Latest / AI Fire Daily / 🎙️ EP 270: The AI Job Market Contraction & The Video Moat
Transcript
- 0:00The future of work isn't a what -if anymore.
- 0:02It is happening right now. Technical writers
- 0:04and DJs are actively being replaced today. Intro
- 0:08music. Welcome to this deep dive. I am really
- 0:12glad you are here with us. We are looking at
- 0:15a fascinating shift today. Our mission is to
- 0:17explore the displacement and the rapid development
- 0:19of the global AI economy. Yeah, we've got some
- 0:22incredibly striking new data to parse. We are
- 0:25looking at real job losses. We are seeing...
- 0:27insane AI capital deployment. And we are analyzing
- 0:31a surprising new battleground. It's an area where
- 0:34the U .S. is actively losing ground to China.
- 0:37It's a profound moment in human history. We're
- 0:39trying to really understand this transition.
- 0:41So we want to stay clear and grounded. We're
- 0:43going to look at the raw facts. Exactly. We have
- 0:45to follow the data. Right. Let's start by examining
- 0:48the labor market. The Bureau of Labor Statistics
- 0:49just released new data. It covers May 2024 to
- 0:53May 2025. Right. And the numbers are incredibly
- 0:56revealing. They really are. They tracked 18 specific
- 0:58occupations. These are roles that are highly
- 1:00exposed to AI. That represents about 10 million
- 1:03workers in the U .S. economy. Which is a huge
- 1:05slice of the workforce. It is. Now, overall,
- 1:08U .S. employment actually grew by 0 .8 % in that
- 1:11time frame. But those 18 AI -exposed fields,
- 1:15they saw a 0 .2 % drop. That divergence is the
- 1:19real story here. The broader economy is growing.
- 1:21But this specific slice. Yeah. And it gets even
- 1:25starker when you dig deeper into the numbers
- 1:28because the health care sector is booming right
- 1:30now. Right. Health care is just pulling everything
- 1:32up. Exactly. It's lifting the averages. If you
- 1:34remove health care from the equation entirely,
- 1:37the reality shifts. Employment in the other 17
- 1:40AI exposed fields actually fell 1 .6 percent.
- 1:45And that for the second consecutive year. Let's
- 1:48pause and really think about that. A 1 .6 % drop
- 1:51across 17 massive industries. Yeah. It is not
- 1:55a sudden collapse. It is a slow, steady contraction.
- 1:58We can look at the specific roles affected to
- 2:01really understand the mechanism here. I think
- 2:03that's where it gets real for people. Definitely.
- 2:05Technical writers saw an 18 .1 % decline in just
- 2:08one single year. That is a staggering number.
- 2:12Nearly one in five technical writing jobs. Just
- 2:16vanished. Gone. Think about what a technical
- 2:18writer actually does all day. They talk to engineers.
- 2:21They read raw software code. Then they synthesize
- 2:25that into readable manuals for end users. It
- 2:28makes complete sense when you understand the
- 2:30underlying technology. Large language models
- 2:33excel at synthesizing complex, structured information.
- 2:37Right. That is exactly what they do best. They
- 2:39can ingest an entire GitHub repository in seconds.
- 2:42They output perfectly formatted documentation
- 2:44instantly. Humans simply cannot compete with
- 2:48that raw speed. No, they can't. It's just not
- 2:50physically possible. And it's not just tech writing.
- 2:52We also saw radio announcers and DJs drop by
- 2:5511 .1%. Voice cloning technology has become practically
- 2:59flawless over the last year. Local radio stations
- 3:02are operating on razor thin margins anyway. Right.
- 3:04They're desperate to cut costs. So they're automating.
- 3:07their entire daily broadcasts. Graphic designers
- 3:09also fell 7 .7%. Yeah. Generative image tools
- 3:13are handling the heavy lifting for basic corporate
- 3:16layouts now. You don't need a designer for a
- 3:18simple flyer anymore. We also need to look closely
- 3:21at customer support. This is a massive segment
- 3:24of the global workforce. Huge. Customer support
- 3:27departments shed over 130 ,000 jobs. That is
- 3:31a 4 .8 % drop in a single year. conversational
- 3:35AI tools are completely taking over. Right. And
- 3:38it isn't just basic, frustrating chat bots anymore.
