The AI Daily Brief: Artificial Intelligence News and Analysis - AI Skepticism is CANCELLED
Episode Date: September 11, 2025Wall Street just delivered one of the strongest signals yet that the AI boom is real and accelerating. Oracle revealed a record-breaking $300B cloud deal with OpenAI, sending its stock soaring and res...haping the narrative around AI infrastructure. In today’s episode, we break down why this moment marks the end of the AI skepticism cycle, explore how new coding agents are pushing the frontier of autonomy, and highlight breakthrough research on solving LLM randomness. Together, these stories show why AI’s future is no longer in question—it’s already here.Brought to you by:KPMG – Discover how AI is transforming possibility into reality. Tune into the new KPMG 'You Can with AI' podcast and unlock insights that will inform smarter decisions inside your enterprise. Listen now and start shaping your future with every episode. https://www.kpmg.us/AIpodcastsBlitzy.com - Go to https://blitzy.com/ to build enterprise software in days, not months Robots & Pencils - Cloud-native AI solutions that power results https://robotsandpencils.com/Vanta - Simplify compliance - https://vanta.com/nlwThe Agent Readiness Audit from Superintelligent - Go to https://besuper.ai/ to request your company's agent readiness score.The AI Daily Brief helps you understand the most important news and discussions in AI. Subscribe to the podcast version of The AI Daily Brief wherever you listen: https://pod.link/1680633614Interested in sponsoring the show? nlw@aidailybrief.ai
Transcript
Discussion (0)
Today on the AI Daily Brief, why AI skepticism is officially canceled.
Before that in the headlines, AI moves to production mode.
The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.
Hello, friends, quick announcements before we dive in.
First of all, thank you to today's sponsors, robots and pencils, Superintelligent, and Vanta.
To get an ad-free version of the show, go to patreon.com.
Where ad-free starts at just three bucks a month.
And of course, to learn about sponsorship opportunities, shoot us in
note at sponsors at AIdailybrief.aI. We will send you all the goods, and you can reach this excellent
audience of AI practitioners. With that, though, let's dive in. Welcome back to the AI Daily Brief
Headlines edition, all the daily AI news you need in around five minutes. The theme of today, at least where
we're kicking off, is AI moving into production mode. One of the big things that I think is happening
right now is that all across the industry, we are moving from pilots and experiments, and more
than that as activity, that as mindset, into thinking about what is it going to take to bring these
technologies to production for big, extensive, important use cases. We're going to talk about that
a little bit in the context of agents and how long they can be autonomous in the main episode,
but we kick off with that theme today with the headlines as well, with OpenAI, who have now
added full support for MCP tools within ChatGBT. The company wrote, we've finally added full support
for MCP. In developer mode, developers can create connectors and use them in chat for right actions,
not just search and fetch. Update Jira tickets, trigger Zabior workflows, or combine connectors for
complex automations. Now, on some level, this is a small story. This was always coming. OpenAI had already
committed to this full support, but it now exists. It's a meaningful improvement to the U.S.
And it means users are going to have access to the full suite of connectors and workflows,
regardless of how they access these models. The practical effect of that is going to be more
real use cases moving to production. Next up, YouTube is rolling out their AI audio dubbing feature
to millions of creators on the platform.
Now, this feature has been going through pilots since 2023,
including testing with large creators like Mr. Bees, Mark Robere, and Jamie Oliver.
Video can be automatically dubbed into dozens of languages,
including Hindi, Korean, and Portuguese.
With two years of data, YouTube has reported incredible results
in delivering English-language content to the globe.
They claim that on average,
creators uploading multilanguage audio have seen more than 25% of watchtime
come from views in other languages.
Jamie Oliver's cooking content saw a 3x viewership boost
by making it available in multiple languages.
YouTube's next pilot is testing automated thumbnail translations
to make that content a more natural fit in other languages as well.
Now, between this rollout and Apple's built-in translation for the new AirPods,
it definitely feels like multi-language AI is hitting production quality.
The implied cost reductions behind this broad rollout are also notable.
In 2023, when the technology was being developed,
it may have only made sense to deploy it for videos that get millions of views.
Now it is going to be available to millions of smaller creators.
Hopefully soon you'll be able to listen to the AI Daily Brief in Spanish, Arabic, or Russian.
The Enterprise also continues to get production-grade tools.
