Blake Marggraff

It's 10x easier to point out risks than to propose solutions. It's 10x easier to propose solutions than to build them IRL.

NeoSaaS ☠️

Using LLMs to do analysis of text or data without code as a middle step feels like the cardinal sin of AI in business. Like firing a machine gun at a hunk of granite to carve a statue.

Now, the main work tasks for which I'm *not* using LLMs are (i) writing natural language content and (ii) summarizing natural language content other humans have written for my direct consumption. Corollary: my business use of LLMs is converging on deterministic work products.

2027 will be the year of tokens beyond screens

First time in my career building stuff that is perhaps multiple step ahead of multiple research publications that were all published <30 days prior to today.

No joke, this will force governments to rethink immigration. Before autonomous humanoids are omnipresent, there will be real labor cost advantages to real-time tele-operated physical AI capable of real work. Pithily: if this walks across a border, is it an immigrant?

Robots beget robots. The Overton window for consumers will move as a function of interaction with physical AI. That might lead to counterintuitive adoption patterns, where certain less-traditionally-tech-savvy demographics are exposed to, and therefore open to, embodied AI.

Administrative spend will collapse to the cost of tokens. Workers closest to the creation of actual utility should, and will, be able to be better-compensated because high-performing companies will aggressively cut bloat.

From "plenty of room before the five hour limit" to "quick, buy more usage credits!" in no time flat

Consumers who adopt physical AI might eventually start feeling like life is subsidized. For instance, a humanoid (or even autonomous vehicle) in the home can become an edge compute device many hours each day, possibly to the degree that non-commercial physical AI becomes free.

Nice. Extending it out an additional step, since data are not intrinsically valuable, the uniqueness and value of tasks that training data enable will, to some degree, flow upstream to the market-assigned value of data, since that's where WAM-building resources will concentrate.

A little late, but great to spend the evening w the @nvidia Inception team for @NVIDIAGTC Taipei / @computextaipei – @LOKarpas opening remarks. Super fun discussions w MetAI team about generative social-physical sim validation. So many familiar faces from San Jose GTC!

Thanks again for hosting us @itseyh @robot_health, @Robo_Tuo! Excited to visit your new space coming online in a couple months, and pumped for the vision you're realizing: high-density innovation with global collab. @WashU @SkandalarisCtr the Shenzhen alumni network 🦾

The more deflationary the offering, the more valuable the company producing it. This has always been true to a degree with the classic winning formula of "higher quality, lower cost," but AI and soon robotics will make this especially clear.

Seems like companies' valuations will increasingly converge on the amount of economic utility they create per Wh. And the root of economic utility is human flourishing (time, health, agency). So... maybe the trick to building generational businesses is QALY-maxxing.

Building & tuning new functionality for robots is deeply gratifying.

Phase one of the Shenzhen Robot Valley Showroom is ready. DM or email us if you’d like to visit. It’s still early days for robotics, but Robot Valley is helping accelerate progress globally. @TraceyWu9062

LLMs were priced on cost from the start, but robotics will be priced on value. Instead of token+margin, embodied AI will be human labor divided by x. The bar will be high, but app-layer physical AI with stellar deployment will be the new "SaaS margin."

Someday, a robot will be able to walk over to you when you get home and, just by looking at you, check on your pulse and respiratory rate. "Someday." 🤖 Booster K1 via @boosterobotics @ErenChenAI + @PresageTech PPG