I build the thing, not the deck about the thing. Product Operator. Claude Code, MCP, agent loops. Reading the AI market: capital, compute, unit economics.
Q1 2026: $300B in venture capital globally, fewest venture deals since 2016. Both describe the same three months.
Four rounds took $188B: OpenAI, Anthropic, xAI, Waymo. 65% of every venture dollar deployed globally last quarter.
SVB's H1 2026 State of the Markets report calls it a "barbell effect": massive late-stage rounds in a handful of mostly-AI companies at one end, early-stage investing at the other, a hollowed-out middle in between.
PitchBook titled its Q1 2026 report "The Magnificent Few."
Venture is not one market. It is two. Late-stage mega-rounds for a handful of frontier labs. A compressed Series A-and-earlier environment for everyone else.
Know which one you are raising in. Pitch to that one.
Remote work promised borderless collaboration. Best people, best problems, anywhere.
The tools finally caught up. The talent is everywhere, already working in public.
So why are the structures retreating.
2025: 92% of OpenAI subscribers on $20 ChatGPT Plus.
2026: 92% projected on $8 ChatGPT Go.
Same number. Complete inversion. One year.
Ad pilot hit $100M ARR in under 2 months. By 2030, ads projected to be a third of revenue. The company that raised $40B to build AGI is building an ad business.
@AndrewYNg named the PM bottleneck a year ago. Last week he updated it: it didn't disappear, it multiplied. Every function downstream of code is the next constraint.
I ship 0→1 in Claude Code, built my own MCP, run agent loops. That's the team I'm building toward.
@WillowVoiceAI if I thought I couldn't love you more, you proved me wrong.
Icon now moves to any corner or edge (no more showing up in every screen recording).
Double-click to talk (no more holding the button).
Locked in for life.
The PM role isn't dying. It's bifurcating.
Senior PMs at AI labs: $468K-$860K median. Non-AI companies: $150K-$200K.
I'm in the hybrid group. Ship 0→1 in Claude Code, align async around it.
The shift isn't a flip. It's a slow rebalance.
Wrote this a month ago.
The deal closed today. The pressure dynamics behind it (GPU crunch, model dependency, app-layer margins) are getting worse.
Cursor was the first acquisition in this cycle. The next ones will be the same shape.
Cursor: $2.7B ARR. 24x growth. Negative 23% margins. Investors lined up at $50B.
They chose a $60B acquisition right with SpaceX instead.
SpaceX solves two problems at once: compute access and the scale to compete when your model provider launches a rival product.
The fastest-growing AI companies are making moves that only make sense when you see the infrastructure layer from the inside.
@meetgranola has a tiny detail I love.
When you zoom out in Mission Control on Mac, the window changes to read "Granola is here."
Almost no user will see it. They built it anyway.
That kind of choice is the difference between a tool and a product.
When you pay people for a metric, they will produce that metric.
@AIatMeta tied employee bonuses to AI use this year. Token consumption jumped 22 percent in one month.
Now Meta is reversing. Six months from "use more" to "use less."
Tokens are easy to consume. Productivity is hard to measure.
While the hyperscalers burn $180B on AI data centers, @Apple is betting on the 1.25 billion devices already in people's pockets.
On Monday Apple confirmed Siri AI runs on Google's Gemini models on Nvidia GPUs. Apple silicon couldn't run a model that big.
Decentralized AI vs centralized capex. Both bets at once.
CFOs are modeling compute costs with 400% sensitivity ranges. CME is launching compute futures to fix that.
GPU rentals jumped from $2.75 to $4.08/hr in two months while older H100s fell from $8 to $1-4. Volatility in both directions.
Compute is becoming an asset class.
A fusion power plant just got its first commercial customer.
Helion raised $465M at $15.5B, almost 3x last year. Microsoft signed the off-take agreement. Plant is under construction in Washington state.
Fusion as AI infrastructure is moving toward deployment.
Mike Schroepfer, ex-Meta CTO, just raised $250M for the physical layer underneath AI: energy, grid, critical minerals.
Panthalassa got $140M from Thiel for ocean-powered data centers. Heron Power got $140M to rebuild the transformer.
AI's bottleneck is steel, seawater, and a 100-year-old part.
Alphabet announced an $80B equity raise yesterday. Upsized to $84.75B this morning. Berkshire Hathaway anchored it with $10B.
Pichai expects $180-190B in capex this year. The world's biggest cash machine is diluting shareholders to fund AI compute. That's the scale we're at.
120 years ago, factories got electricity. The 10x didn't come when they lit the lights or electrified one machine. It came when every machine got electrified and the factory rearranged itself.
AI is at phase 2 right now. The factory hasn't rearranged itself yet.
I have been using Willow Voice every day for months and I am not going back to typing.
The interface for AI agents is starting to look less like a chat window and more like a co-worker you trust enough to murmur to.
Most people I know stopped calling customer support years ago. You only call when you're fully blocked.
That's pent-up defection, not lost demand. When voice AI cuts the friction, the unspent volume comes back.
Jevons paradox: efficiency rarely shrinks demand.
The dashboard is losing the bank account.
Stripe shipped a CLI. Ramp shipped a CLI (275 → 2,500+ users in months). Mercury shipped one. Brex went agent-first.
Finance is moving where the work already happens — Claude Code, Codex, terminal.
@omergero@ivanburazin Many, many years ago in NYC, I met a very nice Israeli lady and the first thing she said to me was: “Dunja, I am going to make you my hummus.”
She did give me the recipe 😊
Closed AI models are the operating systems. Open source is Linux on the back end.
Jai Das (Sapphire) on @theinformation: the split is coming to AI. Frontier labs win the premium reasoning layer. Open source wins the high-volume, cost-sensitive layer.
Different jobs.