Stability AI, for example, has released a new version of their audio model, which they say is up to the task of enterprise-grade sound production.
They write,
Audio influences brand engagement by 86%, but few enterprises are leveraging audio as an extension of their brand,
making customized sound an untapped differentiator.
The company's stable audio 2.5 model is capable of generating full songs within seconds.
It also has a feature the team is called audio in painting, where users can upload a few bars of a track and let the model fill in the rest based on that context.
Stability is also offering to fine-tune the model so companies can dial in the right sound for their brand.
Zach Evans, head of audio research at Stability said, 2.5 isn't just an iteration on 2.0.
It reflects our shift towards enterprise-grade capabilities, professional quality audio, faster performance,
and the advanced control needed for commercial use cases and the multi-step iterative workflows of creative professionals.
Now, control seems key, and this is the same evolution we've recently been seeing with
image and video generation. Creative AI workflows have been moving from essentially trying to get
lucky with a one-shot generation to iterative AI editing, allowing users to hone in on what they need.
All of this adds up to, I think, what we are moving into a much more production-focused era of
AI.
Now, in this particular set of headlines, there were a bunch of things outside this theme as well,
including a couple stories on the regulatory front.
Texas Senator Ted Cruz, perhaps unsurprisingly to you, wants to keep the government's
hands off of AI development. On Wednesday, he introduced a new bill called the Sandbox Act. The bill would
require the White House Office of Science and Technology policy to create a sandbox for AI model
testing with minimal regulatory standards. Companies that participate in the program could then
release AI products under regulatory waivers. Now, notably, this is not just about exempting
companies from AI regulations. The government could move any regulatory barrier as they see fit. For example,
a company working on AI cancer screening software could seek an exemption from HIPAA laws that protect
patient privacy. Waivers could be granted for two years at a time or for up to 10 years in total.
The bill is part of a five-pillar plan that Cruz is set to introduce during Wednesday's
committee hearing, Cruz said, a regulatory sandbox is not a free pass. People creating or using
AI still have to follow the same laws as everyone else. Now, the process of actually getting
to federal regulation is going to be extremely controversial. The verge, for example, characterized
the bill as letting AI companies set their own rules for up to 10 years, and while that's probably
a bit strong, it's going to represent a pretty prominent opinion on how the Sandbox Act would play out
in practice. In the meantime, as the government figures out what it's going to do with AI regulation,
there are many efforts at some sort of self-regulation as well. One recent example is that a group of
leading web publishers have announced a new standard for content licensing for the AI-first
internet. Called really simple licensing or RSL, this new standard builds on top of concepts
like robots.t.TXT and the RSS standards for content feeds. A new nonprofit called the RSL collective
will be the steward for the standard, with support from Reddit, Yahoo, People, Quora, Medium, and O'Reilly.
Said Tim O'Reilly, RSS was critical to the Internet's evolution as an information ecosystem,
giving early online publishers a simple open standard to syndicate their content and reach audiences at Internet scale.
That spirit of openness is what helped the web thrive.
Today, as AI systems absorb and repurpose the same content without permission or compensation,
the rules need to evolve.
RSL builds directly on the legacy of RSS, providing the missing licensing layer for the AI-first Internet.
It ensures that the creators and publishers who fuel AI innovation are not just part of the
conversation but fairly compensated for the value they create. The idea of the new standard
is to allow Robots.txti to go beyond simple yes-no permissions for web scraping to define
a new automated licensing layer for the internet as a whole. The RSL collective writes,
RSL, services supports a range of licensing, usage, and royalty models, including free,
attribution, subscription, paper crawl, and paper inference, i.e. publishers get compensated
every time an AI application uses their content to generate a response.
There are now two distinct approaches to solving the issue of AI crawlers serving up content for free,
with the other being offered by Cloudflare. Cloudflare system includes a proprietary licensing
marketplace that handles the required micropayments and has been pretty controversial. The RSL
collective is betting on AI companies opting into a well-organized standard instead. Now, obviously,
I think this is the beginning and at the end of a conversation, but shows how groups are trying to
figure out how to solve some of the potential negative externalities of AI without waiting around
for the government to solve it for them.
Lastly today, one more on the fundraising front, Perplexity has officially closed their latest round of funding at a $20 billion valuation.
The information reports that Perplexity raised $200 million in fresh capital.
The new round comes shortly after the company's previous round in July, which saw $100 million raised at an $18 billion valuation.
