Shawn Chauhan
@shawnchauhan1.bsky.social
I write interesting things about AI, Startups, Founders, and the Future of Work | Generative AI Consultant | Building Something Big...
Pinned
Shawn Chauhan
@shawnchauhan1.bsky.social
· Dec 22
Most companies try to save themselves by adding more.
LEGO saved itself by ruthless subtraction.
They shut down theme parks, clothing lines, and video games and went from $800 Million in debt to an $18 Billion dollar company.
Here's exactly how they did it:🧵
LEGO saved itself by ruthless subtraction.
They shut down theme parks, clothing lines, and video games and went from $800 Million in debt to an $18 Billion dollar company.
Here's exactly how they did it:🧵
Meta's AI glasses now amplify your conversation partner's voice in noisy environments
Sounds helpful.
But augmented hearing in consumer glasses will be banned in most workplaces within 24 months.
Why?
Employers can't verify whether you're recording conversations or just "boosting" them.
Sounds helpful.
But augmented hearing in consumer glasses will be banned in most workplaces within 24 months.
Why?
Employers can't verify whether you're recording conversations or just "boosting" them.
December 17, 2025 at 12:30 PM
Meta's AI glasses now amplify your conversation partner's voice in noisy environments
Sounds helpful.
But augmented hearing in consumer glasses will be banned in most workplaces within 24 months.
Why?
Employers can't verify whether you're recording conversations or just "boosting" them.
Sounds helpful.
But augmented hearing in consumer glasses will be banned in most workplaces within 24 months.
Why?
Employers can't verify whether you're recording conversations or just "boosting" them.
OpenAI released new audio models with 89% fewer hallucinations and 22% better instruction following.
Impressive benchmarks.
But enterprises won't switch.
Here's why:
The migration cost includes retraining every downstream system that parses the output.
Impressive benchmarks.
But enterprises won't switch.
Here's why:
The migration cost includes retraining every downstream system that parses the output.
December 17, 2025 at 11:00 AM
OpenAI released new audio models with 89% fewer hallucinations and 22% better instruction following.
Impressive benchmarks.
But enterprises won't switch.
Here's why:
The migration cost includes retraining every downstream system that parses the output.
Impressive benchmarks.
But enterprises won't switch.
Here's why:
The migration cost includes retraining every downstream system that parses the output.
Adobe just integrated text-prompt video editing, FLUX.2, and 4K upscaling into Firefly.
Sounds like innovation.
It's actually a ransom note.
Pay for the full Creative Cloud or get left behind.
Standalone tools can't match the training data moat of an ecosystem player.
Sounds like innovation.
It's actually a ransom note.
Pay for the full Creative Cloud or get left behind.
Standalone tools can't match the training data moat of an ecosystem player.
December 17, 2025 at 10:00 AM
Adobe just integrated text-prompt video editing, FLUX.2, and 4K upscaling into Firefly.
Sounds like innovation.
It's actually a ransom note.
Pay for the full Creative Cloud or get left behind.
Standalone tools can't match the training data moat of an ecosystem player.
Sounds like innovation.
It's actually a ransom note.
Pay for the full Creative Cloud or get left behind.
Standalone tools can't match the training data moat of an ecosystem player.
LG TVs are shipping with a forced Microsoft Copilot tile that users can't delete.
This is the new crapware.
But here's the calculation: you won't return your TV over it.
The friction cost of switching ecosystems is too high.
Manufacturers know this. They're betting on consumer tolerance.
This is the new crapware.
But here's the calculation: you won't return your TV over it.
The friction cost of switching ecosystems is too high.
Manufacturers know this. They're betting on consumer tolerance.
December 17, 2025 at 9:00 AM
LG TVs are shipping with a forced Microsoft Copilot tile that users can't delete.
This is the new crapware.
But here's the calculation: you won't return your TV over it.
The friction cost of switching ecosystems is too high.
Manufacturers know this. They're betting on consumer tolerance.
