they met: about 35% are university classmates, about 25% former colleagues, about 15% repeat founders, about 10% family/siblings

Becoming a founder as a domain expert is the most compelling story: Adrian Kilian (dentist→Mango Medical surgical AI), Robbie Bourke (25 years in aviation→Zymbly), Pamir Ehsas (OpenAI external legal counsel→Arcline), Conor Jones (years in national grid→Squid).

Some Observations:

Deep domain expertise + technically capable co-founders = strongest companies in the batch

The most successful teams either previously built and sold a company together, or worked side by side at the same company solving the exact problem they are now solving

31% of companies have at least one Ph.D. or researcher founder, primarily concentrated in healthcare/biotech, hard tech, and AI infrastructure

How They Found Market Fit

B2B (88% of the batch)

“I lived this pain point” (about 40%): Strongest pattern. End Close founders spent 6 years at Modern Treasury handling over $1 trillion in payments. Squid founders spent years in the national grid. They don’t need customer discovery; they are the customer.

“I built the platform I aim to replace” (about 20%): Mendral co-founded by Docker. Perfectly built by a TikTok ML scientist. They intimately know the stack and see where AI makes a quantum leap.

“50 conversations sprint” (about 15%): Systematic discovery. Ritivel had 50+ pharma conversations before writing code. Ressl AI started from consulting, finding deals had the most glue work.

“Infrastructure Prophecy” (approx. 15%): Thesis-driven. “If agents exist, they need to authenticate” → Agentic Fabriq. Risk: Building for the future 2-3 years out.

“Research to Commercialization” (approx. 10%): CellType (Yale Prof + Google DeepMind). Valgo Co-founder actually wrote a security-critical systems textbook.

B2C (batch's 7%)

“I Am The User” (approx. 50%): Fort founder is a disillusioned weightlifter with wearables. Doomersion founder binge-watched short videos and studied languages, combining them.

“Format Shift” (approx. 25%): Existing behavior + new medium. Pax Historia: Love for strategy games + AI substitutes history.

“Hardware Wedge” (approx. 25%): Physical product creating data loops software cannot replicate.

Meta-lesson: No successful W26 company was born out of a hackathon or “what if we used AI to…” brainstorm. Each stemmed from deep personal experience or obsessively customer discovery.

How They Found Distribution Channels

Data is clear: Founder networks are the #1 driver for the fastest-growing B2B companies. 60% of the Top 15 in growth acquired their initial customers through founder networks or YC network.

B2B Model:

“Sell to Former Employer Peers” (approx. 35%): Three Fed10 alums, their rolodexes were the distribution channel

“YC as a Launchpad” (approx. 25%): Cardinal cold-called 40+ YC companies, Palus Finance closed 33 in weeks

“Open Source” (approx. 10%): 21st has 1.4M developers, effective only for infrastructure

“PE M&A Channel” (approx. 8%): One deal = 50-200 endpoints

“Systematic Cold Calling” (approx. 15%): Limited buyer list with quantifiable pain points

“Wedge Product” (approx. 7%): Narrow entry, broad expansion

B2C: The product is the distribution channel. Doomersion 2 received 15,000 downloads in 2 weeks through zero-cost marketing. Pax Historia achieved tens of thousands of DAU through organic growth. A hardware founder bet on physical presence to drive word of mouth.

Main Takeaway: Companies struggling with GTM are almost always those who build the product first and then ask "how do we sell this?" Winners ask "Who can I already reach, what do they urgently need?" and then build that.

Great Pitch Breakdown

Seven components distinguish a memorable pitch from a blurry one:

1. Hook

Three effective prototypes:

Shock Data: "Bringing a drug to market takes 500,000 days. We aim to make it 5 days" (Rhizome AI)

Reframe: "Every file you've uploaded uses a protocol from 1974" (Byteport)

"I am the problem": "I spent 6 years at Modern Treasury building reconciliation, handling $1 trillion" (End Close)

2. Problem (Specific, Not Generic)

"Engineers spend half their time on paperwork" (Zymbly) is more impactful than "We automate backend workflows."

3. Team (One-Liner Credibility Bomb)

"Andrea wrote the first line of code for Docker" (Mendral). "Our team invented the MPIC standard that secures every HTTPS connection on the internet" (Crosslayer Labs).

4. Market (Inevitable, Not Just Big)

"Satellite power demand: set to increase 500 times by 2030" (Beyond Reach Labs). The best market pitches explain why now and why it's inevitable, not just how big the TAM is.

5. Traction (Speed> Absolute Numbers)

"$33,000 MRR in 4 weeks from 0" (Corvera) is more compelling than "Achieved $100,000 ARR" without a timeframe.

6. Unique Insight

「Parasites have evolved a protein to control the human immune system. We are reading their answers」 (Ditto Bio).「Insurance companies cannot price for autonomous systems as historical claims data is non-existent」 (Valgo).

7. Wild Ending

「The first AI Oscar will be born at Martini.」「Booking the Moon Hotel for 2032」 (GRU Space).

Fuzzy Pitch: Generic "AI for [Industry]," team credentials unrelated to the problem, and (crucially) no wild ending.

Competitive Overlap: YC's Multi-bets

Around 30% of companies have direct competitors in the batch. Only about 5% face true high overlap.

High Overlap: LLM contextual compression (Token Company vs. Compresr), medical legal documents (Wayco vs. Docura Health), robot data (Human Archive vs. Asimov)

Medium: Startup law (Arcline vs. General Legal vs. Vector Legal), AI SRE (IncidentFox vs. Sonarly), agent monitoring (Sentrial vs. Moda), prior authorization (Ruma Care vs. ClaimGlide vs. Beacon Health)

What it tells you: YC bets on markets, not companies. Three law startups = market real and large enough to accommodate multiple winners. Two similar-looking companies on Demo Day will look completely different by Series A. Most differentiated companies zero overlap: Terranox, Zymbly, GrazeMate, Ditto Bio. In each case, founder domain expertise is the moat.

Grossly Absent

· Zero EdTech companies

· Zero GovTech

· Zero Consumer Social

· Zero Mental Health/Fitness

· Almost zero Marketplaces

· Almost zero pure crypto (blockchain used as a pipe, never as a product argument)

· Consumers at an all-time low (only 14 companies in total, only 7 officially classified)

Industrial jumps from 3.6% in W24 to 14.1% in W26, a 4x leap.

The "Atom vs Bit" transformation is real within YC.

Reverse analysis: The composition of W26 is a snapshot of what is fundable now, not what will be valuable in 10 years. The missing legendary companies in this batch are those consumer and social founders who will come in 2-3 batches once AI catches up to their ambition.

What Might Fail

Undifferentiated agent infrastructure. 8-10 companies doing agent monitoring/testing/compression. Infrastructure model providers will natively build these. Enterprise buyers default to existing vendors.

AI-native services without a data moat. Fastest to revenue, lowest defensibility. Core tech replicable in weeks. Traditional companies adopt AI in 12-18 months.

Solo technical founders for a relationship sales market. Construction, insurance, freight: if no one can walk the job site and speak the language, they stall.

Lack of domain-depth "AI for [Industry]." Red flag: Description starts with "We use advanced LLM agent..." instead of the customer's specific pain point.

Revenue-less long-cycle deep tech. Conceptually correct but burn money is the failure mode.

Commoditized workflow wrappers. Single-task AI, GPT-5 might natively do the same in 6 months.