|
Welcome to The Weekly Fizz!
The Drip
OpenAI closed $122 billion in funding at a valuation near a trillion dollars. Apple took years to hit
that mark. OpenAI did it fast, and the product is in your browser, not on your desk, wrist, or in your
pocket... yet.
65% of the AI funding in the first quarter went to OpenAI, Anthropic, xAI, and — wait for it — Waymo.
For all the bubble talk, the money isn't slowing.
But there's a math problem underneath it. The fundamental concept of a capitalist economy is that you
extract resources and turn them into something worth more than the resources themselves. Nobody is
currently doing that in generative AI. Three ways out: prices go up, costs come down, or subsidies
keep flowing.
Meanwhile the ads all show someone saying "clean up my calendar for tomorrow." Fine feature. Justin
uses stuff like that every day. But inbox cleanup across the entire working population isn't the value
equation that justifies trillion-dollar valuations. Kellan called it a nouns and verbs problem — AI is
being packaged for 50 different verbs when what's missing is a handful of clear nouns that generate
value. Until we land there, the gap between valuation and output stays confusing.
Inside The Bottle
You'll see further down — Microsoft is testing OpenClaw-style features in Copilot. Big news.
Kellan, meanwhile, built an office for ours.
There are departments: research, operations, engineering, marketing. Stations where the agents work. A
break area for when they're idle. There is, of course, a robot. His name is Bob. Bob is purple.
The whole thing is called the Command Center — a custom app that functions as a digital org chart for
a team of agents that lives mostly inside Kellan's laptop. Each agent has a peer reviewer, because —
as Kellan put it — AI agents are high on their own supply. They're confident the work is right even
when it isn't. Another agent usually catches what the first one missed.
Is the office more office than the job requires? Yes. Kellan owns this.
"If this were a client build, 80% of it would be cut." But underneath the purple robot is a
real idea about what agentic work has to look like to actually be useful — not more agents, but
managers, consumption layers, a way to keep the pile of agent output from becoming a pile nobody
reads.
We get into all of it on the pod. Worth a listen.
Lab Notes
|
■
|
Justin's note: The names of the tasks on your list matter more than you think.
Vague tasks get punted forever. Make it the actual next action — time-bound, specific — and it
moves. AI is good at helping deconstruct vague items into the specific next thing.
|
|
■
|
Kellan's note: I went too far on the Command Center office. Custom robots,
plants, break areas. Part of it was skill-building, part was getting high on my own supply. If
this were a client build, 80% of it would be cut. Worth knowing where that line is when you're
building for yourself.
|
What Stopped Our Scroll
|