This is the new crapware.
But here's the calculation: you won't return your TV over it.
The friction cost of switching ecosystems is too high.
Manufacturers know this. They're betting on consumer tolerance.
The advertising industry isn't consolidating for efficiency.
Omnicom bought IPG for $13.5B and immediately cut 4,000 jobs.
They shuttered iconic agency brands.
This isn't streamlining. It's a controlled demolition.
AI is about to eliminate 70% of campaign production work.
Omnicom bought IPG for $13.5B and immediately cut 4,000 jobs.
They shuttered iconic agency brands.
This isn't streamlining. It's a controlled demolition.
AI is about to eliminate 70% of campaign production work.
December 17, 2025 at 8:00 AM
The advertising industry isn't consolidating for efficiency.
Omnicom bought IPG for $13.5B and immediately cut 4,000 jobs.
They shuttered iconic agency brands.
This isn't streamlining. It's a controlled demolition.
AI is about to eliminate 70% of campaign production work.
Omnicom bought IPG for $13.5B and immediately cut 4,000 jobs.
They shuttered iconic agency brands.
This isn't streamlining. It's a controlled demolition.
AI is about to eliminate 70% of campaign production work.
OpenAI has admitted something the industry won't say out loud.
They rolled back automatic access to reasoning models for free users.
Not because of product strategy.
Because the economics don't work.
Reasoning models cost too much to run at consumer scale.
They rolled back automatic access to reasoning models for free users.
Not because of product strategy.
Because the economics don't work.
Reasoning models cost too much to run at consumer scale.
December 17, 2025 at 7:00 AM
OpenAI has admitted something the industry won't say out loud.
They rolled back automatic access to reasoning models for free users.
Not because of product strategy.
Because the economics don't work.
Reasoning models cost too much to run at consumer scale.
They rolled back automatic access to reasoning models for free users.
Not because of product strategy.
Because the economics don't work.
Reasoning models cost too much to run at consumer scale.
The image generation wars just ended in an unexpected way.
OpenAI released the fastest model. They topped every leaderboard.
But here's what nobody's talking about:
Standalone image tools are dying because multimodal chat absorbed the use case.
You don't subscribe to an image generator anymore.
OpenAI released the fastest model. They topped every leaderboard.
But here's what nobody's talking about:
Standalone image tools are dying because multimodal chat absorbed the use case.
You don't subscribe to an image generator anymore.
December 17, 2025 at 5:33 AM
The image generation wars just ended in an unexpected way.
OpenAI released the fastest model. They topped every leaderboard.
But here's what nobody's talking about:
Standalone image tools are dying because multimodal chat absorbed the use case.
You don't subscribe to an image generator anymore.
OpenAI released the fastest model. They topped every leaderboard.
But here's what nobody's talking about:
Standalone image tools are dying because multimodal chat absorbed the use case.
You don't subscribe to an image generator anymore.
10,000 MCP servers are now live for agent connectivity.
Anthropic donated the protocol to a Linux Foundation entity backed by OpenAI, Google, Microsoft, AWS.
We're standardizing the infrastructure for AI agents.
But here's the problem: we haven't proven agents are useful yet.
Anthropic donated the protocol to a Linux Foundation entity backed by OpenAI, Google, Microsoft, AWS.
We're standardizing the infrastructure for AI agents.
But here's the problem: we haven't proven agents are useful yet.
December 16, 2025 at 10:00 AM
10,000 MCP servers are now live for agent connectivity.
Anthropic donated the protocol to a Linux Foundation entity backed by OpenAI, Google, Microsoft, AWS.
We're standardizing the infrastructure for AI agents.
But here's the problem: we haven't proven agents are useful yet.
Anthropic donated the protocol to a Linux Foundation entity backed by OpenAI, Google, Microsoft, AWS.
We're standardizing the infrastructure for AI agents.
But here's the problem: we haven't proven agents are useful yet.