Sources also said that the company is approaching $200 million in ARR, up from $150 million in August.
I.e., the funding environment for leading AI startups is still red-hot and growth isn't showing any signs of slowing down,
which I think is a perfect segue
to our main episode,
why AI skepticism is officially canceled.
AI changes fast.
You need a partner built for the long game.
Robots and pencils work side by side
with organizations to turn AI ambition
into real human impact.
As an AWS certified partner,
they modernize infrastructure,
design cloud native systems,
and apply AI to create business value.
And their partnerships don't end at launch.
As AI changes, robots and pencils
stays by your side, so you keep paying.
The difference is close partnership that builds value and compounds over time.
Plus, with delivery centers across the U.S., Canada, Europe, and Latin America, clients get local expertise and global scale.
For AI that delivers progress, not promises, visit robots and pencils.com slash AI Daily Brief.
As a founder, you're moving fast towards product market fit, your next round, or your first big enterprise deal.
But with AI accelerating how quickly startups build and ship, security expectations are higher earlier than
than ever. Getting security and compliance right can unlock growth or stall it if you wait too long.
With deep integrations and automated workflows built for fast-moving teams, Vanta gets you audit-ready
fast and keeps you secure with continuous monitoring as your models, infra, and customers evolve.
Fast-growing customers like Langchain, writer and cursor trusted Vanta to build a scalable foundation
from the start. And look, as someone who lives in the world of enterprise procurement,
I love how Vanta makes it easy to get compliance right. The last thing you need when you're trying to win
that big deal is to have it scuttled by something that Vanta has solved for over 10,000
companies. Go to Vanta.com slash NLW to save $1,000 today through the Vanta for Startups
program and join over 10,000 ambitious companies already scaling with Vanta. That's VANTA.com
slash NLW to save $1,000 for a limited time. If you are a regular listener, you will have heard
about superintelligence agent ratingness audits at this point. But I wanted to tell you today about
the full suite of agent readiness products that go beyond just the initial readiness report.
Over the last six months, Super Intelligence has built out an entire agent planning suite.
We help you move from discovery to planning to implementation.
After you've completed your agent readiness audits, we help you double-click on your most
important use cases with what we call our use case planning reports.
These reports are going to help you understand what sort of technical preparation you need
to do to be ready for a use case, what challenges you might face in implementation, and whether
you should be thinking about building, buying, partnering, or some combination. After that, you can
even get a spec document in what we call our technical blueprint that gives either your developers
or the developers of the partner you work with what they need to build exactly the agent that
you're looking for. If you want to learn more about superintelligence agent planning suite,
we've built a custom GPT to answer your questions. Just go to bit.ly slash super super super agent.
That's bit.it.l.ly slash super super agent, all one word. And if you have any questions,
questions, the agent can even help you book an appointment with our team.
Welcome back to the AI Daily Brief. For weeks now, there has been this skepticism narrative
surrounding AI. It has had multiple dimensions, obviously the MIT story, the launch of GPT5,
and on Wall Street it's been particularly pronounced. As I've shared, I think at least when it
comes to the market dimension of this, a lot of it has to do with AI taking on the generalized
anxiety of the larger stock market. But boy, yesterday on Wall Street, AI,
skepticism was forcibly removed after Oracle put in one of the most remarkable trading days
in recent memory. Now, we talked already about Oracle's big announcement. On yesterday's show,
we covered how they had reported a backlog of $455 billion in contracts over the next five years.
The stock was up 27% overnight after the announcement and reached as much as 43% up, ultimately
landing at 36% up to reach a new record high. Value at around $600 billion at the beginning of
the week, Oracle is now almost a trillion-dollar company.
So what happened? This was not a story of Oracle wildly beating analyst estimates. In fact, it was a small miss.
Instead, this was all about the future. It was about that $455 billion, in other words, in contracts over the next five years.
By the end of the day, we got more information around who was accounting for all of that demand.
Specifically, we learned that OpenAI was the mystery customer who had contracted for $300 billion last quarter.
The deal begins in 27 and will run over five years.
The contract requires Oracle to deliver 4.5 gigawatts of compute.
This is one of the single largest cloud contracts in history and will require the equivalent
of two Hoover dams to supply the power.
Now, one interesting narrative shift here is the implications of Oracle chairman and founder
Larry Ellison becoming the world's wealthiest man.
