Ideas & Opportunity

How do I spot an idea that AI just made possible that was not viable two years ago?

A starting point

Look for workflows that used to need a skilled human for every step and ask which of those steps a model can now do at 80 percent quality for a fraction of the cost. The opportunity is rarely a raw chatbot, it is a specific painful job (drafting, reconciling, triaging) where good-enough automation changes the economics. Beware building a thin wrapper that a platform update erases overnight, so anchor on proprietary data or workflow depth.

Go deeper

Hand-picked from around the web, each with a note on why it earns your time.

3 resources 3 link-checked Watch Read

Watch

▶️ Video
✓ Link checked Free Beginner

Why we picked it YC's partners walk through concrete business models that were not viable before large language models: full-stack law firms, personalized tutors, recruiting and technical screening that can finally scale. It is a current, specific view of AI-enabled problem spaces from people who see thousands of applications a batch. Treat their examples as patterns to reason from, not ideas to copy outright.

Startup Ideas You Can Now Build With AI (Lightcone Podcast)

On Y Combinator (Lightcone Podcast) by Garry Tan, Harj Taggar, Diana Hu, Jared Friedman ~45 min

  • The strongest AI ideas are ones where the old cost structure blocked a real business that now works, for example AI running technical interviews or one-on-one tutoring at scale.
  • Look for services that were too expensive to deliver per customer and are now cheap enough to build a company around.
  • The partners frame many of these as second chances at markets that failed before, so a previously dead idea is worth re-examining against today's model capabilities.
Open ycombinator.com

Read

✍️ Essay
✓ Link checked Free Intermediate

Why we picked it This is a16z's clearest map of where AI genuinely opens new application-layer companies versus where it just produces thin wrappers around a commodity model. It names the actual openings (counterpositioning on business model, workflow orchestration, embedded domain expertise) and spells out why a slick interface over ChatGPT is not one of them. Read it as a lens for judging your own idea, not a shopping list.

Good News: AI Will Eat Application Software

From Andreessen Horowitz (a16z) by Alex Immerman and Santiago Rodriguez ~20 min read

  • The real openings come from owning a workflow and embedding domain expertise, not from a nicer UI on top of a general model.
  • New entrants can win by counterpositioning on business model (for example per-conversation pricing instead of per-seat), which incumbents struggle to copy.
  • A frontend that mostly re-skins commodity functionality stays vulnerable, so ask what your product still delivers if the model layer became free tomorrow.
Open a16z.com
✍️ Essay
✓ Link checked Free Intermediate

Why we picked it Elad Gil is one of the most trusted operators-turned-investors on this exact question, and this piece cuts through the wrapper panic honestly. His core point is that most startups (AI or not) start non-defensible, and durable positioning is built after launch through data, integrations, and relentless execution, not claimed on day one. It is a grounding read for a founder worried their idea is too easy to copy.

Defensibility & Competition

From Elad Gil (Elad Blog) by Elad Gil ~12 min read

  • Serving a real customer need well usually matters more than having a moat on launch day, and defensibility tends to accrue over time.
  • Building on top of a model like GPT is fine, but the less you keep building and expanding after launch, the faster you get commoditized.
  • Durable positioning comes from proprietary data, deep integrations, and execution velocity, so pick an idea where using the product compounds an advantage.
Open blog.eladgil.com

People also ask