Copywriters are reporting systematic income collapse.
From six figures to near-zero. Clients replacing craft with "good enough" AI output.
But this isn't really about AI capabilities.
It's a revelation: most clients never valued writing craft. They valued cheap words.
From six figures to near-zero. Clients replacing craft with "good enough" AI output.
But this isn't really about AI capabilities.
It's a revelation: most clients never valued writing craft. They valued cheap words.
December 16, 2025 at 8:30 AM
Copywriters are reporting systematic income collapse.
From six figures to near-zero. Clients replacing craft with "good enough" AI output.
But this isn't really about AI capabilities.
It's a revelation: most clients never valued writing craft. They valued cheap words.
From six figures to near-zero. Clients replacing craft with "good enough" AI output.
But this isn't really about AI capabilities.
It's a revelation: most clients never valued writing craft. They valued cheap words.
Meta tried to play both sides with Llama 4.
One foot in open source, one foot in proprietary advantage.
The result? Chinese labs now define the open AI frontier.
Qwen overtook Llama in downloads. DeepSeek owns reasoning benchmarks.
You cannot hedge on foundational strategy.
One foot in open source, one foot in proprietary advantage.
The result? Chinese labs now define the open AI frontier.
Qwen overtook Llama in downloads. DeepSeek owns reasoning benchmarks.
You cannot hedge on foundational strategy.
December 16, 2025 at 7:30 AM
Meta tried to play both sides with Llama 4.
One foot in open source, one foot in proprietary advantage.
The result? Chinese labs now define the open AI frontier.
Qwen overtook Llama in downloads. DeepSeek owns reasoning benchmarks.
You cannot hedge on foundational strategy.
One foot in open source, one foot in proprietary advantage.
The result? Chinese labs now define the open AI frontier.
Qwen overtook Llama in downloads. DeepSeek owns reasoning benchmarks.
You cannot hedge on foundational strategy.
NVIDIA just became the "open source king." Or did they?
Nemotron 3 Nano is fully open - code, data, training recipe, everything.
But every performance optimization is locked to NVIDIA hardware.
This isn't democratization. It's strategic enclosure.
True openness would mean hardware-agnostic gains.
Nemotron 3 Nano is fully open - code, data, training recipe, everything.
But every performance optimization is locked to NVIDIA hardware.
This isn't democratization. It's strategic enclosure.
True openness would mean hardware-agnostic gains.
December 16, 2025 at 6:30 AM
NVIDIA just became the "open source king." Or did they?
Nemotron 3 Nano is fully open - code, data, training recipe, everything.
But every performance optimization is locked to NVIDIA hardware.
This isn't democratization. It's strategic enclosure.
True openness would mean hardware-agnostic gains.
Nemotron 3 Nano is fully open - code, data, training recipe, everything.
But every performance optimization is locked to NVIDIA hardware.
This isn't democratization. It's strategic enclosure.
True openness would mean hardware-agnostic gains.
Zoom just beat Gemini and GPT-4 on one of the hardest reasoning benchmarks.
Not with a bigger model. With federated orchestration of smaller ones.
The entire industry bet billions on "bigger is better."
What if orchestration matters more than scale?
Not with a bigger model. With federated orchestration of smaller ones.
The entire industry bet billions on "bigger is better."
What if orchestration matters more than scale?
December 16, 2025 at 5:13 AM
Zoom just beat Gemini and GPT-4 on one of the hardest reasoning benchmarks.
Not with a bigger model. With federated orchestration of smaller ones.
The entire industry bet billions on "bigger is better."
What if orchestration matters more than scale?
Not with a bigger model. With federated orchestration of smaller ones.
The entire industry bet billions on "bigger is better."
What if orchestration matters more than scale?
OpenEvidence is raising at a $12B valuation.
A medical AI tool. No proven clinical superiority yet.
Twelve billion dollars.
Investors aren't pricing patient outcomes. They're pricing the narrative that AI will disrupt healthcare.