Yesterday's rally in Oracle stock pushed Ellison's net worth up about $100 billion,
maybe the biggest single-day gain in wealth in history.
With a net worth of nearly $400 billion now, he's basically neck-and-neck with Elon Musk,
and in fact is likely back to second place by the time you're listening,
but that's a little bit besides the point.
The reason it matters at all is that the title of the world's richest man
has tended to follow the dominant tech category of the time.
Elon Musk first took the title in 2021 when green tech and EVs were the big narrative.
Before that, it was Jeff Bezos on the back of e-commerce.
Prior to Bezos, Bill Gates was the wealth leader from software.
Now the world's richest man is an AI infrastructure baron.
Now, of course, there is still plenty of skepticism out there.
Many of the reports have pointedly noted that open-eastern,
AI's current annual revenue was $10 billion in June. The company, therefore, has a long way to go
before they can afford to cash flow $300 billion in infrastructure spending. Stock analyst Thomas Chua
posted, the entire growth narrative hinges on a cash-burning customer's ability to raise
unprecedented amounts of capital. If OpenAI's funding dries up, Oracle could be left with
massive stranded investments in AI infrastructure and a collapsed growth story. There are also plenty of
people pointing out that AI revenue is looking a little circular, with OpenAI's revenue flowing into
Oracle's balance sheet, which then flows into Nvidia's balance sheet and back in
to GPU financing deals. Shortseller Jim Chanos wrote an entire thread poking holes in the open
AI deal, and yet he didn't declare that he's short Oracle. And that's pretty much where we're
at with AI bears on Wall Street. It is very cheap right now to write a research note about the AI
bubble, but it is starting to get extremely expensive to be stuck on the sideline as AI names
rip. The Kobayisi letter simply pointed out the gigantic green candle writing, as we continue
to reiterate, we are still so early in the AI revolution. Take him, the author of the
NVIDIA Way posted, ridiculous how outlets amplified and platformed that shoddly written MIT paper,
conflated Altman's bubble quote, and are now barely mentioning the biggest AI infrastructure
print since NVIDIA's May 2023. Again, ignore the noise and focus on fundamental reality.
Meanwhile, Wall Street analysts are simply revising their targets up as fast as possible.
Strategists for Wells Fargo, Barclays, and Deutsche Bank all boosted their targets for the S&P 500 this
week, citing resilient earnings and an unrelenting AI investment cycle.
Deutsche Bank now sees another 7% move coming by the end of the year, added to an 11% gain year-to-date.
And importantly, Oracle is the biggest move, but it's by no means the only one.
Broadcom is also up more than 20% over the past month on the back of a $10 billion deal to manufacture chips for OpenAI.
Tiny cloud provider Nebius is up more than 40% this week after signing a $17 billion partnership with Microsoft.
Increasingly, the AI rally is no longer just about the mega-cap tech stocks pushing higher.
there is an entire AI supply chain getting dramatically repriced as deals are put in place to power the next five years of development.
And what's more, as we discuss AI jobs disruption, we're basically completely ignoring all that's going to go into servicing these big deals.
Usually when you have a moment of creative destruction, the destruction comes before the creative.
It's why new technology often faces resistance.
It's easier to see what it disrupts before we understand fully what it creates.
In this case, because of infrastructure, we have the potential to see tens of thousands of new jobs
materialize, entire categories of new skills that become extremely important and high leverage,
very early on in the technology's lifecycle because of the need for infrastructure.
You don't hear much about that because there's still such an impulse towards skepticism
and not wanting to count chickens before they're hatched.
But at least when it comes to the market, it is very clear that for the moment now,
AI skepticism is firmly off the menu.
And I think that if you look around even beyond the markets, there's reasons to think that
this shift is more broad-based as well.
First of all, let's talk about Replit's new Agent 3.
One of the things that was really interesting about the discourse around GPT5 was the extent to which
it associated AI progress with simply the performance of the base model.
Meanwhile, companies are creating entire systems and agents that build on top of models and take
advantage of them that are getting more and more performance, even if it is true that
base model performance is plateauing, which itself is a highly questionable proposition. But the
point is that even if it were plateauing, there are so many other places where we're seeing
gains in actual practical production and performance. Agent 3 from Replit is a major step
up in what autonomous coding agents can do. In fact, they're claiming that it's 10x more autonomous
than their previous agents. The agent is capable of running tests and fixing bugs as it goes.
