Luis von Ahn, co-founder and CEO, took to LinkedIn on Thursday to walk back a previous stance pushing AI use over human employees.

“To be clear: I do not see AI as replacing what our employees do (we are in fact continuing to hire at the same speed as before),” he wrote. “I see it as a tool to accelerate what we do, at the same or better level of quality. And the sooner we learn how to use it, and use it responsibly, the better off we will be in the long run.”

He added, “No one is expected to navigate this shift alone. We’re developing workshops and advisory councils, and carving out dedicated experimentation time to help all our teams learn and adapt.”

The clarification is a 180-degree turn from the company’s position a week ago, when it declared it would “gradually stop using contractors to do work AI can handle,” evaluate AI fluency in workers’ annual reviews, and only add new employees “if a team cannot automate more of their work.” 

Von Ahn also appeared to throw his weight behind AI over human teachers in a podcast appearance. Speaking on No Priors with Sarah Guo, he predicted that AI would soon be able to teach any subject, at a greater scale, and create “better learning outcomes” than human teachers, but added that schools would continue to exist “because you still need childcare.” 

The criticism flew in. On the company’s popular TikTok and Instagram accounts, commenters piled on to bash AI on every recent post. (On one video where a baby owl plushie asked “mama, may I have cookie,” the top comment read: “mama may I have real people running the company 💔”)  The company even put von Ahn in his own TikTok, opposite a masked, hoodie-wearing person to explain that “AI will allow us to reach more people.”

A Duolingo spokesperson told Fortune: “We’re still growing our team, and we’re training and developing our talent so they benefit from using AI.” He added, “All AI content is created under the direction and guidance of our learning experts. We have rigorous quality standards in place to ensure that any content we publish is safe, accurate and aligned with the CEFR,” referencing an international standard to measure language ability.

Startups curb their enthusiasm

Duolingo’s self-correction is just the latest in a recent trend. 

Fintech app Klarna had its own turnaround on AI last month. After publicly touting the superiority of its AI chatbot, saying it hadn’t hired humans in a year, the company’s CEO revealed that the “lower quality” of the chatbot meant it would start hiring humans again after all. 

Shopify faced similar criticism after a memo essentially said that AI-driven productivity would replace new hires.

The backlash to Duolingo is the latest evidence that “AI-first” tends to be a concept with much more appeal to investors and managers than most regular people. And it’s not hard to see why. Generative AI is often trained on reams of content that may have been illegally accessed; much of its output is bizarre or incorrect; and some leaders in the field are opposed to regulations on the technology. 

But outside particular niches in entry-level white-collar work, AI’s productivity gains have yet to materialize. An IBM survey of 2,000 leaders found that 3 in 4 AI initiatives fail to deliver their promised ROI. A recent National Bureau of Economic Research study of 25,000 workers in AI-exposed industries found that the technology didn’t make workers massively more productive and had next to no impact on earnings as well as hours.

That “this tool that’s been adopted so fast, where the expectations are so high, [was] not making a difference in earnings was a surprise to me,” University of Chicago economics professor Anders Humlum, one of the NBER study authors, told Fortune.

“It seems it’s a much smaller and much slower transition than you might imagine if you had just studied the technology’s potential in a vacuum.”