The gap between valuation and validation has never been wider.
A medical AI tool. No proven clinical superiority yet.
Twelve billion dollars.
Investors aren't pricing patient outcomes. They're pricing the narrative that AI will disrupt healthcare.
The gap between valuation and validation has never been wider.
December 15, 2025 at 10:00 AM
OpenEvidence is raising at a $12B valuation.
A medical AI tool. No proven clinical superiority yet.
Twelve billion dollars.
Investors aren't pricing patient outcomes. They're pricing the narrative that AI will disrupt healthcare.
The gap between valuation and validation has never been wider.
A medical AI tool. No proven clinical superiority yet.
Twelve billion dollars.
Investors aren't pricing patient outcomes. They're pricing the narrative that AI will disrupt healthcare.
The gap between valuation and validation has never been wider.
Unconventional AI is Pre-product. Stealth mode. Focused on energy-efficient compute.
VCs don't write checks like that for incremental improvements.
They write them when they believe the current infrastructure is unsustainable at scale.
The AI stack might be structurally broken.
VCs don't write checks like that for incremental improvements.
They write them when they believe the current infrastructure is unsustainable at scale.
The AI stack might be structurally broken.
December 15, 2025 at 8:30 AM
Unconventional AI is Pre-product. Stealth mode. Focused on energy-efficient compute.
VCs don't write checks like that for incremental improvements.
They write them when they believe the current infrastructure is unsustainable at scale.
The AI stack might be structurally broken.
VCs don't write checks like that for incremental improvements.
They write them when they believe the current infrastructure is unsustainable at scale.
The AI stack might be structurally broken.
Intel is in talks to acquire SambaNova for $1.6B.
Not R&D. Not a partnership. An acquisition.
When you can't out-engineer the competition, you acquire someone who might.
That's not confidence in your roadmap. That's admission you've already lost the architecture war.
Not R&D. Not a partnership. An acquisition.
When you can't out-engineer the competition, you acquire someone who might.
That's not confidence in your roadmap. That's admission you've already lost the architecture war.
December 15, 2025 at 6:45 AM
Intel is in talks to acquire SambaNova for $1.6B.
Not R&D. Not a partnership. An acquisition.
When you can't out-engineer the competition, you acquire someone who might.
That's not confidence in your roadmap. That's admission you've already lost the architecture war.
Not R&D. Not a partnership. An acquisition.
When you can't out-engineer the competition, you acquire someone who might.
That's not confidence in your roadmap. That's admission you've already lost the architecture war.
After years of consumer-first messaging, that's a meaningful pivot.
It might also be an admission.
If your moat with everyday users was as deep as claimed, you wouldn't need to reposition.
Enterprise isn't a strategy. It's a hedge.
It might also be an admission.
If your moat with everyday users was as deep as claimed, you wouldn't need to reposition.
Enterprise isn't a strategy. It's a hedge.
December 15, 2025 at 5:26 AM
After years of consumer-first messaging, that's a meaningful pivot.
It might also be an admission.
If your moat with everyday users was as deep as claimed, you wouldn't need to reposition.
Enterprise isn't a strategy. It's a hedge.
It might also be an admission.
If your moat with everyday users was as deep as claimed, you wouldn't need to reposition.
Enterprise isn't a strategy. It's a hedge.
Citation systems in AI outputs create liability exposure that will limit deployment more than capability limits do.
Google's Gemini Deep Research generates claim-level citations.
That's not a feature - it's a legal surface area.
Every citation is a potential misattribution.
Google's Gemini Deep Research generates claim-level citations.
That's not a feature - it's a legal surface area.
Every citation is a potential misattribution.
December 14, 2025 at 10:45 AM
Citation systems in AI outputs create liability exposure that will limit deployment more than capability limits do.
Google's Gemini Deep Research generates claim-level citations.
That's not a feature - it's a legal surface area.
