YouTube Deep Summary
Extract content that makes a tangible impact on your life
📝 Transcript (155 entries):
eastern and streaming on CNBC plus. >> Tech giants like Microsoft and Google are outsourcing more
and more coding to AI in a productivity push. But some new research shows the tools might not be as
helpful as some expect. Deirdre Bosa is digging into that for today's tech check. Happy Friday.
>> D Happy Friday. Good morning Carl. So this is some cold water poured on the AI productivity hype.
Researchers at meta, which is an AI nonprofit research firm, ran a real world trial and found that
seasoned engineers were actually 19% slower when using AI tools like cursor. Instead of speeding
them up, the AI often gave suggestions that looked helpful, but actually required time consuming
corrections.
Now, this undercuts a core Wall Street narrative that AI will supercharge white collar efficiency
and unlock a wave of productivity gains across the enterprise. Instead, the study suggests that the
return on AI coding it may be more uneven, less immediate than investors have priced in. Now, there
is some nuance here. Prior studies from meta have shown a more straightforward benefit to junior
engineers from AI tools, particularly for simpler, well scoped tasks. Now, this latest suggests that
while it can help that group level up, it may actually be increasing reliance on senior talent
because someone still needs to debug, refine, and ship the final product.
So that helps explain the current talent wars where Zuckerberg is throwing $100 million offers at
top AI engineers. They're more essential than ever. Meanwhile, new data shows that AI adoption
appears to be stalling more broadly. Ramp this is a platform that tracks enterprise software spend.
It shows that paid AI tool usage was flat at about 40% after a very steep run up over the past year.
Now, in the most aggressive sectors for adoption, tech and finance, there was actually a slight
pullback. I spoke yesterday to CEO Eric Lyman, who told me that companies are trying these tools.
They're not always working, and so they're asking eventually, where's the value. At the same time,
though, he says that the pullback comes after explosive adoption at the start of 2023. He says maybe
1 in 20 companies were using AI tools today.
It's almost one out of two companies on ramp that are spending on them now. When you look across the
tech giants themselves, AI coding, it is already embedded. Google and meta say that around 50% of
their code is now written by AI. So the takeaway here, the tools, they're certainly being used.
They're here.
The payoff may just be more uneven than the hype suggests, and perhaps it plateaus at a certain
point, justifying those huge paychecks for the most senior research analysts. >> What's really
interesting, though. >> Is. >> I mean, if you have these companies that go in and say, yeah, but our
engineers are spending all of this time going back in and revising the suggestions from the AI
tools, does that feedback make it back to the providers of the technology and therefore the kinks
get worked. Maybe it's by AI, but the kinks get worked out, and that's the very feedback that helps
them.
Maybe not this year, but next year and year after. Improve that performance. >> That's a great
point, right. The AI is also learning from these senior engineers that are debugging that are
refining the algorithm. So maybe it does get better.
And I think that's why you see this huge push for AGI or superintelligence, the idea that eventually
the models are smarter than even humans. So they can do 90 or even 100% of the coding. I think the
takeaway here, too, is, though, that these tools are used at different levels. For junior engineers,
they can be really useful. It's the engineers who go into a new job and are able to prompt that
levels them up.
But at the very top, when you know the code source so well and your senior engineer, you're spending
a lot of time rechecking. So maybe it's the middle that. >> Gets squeezed. And the point that we
were making last hour, we were actually talking to Deirdre about whether the openness to change is
different when you're a senior engineer than when you're a junior. Junior engineers may just be also
more open to using those kinds of tools in their job so that we're watch it.
That's dangerous. >> Too, though, right. Because the junior engineers may be accepting code that
isn't 100% correct or, you know, creating more. Work later on for the senior engineers if that code
hasn't been checked. >> Early on.
Nobody ever thought of that.