It can also build other agents and automations, and can be used to build workflows within other
apps. The biggest improvement is a 200-minute maximum runtime. Why Combinators Paul Graham wrote,
it seems a bit counterintuitive at first, but one of the most important tests of an AI is how long it
can continue thinking about something productively. Replet is now up to 200 minutes. CEO Amjad Masad called
this the full self-driving moment of software. One of the big benchmarks for agent autonomy over the
past year has been meter's study of the time horizon for agentic tasks. The study measured agent
performance against tasks that would take a human developer various lengths of time, and graded them
at 50% and 80% accuracy. The study found that the agent time horizon has been doubling, on average
every seven months, and seems to be increasing. Back in July, SWIX wrote, Ambien agents are going
to completely dominate the rest of 2025. Human deep work and focus requires at least one to two hours
uninterrupted. By end of year, all next-gen models will pass the one-to-two-hour autonomy meter barrier,
meaning they will be used in completely different ways than the current one to 15-minute autonomy frontier.
Now, Massad brought up this study in the context of the release of Agent 3, writing,
the meter paper that says that the length of tasks AI can do is doubling every seven months,
radically undersells the scaling that we're seeing at Replit.
It might be true if you're measuring one long trajectory for a single model class,
but this is where an Agent Research Labs Alpha is.
We build multi-agent architecture and use different models from various providers
to tap into their latent abilities across various tasks.
Another way of saying what he's saying is that while the meter paper might be a good benchmark for
general progress of the underlying models, it doesn't do a very good job of measuring the state of
the art as a gentic scaffolding gets more complex. With this release, we're starting to move
beyond coding agents that produce one-time apps and towards continuous AI workers. Gerard Lipscomb,
a developer working with the Replit Stack, posted, one-shining apps was never the test of a
vibe coding platform. The real benchmark is maintaining and improving products over time, while serving
real users without breaking on every new change.
Replit is moving faster in this direction than anyone else I'm aware of.
As someone building my business on Replit, Autonomous App Testing and Automations is music
to my ears.
It's yet another signal that they're building a special kind of full-stack ecosystem.
The kind that one day might allow you to build not just a product, but a business that
runs and improves itself.
This is, I think, exactly what all of the discourse around how much better a base model
is than the number that came before it misses.
It misses the fact that the improvements are coming from these systems.
which are being put together to accomplish ever bigger and more significant groups of work all at once.
Now, the last thing that I want to point to as another indication that we are at the beginning
of a narrative shift once again away from skepticism and towards excitement again, is that
Thinking Machines Lab came out with its first shared research and people are really excited.
On Wednesday, the company put out a paper on the issue of nondeterminism or randomness in AI
outputs. Basically, the idea that you could prompt the same AI two separate times with the
exact same prompt and the exact same model and come up with two very different answers.
This has been treated as just sort of the cost of doing business with LLMs and something that
couldn't necessarily be fixed. However, the team at Thinking Machines Lab, led by Horace He, who
was recruited from Meta, argued that actually the problem here is something which they
call batch invariants. The neuron actually had a really interesting layperson, ELI-5 kind
of analogy. They said, imagine ordering the same coffee at Starbucks, but it tastes different
depending on how many other customers are in line. That's essentially what's happening with AI
models. When an AI server is busy handling lots of requests, it processes them in batches.
Your request gets bundled with others because that's more efficient, and somehow this changes
your specific answer even though it shouldn't. Follow the logic and the busier the server,
the more your results vary. Thinking Machines Labs, or Thinky for short, release their approach as
open source code, which they're calling batch invariant kernels. Now, hold aside all the specifics
here. This is a big, thorny, technical research problem that a lot of people had just given up on,
And so the fact that this lab dropped a solution in a way that anyone else can interact with it has, I think, jogged people into remembering that there's actually still so much exciting science and research to be done that will fundamentally improve what we get out of AI.
So summing this up, yesterday we got one of the biggest days in stock history because of a massive AI infrastructure deal and a halo effect that came over into other AI stocks as well.
We got a brand new agent that totally pushed the frontiers of how much autonomy an agent could actually behave with.
and we got some really exciting new science.
That is why I am saying officially AI skepticism is canceled.
RIP, hope you enjoyed your month in the spotlight.
For now, that's going to do it for the AI Daily Brief.
Appreciate you listening or watching as always.
And until next time, peace.