- 3:40People need to realize these are sophisticated,
- 3:43autonomous agents. They plug directly into inventory
- 3:46databases. They handle complex customer routing,
- 3:49authorized refunds, and they even negotiate shipping
- 3:52instantly. It's remarkable. But we should clarify,
- 3:54it isn't bad news across the board. We are seeing
- 3:56something researchers call the jagged frontier.
- 3:59The Jagged Frontier is a deeply fascinating concept.
- 4:02I really love this framework. It describes an
- 4:05uneven landscape of AI capability. AI is terrible
- 4:08at some highly complex human tasks. It completely
- 4:11fails at basic logic puzzles sometimes, but it
- 4:14is superhuman at routine, structured, repetitive
- 4:16tasks. Exactly. It's highly situational. In fields
- 4:20where AI acts as a collaborative assistant, we
- 4:24are actually seeing a very healthy transition.
- 4:27Right. It is shifting human talent toward complex
- 4:30work. We are seeing more strategy driven and
- 4:33empathetic roles emerge. It's like water finding
- 4:35the cracks in a foundation. AI is eroding the
- 4:39routine, repetitive tasks first, forcing human
- 4:42talent to seek higher, more complex ground. That
- 4:45water analogy is spot on. It really shows how
- 4:48the erosion is silent. You don't realize the
- 4:50foundation is gone until the entire house shifts.
- 4:52Yeah, the simple copy pasting tasks are being
- 4:54completely washed away. But the high level architecture,
- 4:57that remains human. When you really look at these
- 4:59DLS numbers, it's not just a dip. We are watching
- 5:02a fundamental restructuring of work. So this
- 5:05isn't a massive cliff, but a forced reallocation
- 5:07of skills. Exactly. The repetitive work is vanishing,
- 5:10but the high -level strategic work is expanding
- 5:13rapidly as a direct result. Right. Routine jobs
- 5:16shrink, but complex empathetic roles grow. And
- 5:19that shifting labor force brings us directly
- 5:21to the machinery replacing it. If human labor
- 5:24is shifting this dramatically, we have to follow
- 5:27the money. The scale of capital deployment right
- 5:30now is truly mind -bending. It is hard to even
- 5:34comprehend the numbers involved. We are seeing
- 5:36massive historic investments in raw compute power.
- 5:40Yeah. Just 34 top AI startups now generate $80
- 5:44billion in annualized revenue. Let me repeat
- 5:47that for context. $80 billion. That marks a 112
- 5:51% increase from just six months ago. That is
- 5:54unreal. The growth curve is practically vertical
- 5:56right now. And the market is incredibly concentrated
- 5:59at the very top. OpenAI and Anthropic alone capture
- 6:0389 % of that enterprise revenue. It is a massive
- 6:06duopoly in the enterprise space. Companies are
- 6:08throwing unprecedented amounts of money at these
- 6:10models. They really are. Salesforce is a prime
- 6:12example of this spending trend. They are projected
- 6:15to spend $300 million on Anthropic tokens alone.
- 6:18Two sec silence. $300 million. Just on API calls.
- 6:22That shows an incredible level of enterprise
- 6:24trust in the system's output. Mark Benioff actually
- 6:27called the impact unprecedented for his engineering
- 6:30teams. They are using autonomous AI coding agents
- 6:34across their entire software stack. They're seeing
- 6:37massive productivity boosts across the board.
- 6:40The code generation is just insanely fast. And
- 6:43the models are getting cheaper and smarter simultaneously.
- 6:46GPT 5 .5 just hit a major engineering milestone.
- 6:49It became the first AI model to solve a real
- 6:52program bench coding instance. Yeah. And for
- 6:56anyone unfamiliar, ProgramBench isn't just a
- 6:58simple code riddle. No, it's not. It is a benchmark
- 7:00that mimics massive, complex, real -world software
- 7:03engineering environments. And GPT -5 .5 fully
- 7:07beat Opus 4 .7 across every single metric. Wow.
