Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
Tokenmaxxing is for Fools - Freestyle Friday (May 1, 2026)
TL;DR
Joe Reis says “token maxxing” is mostly fake productivity — after speaking at DeepLearning.AI’s Dev and AI event in San Francisco with roughly 3,000 attendees, he argues that constantly running agents and chasing every new tool often feels productive while producing a “graveyard of projects.”
His antidote is deliberate constraint, not more AI — for that SF trip, he brought only a reMarkable tablet, a phone, and a hardcover book, leaving the laptop behind so he could think in “slow but deep cycles” instead of shallow, rapid-fire iteration.
AI is useful, but the real differentiator is problem depth — Reis says agents can amplify you like an “infinite number of employees,” but if you’re a bad manager or don’t understand the problem deeply, more agents just multiply mediocre output.
The skills that actually keep people from being “left behind” are still the basics — he names reading, math, communication, selling, and negotiation as more important moats than staying current on every model release, point solution, or IDE workflow.
He thinks the industry will look back on this burnout phase and ask, ‘what the hell were we doing?’ — echoing a conversation with DBT Labs CEO Tristan Handy, he compares today’s AI obsession to running faster on a treadmill or hamster wheel without actually going anywhere.
His practical advice is a Pareto-rule approach to AI adoption — use the 20% of tools that get you 80% of the value, ignore the doom-mongering and grifters, and focus on building non-incremental, meaningful things the world actually needs.
The Breakdown
A panel in San Francisco surfaces the real anxiety
Reis opens from Salt Lake City after a week in San Francisco at DeepLearning.AI’s Dev and AI event, which he says had about 3,000 attendees. On a panel with leaders from Replit, Landing AI, Oracle, and moderator Silicon Valley Girl, the big question was whether software developers still have a future — and the vibe he picked up was a lot of fear that if you’re not using AI nonstop, you’re toast.
Why he’s pushing back on “token maxxing”
He takes aim at the internet mantra that everything has to be “maxed,” including AI usage, and says he’s moving in the opposite direction: token minimizing. His example is personal and concrete — for the trip, he traveled with a reMarkable tablet, his phone, and a hardcover book, no laptop, because the constant pressure to have agents running all the time was burning him out.
The hidden cost of letting agents run wild
Reis isn’t anti-agent at all; he says he uses agent workflows across his business and life and loves the speed. But he also describes the all-too-familiar mess: you kick off Claude Code, Codex, or some IDE workflow, wander off, and come back to a half-forgotten idea in a graveyard of generated projects. The point isn’t that the tools are bad — it’s that speed can create the illusion of progress without the substance.
Tristan Handy, treadmill productivity, and doing things at human speed
He brings in a fresh conversation with Tristan Handy, co-founder and CEO of DBT Labs, who agrees this whole phase may look absurd in hindsight a year from now. Reis’s metaphor lands cleanly: AI overuse can feel like running faster on a treadmill or hamster wheel — intense motion, same place. What actually helps him think is the opposite: walks with just an Apple Watch, leaving the phone behind, and Tristan’s own low-screen setup at home, sometimes using only voice ChatGPT.
AI was supposed to free us up — instead we’re working 15-hour days
This is where the rant gets philosophical. Reis references John Maynard Keynes’s 1930 letter imagining a future of 15-hour workweeks, then points out the modern irony: people now work 15 hours a day while supposedly being “freed” by technology. For him, the novelty of raw AI speed has worn off; what matters now is “brain max” — making higher-quality decisions, not just more outputs.
Real leverage comes from depth, not more incremental slop
He says the world doesn’t need more incremental solutions, and that his own book took a long time because he didn’t want to write the shallow, obvious version about star schemas and data modeling. The work that matters is the work that doesn’t exist yet, and AI should help people build that future — not crank out endless clever-but-forgettable iterations.
The people who win will be the ones with real moats
Reis is especially blunt about the “you’ll be left behind” narrative: the true risks are weaker fundamentals like poor reading, math, communication, selling, and negotiation. The person to fear competitively, he says, is the one who understands a domain and problem so deeply that it becomes a moat; anybody can ask Claude to “make me a billionaire,” but not everybody can do the hard human work agents still can’t.
Final message: stop doomscrolling, use Pareto, go live your life
He closes by telling viewers, basically, you’re going to be fine. Ignore the “artists and grifters” trying to manufacture FOMO, use the Pareto rule to get 80% of the value from 20% of the tools, and stop acting like you need to be glued to a screen 14 hours a day. Then he signs off with upcoming travel — Sweden for Data Innovation Summit, AI Council, London for Confluence Current, Detroit — while making it clear he also plans to chill this summer.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
