How to use AI in your startup: a practical framework from YC

Executive overview

Most founders asking "should I pivot to AI?" are asking the wrong question. Every new company today should be leveraging LLMs — the question is whether the fundamentals are in place to make the technology produce value.

We are in a moment analogous to the cloud shift of the early 2010s: a window where rebuilding existing software natively with AI will produce something categorically better. Big companies are slow. That gap is the opportunity.

The core insight: switching your idea to make calls to OpenAI does not change your fate — only better customer insight, better execution, and the right environment do.

Should you pivot to AI?

• Don't pivot just because AI seems like a good idea — that alone changes nothing.
• Do build something with LLMs at the heart of it; using AI internally counts too.
• Avoid obvious, crowded angles (e.g. "customer support agent #50") with no new insight.
• Founders who succeed in pivots also change their environment, find new customer insights, and embed in the right communities.
• The WAPI story: a team pivoted from a Zoom productivity tool, moved from Toronto to SF, and grew to power a significant portion of voice-AI companies in under 18 months.

Why location still matters

• San Francisco is where the state of the art is being built, debated, and shared in real time.
• Proximity lets you learn from the best instantly — the equivalent of walking downstairs to ask Pinterest about SEO.
• If you can't relocate, spend 3–4 weeks embedded here: attend hackathons, meet other founders.
• Not being here means you may not know what's impressive, what's not good enough yet, or what deals won't close.

How to find AI opportunities

• Look for legacy software that moves data between systems via a human in the middle — those tasks are fully automatable with agents.
• Spend a few hours watching someone do their job; repetitive tasks that are worth automating become obvious quickly.
• Domain knowledge unlocks edge: the Medicare Advantage copilot worked because the founders had previously worked inside that workflow.
• US healthcare alone has ~$1.3–1.4 trillion in admin spend, most of it legacy software plus manual data transfer.

Examples from recent YC batches

• Replexica: automates UI localisation across languages, putting it inside the software release cycle.
• Gecko Security: AI security engineer that automates previously specialised work for software teams.
• Pre-auth automation (TiVara): pulls doctor notes, summarises them, and auto-generates prior-authorisation requests.
• Voice AI patient follow-up: calls patients between visits to check in and book appointments — better for patients, more revenue for practices.

What the big technology shift means for timing

• Major companies are typically preceded by major technology shifts; this one began roughly two years ago and is accelerating.
• Large incumbents are moving slowly — ChatGPT is over two years old and Alexa still hasn't caught up.
• There are things you can build today that were impossible 18 months ago; the window is open but won't stay this wide.
• Founders who haven't lived through a prior platform cycle (mobile, cloud) may not recognise the scale of what's happening.