Key Takeaways

  • 18 customer support AI startups raised over $100 million in the past 18 months, creating a crowded application layer led by Sierra and Decagon.
  • Sierra succeeds through enterprise distribution and Bret Taylor's executive access, not internal voice model training.
  • The foundational voice API market contains only five major players: Speechify, ElevenLabs, Cartesia, Google Gemini, and xAI's Grok.
  • Voice acts as the wedge to capture user attention, but back-end tool calling across enterprise databases creates actual software retention.
  • Voice conversation is positioned to replace visual screens as the default interface for software, echoing Google's replacement of complex search portals.

The Disagreement: Crowded Support Apps vs Scarce Voice Models

Stebbings remains skeptical about pure-play customer support startups. In his view, the sector is dangerously overheated. “I just think the customer support market's a challenging market to really get behind,” Stebbings argues. “You have Sierra and Decagon out in front with the majority of funding and attention. But to say that there are 18 companies have now raised over 100 million in the last 18 months.” Stebbings warns that enterprise buyers will build internal tools or stick with existing CRM vendors rather than paying new software startups.

Weitzman draws a line between customer support software and foundational voice infrastructure to explain why certain layers remain defensible. Sierra is not an audio research lab. “Sierra doesn't have their own model team,” Weitzman notes. “They use other people's models, right? Because the value of Sierra is the go to market. It's Bret Taylor.”

While hundreds of startups fight for customer service budgets, the raw voice model market has very few contenders. “In the API business, you have Speechify, ElevenLabs, Gemini, Grok, so SpaceX is now in the race, and Cartesia, and that's kind of it,” Weitzman says. “And so that's not that competitive of a space compared to the B2B customer support thing.”

Who's Right (and When They're Wrong)

Both views identify real risks, but they describe different battlegrounds.

Stebbings is right about mid-tier support wrappers. If your product only translates text into synthetic speech and answers basic FAQ tickets, you will get crushed between Sierra's executive relationships and native CRM features. 18 well-funded competitors chasing the same customer support budget means customer acquisition costs will spike while churn stays high.

Weitzman is right about where technical defensibility actually lives: back-end database execution and core infrastructure. “Put another way, if you use a tool like Sierra, the wedge right now is voice, but the important part is tool calling,” Weitzman explains. “ElevenLabs lets you do some tool calling, but that's not the bread and butter.”

Enterprise software value does not come from sounding human. It comes from an agent's ability to pull order records, update billing tables, and resolve database actions without human intervention. Audio realism gets a user to stay on the phone for ten seconds. Tool execution finishes the job.

Long term, Weitzman expects voice to replace visual software interfaces entirely: “Human-computer interface is going to become primarily voice as opposed to a screen. So part of the reason why Google succeeded is it is a very simple interface. There's a text box and a button. That's it.”

What to Do With This

Audit your voice agent roadmap this week. If you are building customer-facing voice tools, stop trying to train your own text-to-speech models from scratch. Instead, buy voice APIs from ElevenLabs, Cartesia, or Speechify, and spend all your engineering hours writing deterministic integrations into your core database so your agent can actually execute tasks.