- 7:10Plus it costs three times less than some specific
- 7:12runs. The capability of these coding agents is
- 7:15staggering. I still wrestle with PromptRift myself,
- 7:17so I can't imagine setting loose autonomous coding
- 7:20agents. It takes a huge leap of faith in the
- 7:22system architecture. especially when the meter
- 7:24is constantly running on API costs. Yeah, that's
- 7:28the scary part. There was a wild story about
- 7:30OpenClaw recently. The creator reportedly burned
- 7:33through $1 .3 million in OpenAI API credits.
- 7:38Oh, my God. And that was just in a 30 -day window
- 7:41running 100 codex agents. Whoa, imagine scaling
- 7:44to a billion queries without realizing the meter
- 7:47is running. That is an incredibly expensive lesson
- 7:50in runaway automation. You script an agent to
- 7:52fix a bug and it loops 10 ,000 times a minute.
- 7:55Exactly. It just burns cash. But that is the
- 7:57chaotic side of this frontier. We see a stark
- 8:00contrast when we look at global public adoption.
- 8:03Some places are trying to smooth out that friction.
- 8:05Right. Europe's first nationwide chat GPT plus
- 8:09giveaway just happened. Yeah. Malta is now offering
- 8:11every single. citizen free access. They give
- 8:14them ChatGPT Plus and Copilot entirely for free.
- 8:16Which is huge. The citizens just have to complete
- 8:18a basic AI literacy course first. It is a brilliant
- 8:21approach to top down national education. But
- 8:24in the U .S., that friction is being smoothed
- 8:26out by consumer hardware instead. Apple's Siri
- 8:29revamp includes deep ChatGPT style conversations
- 8:32directly in the operating system. But they are
- 8:35taking a very different approach to user privacy.
- 8:38Yeah. They're introducing automatic deletion
- 8:40settings right out of the gate. Right. You can
- 8:42erase your chat history after 30 days, one year
- 8:45or keep them forever. They are trying desperately
- 8:48to reduce the friction of adoption. But despite
- 8:51all this enterprise investment, there is deep
- 8:54public skepticism. Oh, absolutely. We are seeing
- 8:56a massive. anti -AI wave forming in the culture.
- 9:00The public relations crisis is growing rapidly.
- 9:03New polls show over 70 % of Americans now view
- 9:06AI negatively. Wow. That is up sharply from just
- 9:10three years ago. People are feeling deeply overwhelmed
- 9:13by the rapid changes. They see the tangible job
- 9:15losses we discussed earlier. Yeah. And they see
- 9:17viral videos highlighting the inherent weirdness
- 9:20of AI. Like that viral Dunder Mifflin video?
- 9:23Someone turned Claude into an intern at the office.
- 9:26I saw that. It showed the AI surviving its chaotic
- 9:29first day in a corporate environment. It fits
- 9:32the corporate absurdity perfectly. It is funny,
- 9:35but it highlights a much deeper underlying anxiety.
- 9:38People feel like they're losing control of their
- 9:40own workflows. Are enterprise budgets outpacing
- 9:43public trust? Undoubtedly. Companies are spending
- 9:46billions to automate operations, while everyday
- 9:49people are increasingly terrified of the societal
- 9:51consequences. Exactly. Massive corporate spending,
- 9:54but huge public skepticism remains. Sponsor.
- 9:58Welcome back. We've been discussing the text
- 10:01and coding domains. The U .S. labs clearly dominate
- 10:04those specific technical areas, but there is
- 10:07a massive blind spot emerging on the global stage.
- 10:09We really need to talk about the video -bounded
- 10:11mode. This is where the geographic narrative
- 10:14completely flipped. China has effectively left
- 10:16the U .S. behind in AI video generation. We are
- 10:19simply reporting the technical data here. The
- 10:21latest benchmarks show Chinese models vastly
- 10:24outperforming Western rivals. Yeah. We are looking
- 10:26at specific models like Stevens 2 .0 and Kling.
- 10:30They're consistently beating US models in photorealism
- 10:33and prompt adherence. They also have vastly superior
- 10:36camera stability during complex movement. This
- 10:38became obvious when OpenAI made a major strategic
- 10:41decision. They quietly shut down their highly
- 10:44anticipated Sora model. That sudden exit. left
- 10:48the field wide open for international competitors.
