Auren Hoffman

went into a clothing store today and there were no prices tags. no prices at all. as an online shopper, just seems incredibly strange and off-putting liked the merchandise but decided to shop and their online store instead

there is a MASSIVE opportunity for jr. employees that has never existed before revenue per employee is going up 3-10x this decade. productivity is not going to land on a senior director who refuses to learn new tools. it is landing on a 23 year old with AI.

@kbarrbuilds Probably yes

@auren none of those three require a degree. so is mine useless?

three kinds of new grads get hired in 2026: the builder (shipped a thing), the operator (ran a thing), the closer (did what they said they would). every great candidate is at least one. if you are none of these, you might get hired but you are not going to be successful. that is the whole picture.

Within a year, almost every first conversation between a founder and a venture capitalist will be agent-to-agent. That doesn't happen today because it would be a huge power imbalance to have an agent talking to a human. For Flex Capital, it would be weird if the founder had us talk to an agent. It might be interesting the first 1 to 4 times, but then after that it would be a little bit weird and have a little bit of power imbalance if one of our humans was talking to their agent. The same thing would go the other way if we had a Flex Capital agent talk to a founder. That would also be pretty weird and have a power imbalance. However, if the first conversation was agent-to-agent, that would make a lot of sense. The founder's agent could see if Flex Capital was a great partner for them, and Flex Capital's agent could see if the company was a good partner for Flex. If that first meeting goes well, then the second meeting (or the second date) would be human-to-human. think we will see almost every first meeting (founder to VC) be agent-to-agent within the next year, and then that would then determine if the second meeting happens (being human-to-human).

LiveRamp founder @auren comes on the pod and explains How to Make LiveRamp Great Again. youtube.com/watch?v=wYXNhq…

@Joscocok with housing and food: most private colleges do

@auren Few colleges cost $400k/4 years.

universities below the top 20 are going to face a massive enrollment crisis. can’t lend $400k against a credential that no longer has a wage premium. math does not work. we're going to see more and more of these schools up for sale. how would you turn around your fav school?

what elementary school your kid goes to will likely not matter. doubtful it will change your kids' outcome by more than 3% (though it can matter a lot if your child is on one of the tails)

@Jeff88Ho that is happening

@auren Convert them to data centers.

am writing skills (.md files) for myself and for people that i work with that we cannot delegate to AI (yet) it is a GREAT to to better understand what i should be doing. this includes preferences around travel, health, food, shopping, tech, and more anyone else doing this?

Super old news. Masters degrees lost their value 2 decades ago

every employee is an executive now with 100 people on their staff (AI + compute). very soon when new talent joins a company they'll ask "what does my AI team look like?". answer should not come from a dinosaur

@Myles_Younger I probably just made it up.

@auren is there data or a reference for this? (actual question; not snark)

Agent autonomy has grown 13X in 6 months. That's insane growth. Now extrapolate out just a few months: the world we will live in is going to look nothing like it was in the past.

@meksikanpijja Simplest thing would be just to focus on the middleware. Focus on the middleware. That's a $100 billion company right there.

@auren what do you think would've been the single most impactful long-term investment into the actual product(s)?

Congrats to LiveRamp and the team for the sale to Publicis. This deal was a steal for Publicis (more on that below). LiveRamp was a shining gem and still is such an important part of the ad tech and marketing tech industry. It has so many amazing people and has a really incredible product, which has not changed that much over the last 10 years or so. It has bittersweet seeing a company that one created and was a part of for almost a decade go into decline. But despite that, the company is still in really good position. Unfortunately they were not able to take advantage of their position. For the last decade they traded long-term positioning for short-term gain. That trade resulted in the ire of their customers and even more so their partners. Now that they will be in a new home in Publicis, I hope that we will see a renaissance of LiveRamp and LiveRamp will continue to be an incredible part of the industry going forward. The company has a lot it needs to do: It needs to focus on the core product. The core product is 70% of their revenues and 500% of their profits. They need to focus on making that product better. The core product really hasn't changed in about a decade. They have to focus on making the middleware product better: making it easier to use, making it so you can onboard partners faster, onboard customers faster, onboard connections. the most important LiveRamp metric The number of connections per customer is the most important metric for LiveRamp. That number is surprisingly low. Most customers only have a small number of connections using the LiveRamp system, not because they don't want to use LiveRamp but because either LiveRamp is way too expensive or takes too long or it's too bureaucratic or too burdensome (or usually all of the above). Now that there is going to be new ownership of LiveRamp, hopefully they can focus more on the product and customers can go from, let's say, 10 connections in LiveRamp to 400 connections, 500 connections, even 1,000 connections. That's how we know LiveRamp is really doing a good job. Also the next way we know LiveRamp is doing a good job is if they can increase the number of customers. Today most of the customers that use LiveRamp are quite large ... which is great. They have still the largest B2C customers in the world. It would be amazing if everybody can use LiveRamp: mid-market companies ... even small companies (would be awesome to have a self-service system). all that said, I'm long the core engineering team and product team of LiveRamp and LiveRamp's position in the market. This deal was a steal for Publicis. It was a great deal for Publicis, assuming they can even make LiveRamp a tiny bit better. this definitely could be a deal that transforms Publicis. There's no reason why LiveRamp can't be a $100 billion market cap company by itself. It should be. It should have that position. It should be well over $100 billion and with the right product vision. it can get there. Let's make LiveRamp great again!

Perplexity is north of $500m ARR and there are only 9 people doing GTM/customer success across 50k enterprise customers. insane and completely impossible a year ago. we all wish we were them

new summation pod with @dmitry140, Perplexity’s Chief Business Officer. we get into how Perplexity Computer started as one Slack channel, why curiosity (not capability) is now AI's scarcest resource, and what every new hire will ask in interviews 6 months from now. youtube.com/watch?v=TyeNSj…

@jbrichman These would be great working on Ovitz.

@auren Michael Ovitz, Chris Miller, Marc Hijink, Demis Hassabis

we just climbed to the 34th most popular technology podcast in the world. woo hoo! any great suggestions of guests that you would love to hear from???

@stalmico Both pretty amazing.

@auren what was the speed difference like?

Codex or Claude? Both of them are pretty amazing. Over the weekend, I ported one of my apps from Claude Code to Codex to see how good Codex was. Codex was definitely a very capable system. thought it did not yet live up to Claude Code. Claude Code was significantly faster. Codex was also just doing the most complex things you could do rather than focusing on more simple tasks. I did find that if we move Codex to Medium performance, it performs BETTER than on High, which to me was really counterintuitive. not exactly sure why you would have Codex on High when Medium seems to perform significantly better. Maybe my tasks were just not complex enough for High. That would be really helpful for Codex to help teach us: when to use it for Medium, when to use it for High. When Codex was on High, it was just doing everything in a super roundabout way, whereas Claude Code just got to the answer very quickly. With Codex, I had to guide it much more and I had to be much more technical and really architect it more when it was on High. When Codex is on Medium, it actually performs pretty well and doesn't need a high-level technical architect.