Jason Preston
Daily

What a $136,000 AI assistant tells us about judgment

Jason Preston · 3 min read ·

Every few years there’s a startup story that sounds like satire until you sit with it for a minute.

This week’s entry: a billionaire paying $136,000 a month for an AI assistant that lives in his text messages.

In his newsletter, Alex Heath describes Poke as “OpenClaw for normies” — a proactive, SMS-native assistant that feels more like texting a friend than prompting a chatbot. The part that caught everyone’s attention wasn’t the interface. It was the pricing. Poke charges based on how you use it and what it thinks it’s worth to you. For one very rich customer, that number landed at $136k per month. And he paid it.

It’s tempting to treat that as the latest entry in the “rich people doing ridiculous things” file. But I think there’s something more interesting happening here: we’re starting to put a price on moving judgment off your plate.

Look at what’s happening at the other end of the spectrum. On the enterprise side, Microsoft’s new Copilot Researcher uses multiple models under the hood — GPT to draft, Claude to critique — so that “two AIs keep tabs on each other” before a memo hits your screen. Microsoft’s pitch to knowledge workers is simple: we’ll not only write the draft, we’ll also do the first-pass review you don’t have time for.

Meanwhile, Oren Etzioni is waving a caution flag in his piece on how to read with AI. If you let AI summarize everything for you, your understanding gets flatter and your advice gets worse. The risk isn’t that AI reads for you. It’s that you quietly give up the habit of thinking for yourself.

Put those three together and a pattern emerges:

  • Consumers are starting to pay real money to outsource judgment in the messy parts of life (“just text Poke, it’ll handle it”).
  • Enterprises are wiring judgment into their tools as a feature (“don’t just draft, also critique and cross-check”).
  • And the people studying this space are warning that if you don’t stay intentional, your own judgment muscle atrophies.

So is $136,000 a month crazy? Maybe. But it’s also a loud signal about where the value is accruing in this wave of AI.

We spent the first phase of the LLM era promising speed: faster drafts, faster summaries, faster slides. The next phase is about trusted delegation. Not “write this email,” but “decide which emails I should even see.” Not “summarize this deck,” but “tell me which two slides actually matter for the board meeting tomorrow.”

That’s a different job. And it lives much closer to the line we’ve historically drawn around our own agency.

For founders, there’s an obvious design question here: where does your product sit on the spectrum between tool and teammate? Tools make things faster. Teammates own outcomes. The more your product acts like a teammate — watching the stream of your life or your company and stepping in unprompted — the more you have to decide how opinionated it should be, how it explains itself, and what happens when it’s wrong.

For the rest of us, the number to obsess over isn’t $136,000. It’s the portion of your day where you’re tempted to hand off thinking entirely. The emails you never read, the documents you only see as bullet points, the decisions you let the model “just handle.”

There’s a version of this future where assistants like Poke and Copilot make us sharper by holding a mirror up to our own judgment: challenging our defaults, surfacing second-order effects, forcing us to articulate what we actually want. And there’s a lazier version where we become the human in the loop in name only, nodding along while the system steers.

The pricing ladder for AI judgment is being built in real time, from free browser extensions to six-figure text threads. The hard question — for companies, for policymakers, for each of us — is not how much we’re willing to pay. It’s how much of our judgment we’re willing to give away.