Joe Fraiman

@joefraiman

Entrepreneur

Marin, CA

Joined July 2011

166 Following

98 Followers

4 Posts

Joe Fraiman

@joefraiman

3 days ago

https://t.co/uYniwfo2nA

Joe Fraiman

@joefraiman

4 months ago

Anthropic has an Uber problem. In the last 6 weeks, they shipped Cowork, Claude in Excel & Powerpoint, Code Security, data connectors for Google Workspace and financial platforms, and two new flagship models. It's the most aggressive product blitz in AI history. Truly incredible. And it broke them. Capacity can't keep up with demand. Services scream "Overloaded." The best AI on the market is also the least available. They can't build data centers fast enough to fix this. So there's really only one lever to pull: Price. Uber faced this exact problem a decade ago. Too many riders, not enough drivers. Their answer was surge pricing. People hated it. Then every ride-sharing company adopted it. Then everyone stopped complaining. The same pattern is coming to AI. Token prices will rise, and the arbitrage between AI and human labor will shrink. And the skill that separates great AI users from everyone else won't be prompting — it'll be cost optimization. That creates four quadrants of AI work: Tinkering (Low cost, low impact) Where most people live today. Useful, but incremental. Summarize this doc. Rewrite this email. The ROI is real but modest. Wasteful (High cost, low impact) Work that's economical at today's token prices but won't survive a price increase. As costs rise, some of this work migrates back to humans — at least temporarily. Commercial (High cost, high impact) The enterprise engine. High-volume, high-stakes tasks where AI pays for itself many times over, even at premium pricing. This is where the big money flows. Genius (Low cost, high impact) The honeypot. Workflows that consume relatively few tokens but produce outsized value. The people who find these will guard them like trade secrets — because they are. The AI users who win long-term won't be the ones with the best prompts. They'll be the ones who know exactly which quadrant they're in — and how to move down and to the right. @chamath @DavidSacks @jason @friedberg @garrytan

joefraiman's tweet photo. Anthropic has an Uber problem.

In the last 6 weeks, they shipped Cowork, Claude in Excel & Powerpoint, Code Security, data connectors for Google Workspace and financial platforms, and two new flagship models.

It's the most aggressive product blitz in AI history. Truly incredible.

And it broke them.

Capacity can't keep up with demand. Services scream "Overloaded." The best AI on the market is also the least available.

They can't build data centers fast enough to fix this. So there's really only one lever to pull:

Price.

Uber faced this exact problem a decade ago. Too many riders, not enough drivers. Their answer was surge pricing.

People hated it. Then every ride-sharing company adopted it. Then everyone stopped complaining.

The same pattern is coming to AI.

Token prices will rise, and the arbitrage between AI and human labor will shrink.

And the skill that separates great AI users from everyone else won't be prompting — it'll be cost optimization.

That creates four quadrants of AI work:

Tinkering (Low cost, low impact)
Where most people live today. Useful, but incremental. Summarize this doc. Rewrite this email. The ROI is real but modest.

Wasteful (High cost, low impact)
Work that's economical at today's token prices but won't survive a price increase. As costs rise, some of this work migrates back to humans — at least temporarily.

Commercial (High cost, high impact)
The enterprise engine. High-volume, high-stakes tasks where AI pays for itself many times over, even at premium pricing. This is where the big money flows.

Genius (Low cost, high impact)
The honeypot. Workflows that consume relatively few tokens but produce outsized value. The people who find these will guard them like trade secrets — because they are.

The AI users who win long-term won't be the ones with the best prompts.

They'll be the ones who know exactly which quadrant they're in — and how to move down and to the right.

@chamath @DavidSacks @jason @friedberg @garrytan

joefraiman retweeted

Kevin Song @kevinjsong

over 4 years ago

Yesterday we officially launched @risewithco and announced years of hard work and $30M+ of funding from amazing investors. For my first tweet ever, I want to share my Founder’s letter and why withco matters. If it resonates, I hope you’ll share it. https://t.co/dVHnmKZxJB

261