Every citation is a potential misattribution.
Google's Gemini Deep Research generates claim-level citations.
That's not a feature - it's a legal surface area.
Every citation is a potential misattribution.
Benchmarking against "human experts" measures test-taking ability, not the judgment and accountability that defines expertise.
GPT-5.2's 70.9% win rate sounds impressive until you realize the benchmark rewards speed and pattern matching, not the messy, high-stakes decisions experts actually make.
GPT-5.2's 70.9% win rate sounds impressive until you realize the benchmark rewards speed and pattern matching, not the messy, high-stakes decisions experts actually make.
December 14, 2025 at 9:30 AM
Benchmarking against "human experts" measures test-taking ability, not the judgment and accountability that defines expertise.
GPT-5.2's 70.9% win rate sounds impressive until you realize the benchmark rewards speed and pattern matching, not the messy, high-stakes decisions experts actually make.
GPT-5.2's 70.9% win rate sounds impressive until you realize the benchmark rewards speed and pattern matching, not the messy, high-stakes decisions experts actually make.
Microsoft says quantum advantage is "years, not decades away."
Majorana qubits + millions of qubits on a chip.
Then hybridize with AI.
We're not just getting smarter models. We're getting models that run on fundamentally different physics.
The next 5 years are going to be absolutely insane.
Majorana qubits + millions of qubits on a chip.
Then hybridize with AI.
We're not just getting smarter models. We're getting models that run on fundamentally different physics.
The next 5 years are going to be absolutely insane.
December 13, 2025 at 11:30 AM
Microsoft says quantum advantage is "years, not decades away."
Majorana qubits + millions of qubits on a chip.
Then hybridize with AI.
We're not just getting smarter models. We're getting models that run on fundamentally different physics.
The next 5 years are going to be absolutely insane.
Majorana qubits + millions of qubits on a chip.
Then hybridize with AI.
We're not just getting smarter models. We're getting models that run on fundamentally different physics.
The next 5 years are going to be absolutely insane.
946,000 layoffs in 2025.
17,000+ explicitly blamed on AI.
But unemployment is "only" 4.3%.
The dirty secret? Companies aren't firing fast. They're just not hiring.
Economists are calling it "The Great Freeze."
Entry-level workers are absolutely annihilated.
17,000+ explicitly blamed on AI.
But unemployment is "only" 4.3%.
The dirty secret? Companies aren't firing fast. They're just not hiring.
Economists are calling it "The Great Freeze."
Entry-level workers are absolutely annihilated.
December 13, 2025 at 10:30 AM
946,000 layoffs in 2025.
17,000+ explicitly blamed on AI.
But unemployment is "only" 4.3%.
The dirty secret? Companies aren't firing fast. They're just not hiring.
Economists are calling it "The Great Freeze."
Entry-level workers are absolutely annihilated.
17,000+ explicitly blamed on AI.
But unemployment is "only" 4.3%.
The dirty secret? Companies aren't firing fast. They're just not hiring.
Economists are calling it "The Great Freeze."
Entry-level workers are absolutely annihilated.
LinkedIn creators are 2.7x more likely to earn $31K+ than short-form creators.
Why? Long-form builds relationships. B2B audiences have budgets.
Entertainment gets reach.
B2B gets revenue.
If you're optimizing for views instead of money, you're playing the wrong game.
Why? Long-form builds relationships. B2B audiences have budgets.
Entertainment gets reach.
B2B gets revenue.
If you're optimizing for views instead of money, you're playing the wrong game.
December 13, 2025 at 9:30 AM
LinkedIn creators are 2.7x more likely to earn $31K+ than short-form creators.
Why? Long-form builds relationships. B2B audiences have budgets.
Entertainment gets reach.
B2B gets revenue.
If you're optimizing for views instead of money, you're playing the wrong game.
Why? Long-form builds relationships. B2B audiences have budgets.
Entertainment gets reach.
B2B gets revenue.
