Complete List of YC W23 Companies With Descriptions

• Batch size: ~282 companies (largest batch at the time)

• Demo day: April 2023

• Historical significance: First post-ChatGPT batch

• AI proportion: ~35% (up from ~15% in W22)

• Funding environment: Tightening — the 2022-2023 correction was underway

• International proportion: ~38%

• Defining question of the batch: "Is your AI product real or is it a GPT wrapper?"

AI AND MACHINE LEARNING

Notable W23 AI companies:

DEVELOPER TOOLS

Notable W23 Developer Tools companies:

B2B SAAS

Notable W23 B2B SaaS companies:

FINTECH

Notable W23 Fintech companies:

HEALTHCARE AND BIOTECH

Notable W23 Healthcare companies:

CONSUMER AND SOCIAL

Notable W23 Consumer companies:

INDIAN-ORIGIN W23 COMPANIES

THE GPT WRAPPER PROBLEM

W23 was the first batch to grapple with what investors called the "GPT wrapper problem" — the concern that many AI startup products were thin interfaces on top of GPT-3/GPT-4 capabilities that would be commoditized as OpenAI improved its own products. This concern dominated W23 investor conversations at demo day and shaped the fundraising outcomes significantly.

Companies that escaped the "GPT wrapper" label in W23 shared these characteristics:

• Proprietary data that improved the AI product with usage

• Deep workflow integration that created switching costs

• A specific domain (legal, medical, financial) where the AI required fine-tuning or specialized knowledge

• Infrastructure that was model-agnostic — not dependent on any single foundation model

W23 AI companies that struggled to raise were typically those where the product was essentially: "we put a nice interface in front of GPT-4 and pointed it at [vertical]." Without a clear data moat or workflow integration story, these companies could not answer the commoditization question convincingly.

THE W23 BATCH SIZE EXPERIMENT

At 282 companies, W23 was a significant increase from W22 (approximately 200 companies). YC's decision to scale batch size reflected confidence in the quality of applications driven by AI startup formation. However, the larger batch also meant more companies sharing the investor attention window at demo day — which may have contributed to the slower fundraising pace for mid-tier W23 companies compared to earlier, smaller batches.

W23 NOTABLE OUTCOMES BY 2025

DEVELOPER TOOLS DOMINATED W23 POST-DEMO FUNDRAISING

An unusually high proportion of W23's strongest fundraising outcomes came from developer tools companies — Neon, Zed, Railway, Depot, Speakeasy — rather than from AI application companies. This counter-intuitive outcome (given the AI excitement surrounding W23) reflected two factors: developer tools had more established metrics frameworks (daily active developers, SDK downloads, GitHub stars) that made evaluation easier, and the developer tools category was less subject to the "GPT wrapper" commoditization concern.

W23 occupies a unique position in YC batch history because it sits precisely at the transition between the pre-LLM and post-LLM eras of startup formation. Understanding this context is useful for founders applying today.

What W23 got right about AI startups:

• Vertical-specific AI with deep workflow integration outperforms horizontal AI

• Proprietary data moats are real and matter for AI product defensibility

• Infrastructure and tooling for AI developers is a large, durable opportunity

• The GPT wrapper concern was legitimate — thin wrappers have largely not survived

What W23 got wrong (in retrospect):

• The pace of foundation model improvement was underestimated — some products that seemed defensible in April 2023 were commoditized by June 2024

• The enterprise AI adoption timeline was underestimated — companies in W23 that projected fast enterprise sales cycles found 2023-2024 enterprise AI procurement was slower than expected

• Consumer AI retention was harder than W23 consumer AI companies projected — the novelty period was shorter and the drop-off steeper than hoped

What W23 teaches founders today: The companies that succeeded from W23 — Neon, Zed, Cohere, Ambience — all had specific, identifiable moats that went beyond their AI capabilities. Infrastructure quality, data network effects, domain-specific fine-tuning, and deep clinical workflow integration are examples. Founders today should be able to answer specifically: "What about this product is defensible in a world where foundation model capabilities double every 12 months?"