When Microsoft released its Work Trends report last week, it identified the emergence of something it termed the “frontier firm,” a company with advanced AI deployment and maturity in possession of a “belief that agents are key to realizing ROI on AI.” (View Highlight)
Fewer than 3% of those polled for the report said they worked at such a firm.
It seems that Duolingo, the language-learning app, is interested in including itself on that short list. (View Highlight)
In an internal email sent by CEO Luis von Ahn (and published on Duolingo’s LinkedIn page), von Ahn said that Duolingo is officially going to be “AI-first.”
• In 2012, he said, Duolingo “bet on mobile,” something that allowed it to thrive for the past decade. Now, another “platform shift” is coming, and Duolingo doesn’t want to get left behind.
• AI, according to von Ahn, “isn’t just a productivity boost. It helps us get closer to our mission. To teach well, we need to create a massive amount of content, and doing that manually doesn’t scale. One of the best decisions we made recently was replacing a slow, manual content creation process with one powered by Al. Without Al, it would take us decades to scale our content to more learners. We owe it to our learners to get them this content ASAP.” (View Highlight)
Still, he said that Duolingo “can’t wait until the technology is 100% perfect. We’d rather move with urgency and take occasional small hits on quality than move slowly and miss the moment.”
As part of the shift, Duolingo will stop hiring contractors to “do work AI can handle.” The people Duolingo does hire will be required to use AI, a requirement that will be evaluated during performance reviews. (View Highlight)
It mirrors a recently leaked memo from Shopify CEO Tobi Lutke, a memo that said that effective AI usage is now a “fundamental expectation” of everyone at the company. Lutke went on to say that teams won’t be allowed to hire additional people unless they can prove the additional work cannot be automated.
Still, we’re at a point in the generative AI integration where the benefits of this approach remain decidedly unclear.
Some companies, like Klarna, have tried something like this, and then been forced to reverse course. Others, like Johnson & Johnson, found that the value of the technology is in narrow applications, rather than widespread use. (View Highlight)
This all is somewhat compounded by a recent working paper, published earlier this month from the Becker Friedman Institute for Economics at the University of Chicago, that identified minimal labor market impacts due to generative AI.
• The paper is based on two massive surveys, each of 25,000 people from 7,000 workplaces across Denmark, and employer-employee data regarding wages, earnings and working hours.
• Though the surveys identified time savings associated with chatbots across the board, “users report average time savings of just 2.8% of work hours,” far from a transformation in productivity. (View Highlight)
The researchers estimate that only 3-7% of workers’ already-modest productivity gains “are passed through to higher earnings … the limited impacts of AI chatbots on workers’ earnings reflect a combination of modest productivity gains and weak pass-through to wages, although employer policies can enhance both.”
They added that, “while adoption has been rapid, with firms now heavily invested in unlocking the technological potential, the economic impacts remain small.” (View Highlight)
On top of this real-world study, a team of researchers at Carnegie Mellon University recently conducted an experiment called TheAgentCompany, where they threw state-of-the-art agents into a self-contained digital environment designed to mimic a small software company, to see how well benchmark performance translates to real-world efficacy.
The best-performing agent only managed to complete 24% of its assigned tasks.
The researchers said that “there is a big gap for current AI agents to autonomously perform most of the jobs a human worker would do, even in a relatively simplified benchmarking setting.” (View Highlight)
von Ahn said that “change can be scary.” Lutke said that this “sounds daunting.” When HP fired 27,000 employees to strategically shift to the cloud back in 2012, the language was: “while these actions are difficult … they are necessary.” When Microsoft laid off 18,000 people in 2014 to make that same transition, the language was a little more succinct: “difficult, but necessary.” (View Highlight)
Again, I don’t think many companies will be legitimately successful there. Johnson & Johnson’s approach seems the most likely example that will be emulated: narrow, evidence-based adoption. In other words, using the tech where and when it makes sense to use it, rather than mandating its use in everything, by everyone, all the time.
Reliability issues and security issues, neither of which has been mitigated to a legitimately acceptable degree despite all the benchmark progress over the past two years, are enough on their own to stymie wide-scale enterprise adoption. (View Highlight)
