One red-hot category in the generative AI space is customer support, which isn’t surprising, really, when you consider the tech’s potential to cut contact center costs while increasing scale. Critics argue that generative AI-powered customer support tech could depress wages, lead to layoffs, and ultimately deliver a more error-prone end-user experience. Proponents, on the other hand, say that generative AI will augment—not replace—workers while enabling them to focus on more meaningful tasks.

Jesse Zhang is in the proponent camp. Of course, he’s a little biased. Along with Ashwin Sreenivas, Zhang co-founded Decagon, a generative AI platform to automate various aspects of customer support channels.

Zhang is well aware of how stiff the competition is in the market for AI-powered customer support, which spans not only tech giants like Google and Amazon but also startups such as Parloa, Retell AI, and Cognigy (which recently raised $100 million). By one estimate, the sector could be worth $2.89 billion by 2032, up from $308.4 million in 2022.

But Zhang thinks that both Decagon’s engineering expertise and go-to-market approach give it an advantage. “When we first started, the prevailing advice we received was to not pursue the customer support space because it was too crowded,” Zhang told TechCrunch. “Ultimately, the thing that worked for us was to aggressively prioritize what customers wanted and maintain laser focus on what customers would get value from. That’s the difference between a real business and a flashy AI demo.”

Both Zhang and Sreenivas have technical backgrounds, having worked at both startups and larger tech organizations. Zhang was a software engineer at Google before becoming a trader at Citadel, the market-making firm, and founding Lowkey, a social gaming platform that was acquired by Pokémon GO maker Niantic in 2021. Sreenivas was a deployment strategist at Palantir before co-founding computer vision startup Helia, which he sold to unicorn Scale AI in 2020.

Decagon, which sells primarily to enterprises and “high-growth” startups, develops what amounts to customer support chatbots. The bots, driven by first- and third-party AI models, are fine-tuned and capable of ingesting businesses’ knowledge bases and historical customer conversations to gain a greater contextual understanding of issues.

“As we started building, we realized that ‘human-like bots’ entails a lot since human agents are capable of complex reasoning, taking actions, and analyzing conversations after the fact,” Zhang said. “From talking to customers, it’s clear that while everyone wants greater operational efficiency, it cannot come at the expense of customer experience—no one likes chatbots.”

So how aren’t Decagon’s bots like traditional chatbots? Well, Zhang says they learn from past conversations and feedback. Perhaps more importantly, they can integrate with other apps to take actions on behalf of the customer or agent, like processing a refund, categorizing an incoming message, or helping write a support article.

On the back end, companies get analytics and control over Decagon’s bots and their conversations.

“Human agents are able to analyze conversations to notice trends and find improvements,” Zhang said. “Our AI-powered analytics dashboard automatically reviews and tags customer conversations to identify themes, flag anomalies, and suggest additions to their knowledge base to better address customer inquiries.”

Now, generative AI has a reputation for being, well, less than perfect—and, in some cases, ethically compromised. What would Zhang say to companies wary that Decagon’s bots will tell someone to eat glue or write an article full of plagiarized content, or that Decagon will train its in-house models on their data?

Basically, he says, don’t worry. “Providing customers with the necessary guardrails and monitoring for their AI agents has been important,” he said. “We optimize our models for our customers, but we do this in a way that ensures that it is impossible for any data to be inadvertently exposed to another customer. For instance, a model that generates an answer for customer A would never have any exposure to data from customer B.”

Decagon’s tech, while subject to the same limitations as every other generative AI-powered app out there, has been attracting name-brand clients as of late, like Eventbrite, Bilt, and Substack, helping Decagon reach break-even. Notable investors have climbed aboard the venture, too, including Box CEO Aaron Levie, Airtable CEO Howie Liu, and Lattice CEO Jack Altman.

To date, Decagon has raised $35 million across seed and Series A rounds that had participation from Andreessen Horowitz, Accel (which led the Series A), A*, and entrepreneur Elad Gil. Zhang says that the cash is being put toward product development and expanding Decagon’s San Francisco-based workforce.

“A key challenge is that customers equate AI agents to previous-generation chatbots, which don’t actually get the job done,” Zhang said. “The customer support market is saturated with older chatbots, which have eroded consumer trust. New solutions from this generation must cut through the noise of the incumbents.”