How To Get AI Startup Ideas

YC AI Startup School Announcement

• Event Details: YC will host its first-ever AI Startup School in San Francisco on June 16th and 17th.
• Speakers: Confirmed participants include Elon Musk, Satya Nadella, Sam Altman, Andre Karpathy, Andrew Ng, and Fei Fei Li.
• Eligibility: The conference is free and exclusive to computer science graduate students, undergraduates, and recent graduates in AI and AI research.
• Support: YC will cover travel expenses for attendees, though space is limited and application is required.
• Core Thesis: Founders are advised to persist with cutting-edge AI startups even without a finalized idea, as the rapid pace of technology increases the probability of a "lucky break" occurring soon.

Strategy for Generating Startup Ideas

• Avoid "Lazy" Ideas: Successful founders are cautioned against "hackathon-style" ideas or simply jumping on current trends, as these often lack long-term viability.
• The "Edge" Principle: Founders must position themselves at the "edge" of human knowledge or industry practice rather than operating within their immediate comfort zone.
• Dual Approach to Ideation:
  • Aggressive Introspection: Deeply analyze one's own unique history, research, and professional experience to identify unmet needs.
  • Aggressive Externalization: Physically enter different industries (government, healthcare, logistics) to understand root problems through first principles.
• Founder-Market Fit: The most viable ideas often arise when a founder possesses a unique, PhD-level depth of understanding in a specific domain that no one else in the startup world possesses.

Case Studies: Ideas Derived from Internal Expertise

• Salient:
  • Problem: Manual, outsourced processes for auto debt collection payments at Tesla Finance Ops.
  • Solution: AI voice agents for loan processing, founded by a former Tesla employee who witnessed the inefficiency firsthand.
  • Outcome: Successfully servicing large banks by automating a niche, esoteric workflow.
• Diode Computer:
  • Problem: Lack of software tools (like Git) for hardware engineering, forcing manual parsing of data sheets for component verification.
  • Solution: AI circuit board copilot.
  • Founder Advantage: The team combined electrical engineering (hardware) and software engineering expertise, creating a unique intersection of skills.
• Spur:
  • Problem: Figma engineers spent excessive time manually writing and maintaining tests for complex front-end interfaces.
  • Solution: An AI QA agent that automatically writes and maintains software tests.
  • Source of Insight: Derived directly from the founder's tenure at a company with notoriously difficult testing requirements.
• DataCurve:
  • Pivot Story: Initially a "toy" hackathon project (Uncle GPT) that failed to find product-market fit.
  • New Focus: AI tools for product managers, derived from the founder's internship at Cohere working with LLMs and synthetic data.
  • Performance: Raised to mid-to-high seven figures within a year of pivoting.

Case Studies: Ideas Derived from External Investigation

• Egress Health:
  • Method: The founder shadowed their mother (a dentist) to understand office workflows.
  • Discovery: Excessive manual admin work regarding insurance processing and pre-authorizations.
  • Solution: LLM-powered back-office automation for dental practices.
  • Trend: Utilizing family connections to gain "access" to underserved industries where software has historically been absent.
• Able Police:
  • Trigger: Personal experience with the police officer's paperwork burden (e.g., San Francisco laws requiring reports for minor stops).
  • Discovery: Police officers spend hours filling out forms that could be automated via LLMs and computer vision using existing camera data.
  • Execution: Founder conducted "undercover" research via ride-alongs and deep investigation into the root cause of clerical inefficiency.
• Happy Robot:
  • Method: Founders used personal charm to access the logistics industry without prior connections.
  • Solution: AI agents for coordinating trucker logistics.
• Medical Billing Company (Undisclosed Name):
  • Method: A co-founder took a remote medical billing job without disclosing their intent to build software.
  • Execution: Built a local AI agent using Llama 3 to automate their own work, eventually pivoting to sell the solution.
  • Legal Note: This approach is legal provided the software is built locally on the worker's own hardware without violating client confidentiality.
• Sweet Spot:
  • Insight: A friend's job involved manually refreshing government websites to find contract bids.
  • Solution: AI platform for government contracting that finds opportunities, generates bids, and optimizes pricing.

Strategic Trends and Market Observations

• Automation of Outsourced Roles: Jobs that have been outsourced to low-wage countries (e.g., Lilac Labs automating drive-thru order takers) signal high potential for AI automation.
• Improving Existing RPA: Companies like Automat identified that existing Robotic Process Automation (RPA) tools like UiPath require expensive consultants to implement; they built an AI-native alternative that works out-of-the-box.
• Vector Database Innovations:
  • PreyDB: Identified a need for real-time sync between Postgres and Pinecone, later pivoting to extend Postgres capabilities to replace separate vector databases.
  • Reducto: Discovered data chunking issues for RAG applications by observing peers at the cutting edge of AI development.
• Recruiting Search: PeopleGPT evolved from a freelancer marketplace concept into an AI-powered people search engine for recruiters, leveraging deep user engagement to find the real need.
• GigaML: Despite entering a "crowded" customer support market, their superior technical execution (fine-tuning models) allowed them to secure a major deal with Zepto, an Indian delivery company.
• Seed Extension Validity: The era of "never do seed extensions" is over; AI companies often require a year or more to pivot and find product-market fit due to the speed of technological change.

Founder Psychology and Advice

• Overcoming Competition Fear: Founders often avoid good ideas because they perceive a market as "too crowded," ignoring the reality that few competitors possess the technical depth to execute complex AI problems effectively.
• The "Toilet" Analogy: Founders who ship products and talk to users possess "real world" data, whereas investors or non-founding team members often rely on "Plato's cave" shadows (social media, news feeds) and fail to understand the product's viability.
• Blinders: Reading Paul Graham's How to Get Startup Ideas is recommended to identify "blinders" that prevent founders from seeing ambitious ideas because they seem too scary.
• Internships as Incubators: A significant percentage of YC's billion-dollar companies trace back to insights gained during internships at bleeding-edge companies (e.g., Cohere, Scale AI).
• Undercover Strategy: Founders are advised to take low-barrier entry jobs (e.g., data entry, clerks) in target industries to gain deep, first-principles understanding of workflows that can be automated.