- 10:51OpenAI killed Sora because of massive infrastructure
- 10:54overhead. Right. Generating pixels is exponentially
- 10:58harder than generating text tokens. It simply
- 11:00didn't make commercial sense for them to run
- 11:02it at scale. Video generation requires an astonishing
- 11:06amount of raw compute. But more importantly,
- 11:09it requires an infinite amount of proprietary
- 11:11training data. That data is the true moat we
- 11:15are discussing here. ByteDance owns TikTok and
- 11:17CapCut. Kuaishu operates massive short video
- 11:20platforms all across Asia. They possess proprietary
- 11:23training libraries that American labs simply
- 11:25cannot replicate. Think about what is actually
- 11:27on TikTok. They have billions of hours of unstructured
- 11:29human motion. Yeah. They have every conceivable
- 11:32camera angle, weird lighting condition and physical
- 11:34interaction. U .S. companies simply do not have
- 11:37that raw physical data to train their physics
- 11:40engines on. And the Chinese companies are moving
- 11:42incredibly fast to monetize this. Kuaishu's Kling
- 11:46model is scaling at an unprecedented rate globally.
- 11:49Right. They're actually exploring a standalone
- 11:52public listing right now just to fuel that expansion.
- 11:55We also have to look at the user experience driving
- 11:57this shift. American tools constantly throw ambiguous
- 12:01errors during the generation process. They actively
- 12:04block pumps due to extremely strict safety guardrails.
- 12:09It creates a deeply frustrating experience for
- 12:11media professionals. Imagine an ad agency trying
- 12:14to render a dynamic car commercial. An American
- 12:18model might block a fast -tracking shot with
- 12:20motion blur. It might trigger a vague safety
- 12:22policy about glorifying reckless driving. Meanwhile,
- 12:26the Chinese models easily handle aggressive camera
- 12:28angles. They process high -speed motion without
- 12:30any hesitation or arbitrary blocking. They also
- 12:34handle complex audio sync flawlessly, matching
- 12:36lip movements to generated speech. Western alternatives
- 12:39simply cannot match their temporal consistency
- 12:41right now. The characters in Western models tend
- 12:44to melt or change clothes between camera cuts.
- 12:47Let's clarify what temporal consistency actually
- 12:49means. Keeping characters and backgrounds looking
- 12:52exactly the same frame by frame. Right. Early
- 12:56AI video was a morphing, hallucinating mess.
- 12:59If someone turned their head, they might grow
- 13:01an extra ear. Yeah, it was terrifying. Now, Kling
- 13:04can generate a character walking through a crowded
- 13:06street, and their jacket stays perfectly solid
- 13:09the entire time. This performance gap is causing
- 13:11a mass - Wow. And Western companies are eagerly
- 13:25paying it. They desperately need these tools
- 13:27to stay competitive in commercial media production.
- 13:29If the American models block their creative workflows
- 13:32with safety filters, they will simply move their
- 13:35budgets abroad. I want to examine the root cause
- 13:37here. Is the U .S. losing simply because of safety
- 13:40ethics or is it fundamentally a lack of raw proprietary
- 13:43video data? It's both. But the proprietary data
- 13:47is the insurmountable wall. TikTok's endless
- 13:50stream of human movement provides training examples.
- 13:53U .S. labs simply cannot scrape from the public
- 13:56web. superior performance and fewer restrictions
- 13:59will inevitably win out. So we have talked a
- 14:01lot about massive corporate moats today. We have
- 14:04talked about $80 billion monopolies and international
- 14:07data wars. Right. It can feel very alienating
- 14:10for the individual listener trying to navigate
- 14:12this space. It is incredibly important to bring
- 14:15this back down to the micro level. Despite these
- 14:17massive geopolitical shifts, independent developers
- 14:20are still building. Yeah, they really are. Individual
- 14:23creators are getting incredibly powerful new
- 14:25tools. The individual toolkit has honestly never
- 14:28been more robust. There's a whole new wave of
- 14:31empowered AI tools hitting the market. They allow
- 14:34individuals to build their own highly customized
- 14:37workflows without relying on massive studio infrastructure.
