Surprised by the step function in "intelligence" since swapping Fable 5 into my agent harnesses. So much for "better models make the scaffolding obsolete." A harness loops model calls so small per-call gains compound. A good harness magnifies the model improvements not killed by.

AI is making teachers’ jobs a nightmare. Despite educators being the main force ensuring future generations remain smart, curious, and capable, every day is a struggle for them. AI enables unprecedented levels of cheating. It's disabling students’ ability to think critically and form their own thoughts and opinions. At Oboe, we want to turn AI on its head. Rather than using it as a shortcut, we want to use it to democratize access to great learning. So our goal is the same as that of teachers: teaching billions of people trillions of things. That's why, starting today, we’re offering completely free access to Oboe Pro for the first 1,000 teachers that email us at teachers@oboe.com from their school addresses. If you know a wonderful teacher who would love Oboe, please send this their way.

Ambitious young people used to join startups before founding their own. Now they skip straight to founding. Every founder says the best people are impossible to hire. All my AI native portfolio companies that fail are getting bought. That talent is too valuable to let walk.

Thrilled to welcome @mikeyyyzhao to @EniacVC! Michael spent 8 years as co-founder & CTO of @vimcal, a deeply technical, AI-native builder who shipped product at the edge of what’s possible. Now he’s bringing that founder’s instinct to the other side of the table. He’s also woven into the YC, Berkeley, and SF founder communities, and has been quietly writing first checks and helping founders go 0→1 long before it was his job title. Founders backing founders is how we started at Eniac. Couldn’t be more excited to have him continuing that tradition. Welcome, Michael 🎉

Some personal news: I'm joining @EniacVC ! I’ll be investing in founders at pre-seed and seed with a focus on applied AI, vertical SaaS, and agent infrastructure. I stepped away from @vimcal earlier this year after eight years as co-founder and CTO. Building, scaling, and leading Vimcal has been the most challenging and rewarding experience of my career. I'm so thankful for my co-founder @jaylbird11 , our customers, investors, and the entire Vimcal team. I'm staying on as an active advisor and board member, and remain incredibly bullish on what the team is building ahead. Why venture? When John and I started Vimcal, we got an unbelievable amount of help from the @ycombinator community and beyond. Since then, I've tried my best to pay it forward, helping founders with everything from YC apps to product, GTM, and recruiting, sharing the lessons and mistakes I've made along the way. Over time, I started to support founders beyond just giving advice. That often meant writing the first check into a company, before they’ve even incorporated. I want to take what I’ve learned and make supporting and backing founders my full-time mission. Why Eniac? Eniac was one of the first seed-only firms, backing founders since 2010 when seed was barely a category. The founding partners (@Hadley , @nihalmehta , and @timy0ung ) were all former founders themselves, with a conviction that founders need partners who understand the 0-1 stage and are willing to roll up their sleeves alongside them. It says something that repeat founders with successful exits, who could raise from anyone, keep coming back to Eniac for their first check. I can't wait to continue Eniac's tradition of being the first believer in founders. If you or someone you know is building something new, I'd love to chat! Don’t hesitate to DM me!

The most important thing I look for in agentic systems these days: usage that compounds into an advantage against future competitors. Not just for that one user, but advantage that accrues to the entire system.

Today, we're rolling out on-demand songs and diagrams in @oboelabs. You remember the alphabet from when you were 4, but you've already forgotten something you read this morning. That's because walls of text are the worst way to retain what you want to learn. Songs lodge in your brain for decades, and a good diagram can help a complicated concept click in seconds. Two new ways to learn, live today.

Excited to have so many people here for NY Tech Week, here's my latest tips on being a real New Yorker this week: 1) You heard real New Yorkers ride the subway or Citi Bike. Wrong. We ride the top deck of the big bus to go from meeting to meeting. To look extra legit wave at the people on the street. 2) Want a Knicks hat for that Chalamet courtside MSG look? Wack. The insta girls of Williamsburg and the West Village wear Jets gear head to toe this season. 3) Need dinner after a long day of networking? Skip the tiny spots with no sign and a line out the door, that's where confused tourists eat. Real New Yorkers go straight to the Times Square Bubba Gump and ask for the window table so the city can see you crushing life. You’re welcome

You just won a 2-week, all-expenses-paid vacation to Staten Island. But there’s a catch: you have to stay within one region the whole time. What are you picking?

Hot take - I don’t really worry about model cos competing w/ enterprise AI startups Yes, they are now doing FDE work, but: (1) AGI/models are ultimate priority; and (2) they won’t build out integrations, workflows, etc for the long tail of use cases (especially vertical ones!)

Stupid question: have any of the startups Claude ‘killed’ this year actually died? Or even been materially hurt?

🌟 Judges announced for our Vertical AI Pitch Competition 🤖 Vertical AI • June 2 • @BrownRudnickLLP Joining the panel: - @Hadley (@EniacVC) - @hayleybay (@firstround) - @jefielding (@EverywhereVC) - @jordanrcrook (@betaworks) Looking forward to an exciting day of pitches and conversations across Vertical AI!

Got a terrible haircut this week because my barber left to start a company. What a world we’re living in.

The traditional VC fund org structure produces slow decisions, politics, and consensus-driven strategy. The current environment is moving way too fast for an org to operate with multiple CEOs. The biggest firms are moving to structures that look more like 500 person corporations. We on the other hand have organized ourselves like an 11 person startup, which in my experience is the most dynamic type of organization.

Post-acquisition founder doing their earn-out

Tokenmaxxing gets expensive on June 15 when Anthropic ends subsidized rates. @coworkerapp just gave teams a way through.

I’ve been wondering why Anthropic has been so slow to launch a native OpenClaw-type harness. Just dawned on me it’s likely because it would massively increase token usage on Pro and Max plans, and they don’t have the compute for that.

OpenRouter just raised $113M at $1.3B routing 100T tokens a month between providers. Google Ventures led. Interesting signal in the labs-own-everything vs commoditize debate when a model lab’s own investor arm is funding the swappable layer.

Turns out my wife does not love getting nudges from my AI assistant about things I’m waiting on her for