Last week, Salesforce announced a fully autonomous AI worker for customer service: the Einstein Service Agent.
The autonomous agent leverages generative AI (GenAI) to handle digital customer queries of varying complexities without the need to pre-program responses.
To do so, contact centers must connect the Service Agent to their trusted knowledge bases, data stores, and relevant external sources.
Built on the Einstein 1 Platform, the solution is currently in pilot, but Salesforce has teased that it will reach general availability before the year’s end.
During the latest episode of CX Today’s Big News Show, Rebecca Wetteman, CEO & Principal Analyst at Valoir, underlined the move’s significance.
"This marks the shift from humans in the loop to humans at the helm," she said. "Salesforce is signaling its confidence in its AI capabilities by making these agents autonomous."
That confidence is relatively newfound. After all, only six months ago, Marc Benioff, CEO of Salesforce, told Bloomberg:
With the current technology, it’s all about augmentation. We can do a lot… [so] long as the human remains in the loop because, as we all know, these things (LLMs) lie pretty bad.
Since then, many high-profile autonomous AI failures have hit the headlines, including the Air Canada, DPD, and New York City Council incidents.
So, while Salesforce may now appear confident in going all-in on GenAI by announcing autonomous agents, most customers will remain cautious and require additional assurances that the technology is reliable before deploying it.
As such, Salesforce must further educate the market about why this shift makes sense now, which it may aim to do at its upcoming Dreamforce event.
With Salesforce, Autonomous AI Failures Are Less Likely
As the incidents at Air Canada, DPD, and New York City Council remind us, autonomous AI requires a sophisticated knowledge management strategy.
Indeed, the quality of Salesforce's customers' knowledge and data stores will most impact the success of autonomous agents.
Also, the New York City incident exemplifies the need for a fallback, so when the knowledge isn’t there, the virtual agent doesn’t make it up but transfers the customer to a human agent.
Thankfully, the Einstein Service Agent offers this capability, making similar AI failures less likely. As Liz Miller, VP & Principal Analyst at Constellation Research, stated:
The autonomous nature of these bots isn't just about handling conversations but also about recognizing when to transfer the customer to a human agent.
That’s critical, as it also demonstrates Salesforce’s commitment to integrating AI in a supportive, rather than fully autonomous, role within contact centers.
Additionally, Salesforce can flaunt this AI reasoning – alongside its natural responses and investment in guardrails, templates, and low-code tools to ease deployments – as a core differentiator.