- 14:41Well, if you get a few specific examples today
- 14:43to see how this works in practice. Vago is a
- 14:45tool specifically for generating narrative videos.
- 14:48It solves the exact continuity problem we were
- 14:51just discussing. It generates consistently compelling
- 14:53narrative videos. But the magic is that it does
- 14:56it without the annoying prompt engineering. Oh,
- 14:58that's huge. Normally, you have to write a massive,
- 15:01complex prompt for every single scene to maintain
- 15:05your aesthetic. Vivago ensures every single scene
- 15:08stays on brand automatically. Then there is the
- 15:11audio space, which is evolving just as fast.
- 15:14SUN to Spotify is a really interesting new distribution
- 15:17platform. Yeah, I love this one. You can create
- 15:20complete AI podcasts and audiobooks entirely
- 15:22from scratch. And you can publish them directly
- 15:25to your Spotify library. The pain point of managing
- 15:28RSS feeds and complex audio distribution pipelines
- 15:31is gone. That saves so much time. It is instantly
- 15:34ready for streaming or offline listening. It
- 15:36bypasses the traditional publisher networks completely.
- 15:39We're also seeing foundational tools for pure
- 15:42developers. The Files SDK is a unified storage
- 15:45system that is gaining huge traction. Right.
- 15:47It works seamlessly for both object and blob
- 15:50backends. It gives developers an escape hatch.
- 15:53Normally, you are locked into whatever backend
- 15:56cloud storage system you use. But this standardizes
- 16:00the inputs and outputs, meaning developers aren't
- 16:02trapped in a single corporate ecosystem. It uses
- 16:05standard web I .O. Nice. It gives developers
- 16:08total control over their data architecture. And
- 16:10for web designers, there's Kirky. WordPress still
- 16:13powers nearly half the Internet. Kirky gives
- 16:15those WordPress designers a free -form, infinite
- 16:18canvas to work on. You can design layouts freely.
- 16:21Without any imposed rigid block structure. It
- 16:24breaks the old UI constraints. Right. Comes with
- 16:26over 100 template kits. And the best part is
- 16:29it is completely free for all users. It's like
- 16:31stacking Lego blocks of data. You pick the exact
- 16:34tool you need to build your custom workflow.
- 16:37That is the real beauty of this current era.
- 16:39The big foundational models provide the massive
- 16:42raw compute engine. But these smaller open source
- 16:45tools. give you the steering wheel to actually
- 16:47drive it. It completely democratizes access to
- 16:50incredibly complex systems. Do these niche tools
- 16:53decentralize power back to the user? Yes, absolutely.
- 16:57They allow a single talented person to execute
- 17:00ideas that used to require an entire studio or
- 17:03development team. Got it. Big tech builds the
- 17:06engine. We drive the car. We have covered a massive
- 17:09amount of ground today. We started by looking
- 17:11closely at the hard reality of job losses. Yeah,
- 17:14we did. The global AI economy is actively...
- 17:18tangibly rearranging the labor market. We saw
- 17:2018 AI -exposed fields shrink while the broader
- 17:23U .S. economy actually grew. We watched technical
- 17:26writers and customer support agents face real
- 17:28immediate displacement as their core workflows
- 17:31were automated. We then explored the $80 billion
- 17:33enterprise value explosion. We saw massive U
- 17:36.S. dominance in text and coding generation driven
- 17:39by historic corporate spending. But we also identified
- 17:42stark geographical lines being drawn right now.
- 17:45The U .S. is hitting a massive wall in... video
- 17:47generation, Chinese tech giants are leveraging
- 17:50their incredibly deep proprietary data moats.
- 17:53They're defining the absolute next frontier of
- 17:56media creation. The technological landscape is
- 17:58shifting faster than anyone predicted. I want
- 18:00to thank you for joining us on this deep dive.
- 18:03We always aim to look past the hype and really
- 18:06examine the underlying data. Absolutely. The
- 18:08data always leads to deeper, more human questions.
- 18:11If Chinese models become the undisputed global
- 18:14standard for generating video and media, How
- 18:17will that subtle shift shape the cultural narratives
- 18:19we consume every single day? Outiero Music.