If you're optimizing for views instead of money, you're playing the wrong game.
Harvard kids just launched Spike - a prediction market where you bet on TikTok videos going viral.
Import any TikTok. Place tokens on its performance. Win if you're right.
Creators earn 1% of ALL volume traded on their videos.
We're literally gamifying the creator economy. What could go wrong?
Import any TikTok. Place tokens on its performance. Win if you're right.
Creators earn 1% of ALL volume traded on their videos.
We're literally gamifying the creator economy. What could go wrong?
December 13, 2025 at 8:30 AM
Harvard kids just launched Spike - a prediction market where you bet on TikTok videos going viral.
Import any TikTok. Place tokens on its performance. Win if you're right.
Creators earn 1% of ALL volume traded on their videos.
We're literally gamifying the creator economy. What could go wrong?
Import any TikTok. Place tokens on its performance. Win if you're right.
Creators earn 1% of ALL volume traded on their videos.
We're literally gamifying the creator economy. What could go wrong?
Disney gave OpenAI $1B and licensed 200+ characters to Sora.
Now fans can generate Mickey Mouse videos.
Disney curates the "best" ones for Disney+.
Translation: Free fan labor → Platform content → Disney profits.
They're not fighting AI. They're monetizing it while keeping total control.
Now fans can generate Mickey Mouse videos.
Disney curates the "best" ones for Disney+.
Translation: Free fan labor → Platform content → Disney profits.
They're not fighting AI. They're monetizing it while keeping total control.
December 13, 2025 at 7:30 AM
Disney gave OpenAI $1B and licensed 200+ characters to Sora.
Now fans can generate Mickey Mouse videos.
Disney curates the "best" ones for Disney+.
Translation: Free fan labor → Platform content → Disney profits.
They're not fighting AI. They're monetizing it while keeping total control.
Now fans can generate Mickey Mouse videos.
Disney curates the "best" ones for Disney+.
Translation: Free fan labor → Platform content → Disney profits.
They're not fighting AI. They're monetizing it while keeping total control.
Trump just signed an executive order to kill state AI laws.
Created an "AI Litigation Task Force" whose ONLY job is to sue states.
If your state has "onerous" AI regulations? You lose federal broadband funding.
California and New York just got told to sit down. Wild times.
Created an "AI Litigation Task Force" whose ONLY job is to sue states.
If your state has "onerous" AI regulations? You lose federal broadband funding.
California and New York just got told to sit down. Wild times.
December 13, 2025 at 6:30 AM
Trump just signed an executive order to kill state AI laws.
Created an "AI Litigation Task Force" whose ONLY job is to sue states.
If your state has "onerous" AI regulations? You lose federal broadband funding.
California and New York just got told to sit down. Wild times.
Created an "AI Litigation Task Force" whose ONLY job is to sue states.
If your state has "onerous" AI regulations? You lose federal broadband funding.
California and New York just got told to sit down. Wild times.
Google just turned research into an API call.
Deep Research now runs 60-minute autonomous investigations - planning, searching, finding gaps, then diving deeper.
Google's Interactions API is one endpoint for:
- Gemini models
- Pre-built agents
- Your custom agents
- File search + web search
Deep Research now runs 60-minute autonomous investigations - planning, searching, finding gaps, then diving deeper.
Google's Interactions API is one endpoint for:
- Gemini models
- Pre-built agents
- Your custom agents
- File search + web search
December 13, 2025 at 5:25 AM
Google just turned research into an API call.
Deep Research now runs 60-minute autonomous investigations - planning, searching, finding gaps, then diving deeper.
Google's Interactions API is one endpoint for:
- Gemini models
- Pre-built agents
- Your custom agents
- File search + web search
Deep Research now runs 60-minute autonomous investigations - planning, searching, finding gaps, then diving deeper.
Google's Interactions API is one endpoint for:
- Gemini models
- Pre-built agents
- Your custom agents
- File search + web search