Patrick Donohoe

@FrankieBorrelli @ForePlayPod Everyone leaving early to take the LIRR

Why is no one at the US Open?

I think this is an oversimplification. Because the main problem is token costs are exploding. But there is a path where frontier labs start experiencing some margin compression especially for larger customers. AI just started “working” this year. The end-state competitive dynamics on the model layer is unknown

Chinese open source will win. Enterprises saying they won’t use Chinese models are starting to adopt as they are now token conscious.

Inference is undoubtably going to explode in the coming years. But people are underestimating the impact of using ai to build “agents” that are orchestrating traditional python executables to deterministically solve business problems. The executables are cpu-bound so opportunity for new bottleneck

The CPU-to-GPU ratio is going from 1:8 to 1:1. Intel CEO Lip Bu Tan @LipBuTan1 says agentic AI inference flips the math that made CPUs irrelevant in the AI build-out. Training needed 8 GPUs per CPU. Inference loops and agent orchestration are CPU-native work. If he is right, Intel's data center business has a tailwind consensus has not modeled. Why it matters and what breaks it, on No Priors: podcastalpha.substack.com/p/intel-lip-bu… Source: No Priors - youtube.com/watch?v=asCgCv…

This is one under-appreciated part of the ai labs revenue growth and scaling. When a startup goes from 0-10bn the big players are interested but don’t really care since you aren’t taking their revenue. From 10-100 you start needing to extract budget from their services which motivates them to fight back

Google, Microsoft and Salesforce are backing a new AI software standard aimed at keeping enterprise apps at the center of workplace automation. This could blunt OpenAI and Anthropic’s push to make ChatGPT and Claude the main gateway for work AI. Read more: thein.fo/4euyi9R

@bindureddy This is directionally correct. If you can get opus + gpt-5.5 on the consumer subscription plans then using them for everything is fine. If you are paying per token at a large scale then you need to have some tough conversations and think about optimizing

Haha - this is unbelievable! I suspect you just need Opus 4.8 and GPT 5.6 for just 10-15% of tasks Kimi and GLM will do the rest Enterprises who are paying eye-watering Anthropic bills are officially OVER PAYING like crazy 🤯

@MarkLevineNYC Nyc is not seeing a tourism boom because the residents have such a high income+want to go to the games so they bid up the ticket prices out of the international fans range (also very high cost of hotels). Similar to Knicks finals games price action

This is 100% the experience in NYC. Hotel bookings are *lower* here than this time last year. Bookings into JFK are down from last year. FIFA had projected 1.2M visitors to NY/NJ. We will be lucky to hit 1/3 of that. Huge disappointment.

@AmFootballin not anymore

The U.S. has not won consecutive matches at a World Cup since the 1930 inaugural tournament 👀

Think Real Madrid should purchase the entire #usmnt. Best team in the world 🌎

@mprkhrst underrated how its more fun to solve hard problems when you get to do it with your friends

This looks a lot lonelier and way less fun than hiring great people

@craigweiss Juiced scale … the 500k a year data labeling army

omg… meta is become scale

@aschiff19 @matt_slotnick It very much depends on the responsibilities of the specific FDE role. There is a very large divergence across companies

@patrickdonohoe @matt_slotnick Top FDE is an oxymoron…no top talent should take that role

if the large consulting firms actually knew how to implement real AI systems they wouldn’t be suffering like this. the demand is enormous, they just don’t have offerings to meet it

@PolymarketSport The Aussie right back has had a tough morning. Someone get him a pint and a lawn chair. If your gonna get blown by all game at least be comfortable

@jenzhuscott @DavidSHolz @midjourney If you have a cash generating machine and really smart people. Seems like a pretty noble thing to do to deploy them against impossible problems in medicine

So, @DavidSHolz is doing what bold founders do: swinging at impossible problems. Medical pros saying @midjourney full-body ultrasound scanner ‘won’t work technically’ - fair skepticism - new hardware always faces physics, validation & regulatory hurdles. And I know from 1st hand: hardwares are super hard. But changes have to start somewhere. From Leap Motion to profitable AI to this: he executes while others talk. Excited to see the data as it ships. Progress demands builders who ignore the ‘impossible’ and just give what might be possible a go. ⚡️❤️‍🔥

@Amank1412 They want to be uber not Lyft

OpenAI allegedly spent 44% of revenue i.e. 5 BILLION dollars on "sales and marketing" according to leaked financial statements. What's going on?

@theinformation At their core these agents aren’t that hard to build, which means distribution is everything. Developers are very fickle. The OpenAI acquisition of openclaw upset developers so they churned off of it

Some copycats are starting to catch up with OpenClaw. Hermes, an agent tool from Nous Research, recently eclipsed OpenClaw in terms of new GitHub contributors in the last 30 days. More details & what it means for the AI agent race: thein.fo/43JJs5H

@simoncorry @tryramp @eglyman @karimatiyeh Good luck with what’s next!

Drum roll… after an incredible run at @tryramp I’m moving on to my next adventure. Eternally grateful to @eglyman and @karimatiyeh for taking a chance and letting me recruit one of the best design teams in fintech. Big news dropping Monday!

@JamesonCamp Selling a vision is the most important skill for founders. To investors, employees, and customers

Every great engineer I've tried to hire is already making $400K+ at Google or Meta with RSUs vesting every quarter and a promotion locked in. They are not browsing your job post. A friend who co-founded a $100M company told me this when I was searching for a CTO: "You need to pull people out of their jobs. That's the only way." He was right. Comp in tech right now is insane. Golden handcuffs everywhere. You will not outbid these companies on salary. So you have to sell a future worth more than their current paycheck. Equity in something they actually believe could win. The best companies in the world were built by founders who could walk into a room and convince someone to bet on them over a paycheck. That's the skill. Learn it.

@BosscherBen Feels like a sample size issue. The bigger problem for Duran is he went ghost in the playoffs

Last Six Finals Teams: Knicks – KAT 7’0 Spurs – Wemby 7’5 OKC – Chet 7’1 Pacers – Mules Turner 6’11 Celtics – Kristaps 7’2 Mavs – Dereck Lively II 7’1” Nuggets – Jokic 6’11 Heat – Bam – 6’9” 6/8 centers are floor spacers 5/8 are 7 footers Duren is neither. Don't give him a max

@theinformation AI+traditional saas are realizing that there is a limited amount of value that can be provided to enterprises. And the company that is most embedded in the workflows will get to capture it

Google, Microsoft and Salesforce are backing a new AI software standard aimed at keeping enterprise apps at the center of workplace automation. This could blunt OpenAI and Anthropic’s push to make ChatGPT and Claude the main gateway for work AI. Read more: thein.fo/4euyi9R