Insights
19
Nov
2026

Notes from a week with ElevenLabs in New York

Karolina's notes from three days with ElevenLabs in NYC

The ElevenLabs Summit NYC welcome screen above four empty chairs on stage

By Karolina Kondrat, VP of Partnerships

A big part of my job is making sure we know our partners' platforms (hence my title) inside and out before we bring them to a client.

That's how I ended up spending three days in NYC with ElevenLabs at their Summit on 11/11. (Coincidence? Think not.)

So what did I actually learn from three days with some very smart people? Glad you asked.

What you’ll find below:

  • 3 key takeaways from the Summit
  • The one piece of advice I kept hearing
  • What I took home (and you can too)

3 key takeaways from the Summit

The ElevenLabs Summit NYC welcome screen on stage

AI IQ isn't the bottleneck anymore

Miles pointed this out a few months ago, but it's even truer today: all the frontier models are very capable; the real work now is making interactions between humans and AI better. If you have to direct your time and attention anywhere, it's here.

Most AI projects that fail don't break down because of the model. They fall apart because of everything around it: what happens when someone changes their mind halfway through a conversation, when it hands off to a real person, and how it actually connects to the systems behind it.

IRL example: a voice agent knows exactly when your dentist has an opening but can't book it, because nobody connected it to their Google Calendar. The model did its job. The experience still flopped. And…now you've got cavities.

Voice is judged on a different scale

Not fair, I know. But people speak way faster than they type and are less patient in conversations. A pause that's fine in a text chat is unforgivable on a call. Latency will kill your credibility. Plus, voice agents often handle more emotional moments, like a missed flight or a billing problem, so tone and empathy matter just as much as accuracy. Nobody wants a robot voice anymore (did they ever?!). If you're going to use AI voice, these are non-negotiables.

Very few people hang up on chatbots (they may forget what tab they left it in but that's a them problem). Almost everyone hangs up on bad phone agents all the time.

Regulated industries are next up

During the Summit, ElevenLabs launched their new HIPAA-compliant model specifically designed and built for healthcare.

While some of us are happy to hand over our data to the latest Opus release, healthcare, finance and insurance have strict compliance and data requirements, and the models you and I use day-to-day weren't built for it, blocking adoption.

They also use a lot of funky words: thousands of complex medical terms, finance's alphabet soup (EBITDA, anyone??). A model that you and I use hears "HbA1c" or "COPD" and guesses, with dire consequences 😬.

Start small

The most consistent advice I heard all week: don't try to do everything all at once. In other words, don't bite off more than you can chew. That's a surefire way to fail.

Pick the types of conversations you have a hundred times a day, like booking appointments. Then decide up front how you'll know it's working (happier customers? shorter calls?). Once you've dialed it in, add the next.

And to be honest, this matches with what we see on a daily basis with our clients: focus on small, repeatable wins early. Then build on those.

If you want to steal something, here's the checklist we run before building any voice agent:

  1. Is this a conversation your team is having a hundred times a day?
  2. Do you know what "good" looks like, and how you'll measure it?
  3. When the agent gets stuck, who picks up?
  4. Can it reach the systems it needs (your calendar, your CRM, your billing tool)?

If you can't answer all four, you aren't quite ready to build. But! You're ready to scope. (Which, conveniently, is something we do as an ElevenLabs partner and Claude Select Partner 😉.)

What I took home (and you can too)

Here's what I'd tell any team looking at it:

  • The model is the easy part. Gemini, Claude, ChatGPT are all very capable. Focus on designing a nice experience for your customers. That's where the difference is: the conversation itself.
  • Start with one use case you can measure. Once you've dialed it in (congrats btw), start on the next one.

If you think AI is still running on the hype train or it's coming for all our jobs, hit reply. I'd love to chat! No question about it, the world is changing pretty fast and the summit was a good example of how quickly things are accelerating.