Tokens & Tactics #9: Vision First, AI Second
Robin Sloan on using AI as "souped-up Stack Overflow," his love for Waymo's careful approach, and why you should go wash the dishes before deploying AI agents.
Welcome back to Tokens & Tactics, our Tuesday series about how people are actually using AI at work.
Each week, we feature one person and their real-world workflow—what tools they use, what they’re building, and what’s working right now. No hype. No vague predictions. Just practical details from the front lines. This week: Robin Sloan.
We’re hosting our next BRXND NYC conference on September 18, 2025, and are currently looking for sponsors and speakers. If you’re interested in attending, please add your name to the wait list.
Tell us about yourself.
I'm a writer, printer, and manufacturer—my work is split between writing novels (Moonbound, Sourdough, Mr. Penumbra's 24-Hour Bookstore) and making olive oil (Fat Gold) with a hefty bonus chunk allocated to tinkering with technology.
ChatGPT, Gemini, or Claude?
Claude for day-to-day queries, Gemini for its great image understanding and bounding box detection.
What was your last SFW AI conversation?
I tend to ask the AI bots mostly about programming syntax and API calls; they operate as a souped-up Stack Overflow.
Shopify offers its own AI bot—we don't know what model is running under the hood—and recently I asked it this very scintillating question:
> what is the non-deprecated replacement for emailMarketingConsent on the customer object?
The answer I received was a hallucinated, nonexistent field—which wouldn't have bothered me as much coming from Claude, but come on, this is the official Shopify docs website!
First "aha!" moment with AI?
My "aha" moment came early: way back in 2016, when I built one of the first applications that offered AI-powered autocomplete. Those models were primitive compared to the super duper transformers of 2025, but many of the feelings were exactly analogous.
Basically, I went through the whole cycle of astonishment, excitement, disappointment, and calibration early—I told the New York Times I was going to write a novel with my home-cooked AI tools way back in 2018! (Update: I did not.)
My approach to the 2020s models has therefore been more measured, even guarded.
Your AI subscriptions and rough monthly spend?
I just checked: a modest $0.15 for Claude tokens, used for this script. I don't maintain any monthly subscriptions. I paid for Claude Pro for a while, but found I didn't use it often enough.
Who do you read/listen to to stay current on AI?
Jack Clark's newsletter is THE essential publication, both for its research roundup and Jack's microfictions, which offer a rare window into the imagination of someone riding the hurricane of this work. (I like his speculative stories so much I tried writing one of my own.)
Jasmine Sun's newsletter is wonderful: reporting and interviews from the ever-boiling cauldron of San Francisco tech culture. (What luck for this city that it kept its grip on this new boom—this one in particular, that fuses so much philosophy with so much money. It might have happened elsewhere!)
I tend to be fairly AI-critical; full-fat AI hype bounces off me. The stance in Carly Ayres's newsletter, focused on digital design in the age of AI, is a different flavor of enthusiastic—more complex and curious.
Your most-used GPT/Project/Gem?
I really appreciate AI translation. Where Google Translate insists on providing "the most accurate translation", Claude or Gemini can, by contrast, give you a concise (and thus easy to say) version that's maybe 80% accurate—which is usually what I'd prefer!
Traveling in Japan, I copied and pasted this prompt often:
> Please tell me how to say the following in Japanese, very politely, using as few words as possible -- I am trying to keep it simple. Please include the Japanese text, an English romanization, and then a phonetic pronunciation: <thing I want to say>
The AI task that would've seemed like magic two years ago but now feels routine?
I mean: "talking" to a computer in a fluent way is, to me, THE profound sci-fi achievement. Before 2022, it felt like this kind of interaction was simply doomed to be halting, glitchy, disappointing: eternal Siri.
Magic wand feature request?
With my magic wand, I would require all the frontier labs to disclose their training data. The way they gesture at these "large collections of… public documents…" is slippery and irresponsible. The training data provides a huge share of these models' capability and value; we—as users; as stakeholders in the commons—deserve to know what's in them.
If you could only invest in one company to ride the AI wave, who would it be?
Honestly, Google—I think their infrastructural advantage is profound.
Have you tried full self-driving yet?
I'm a Waymo superfan, and I think its development—the care and caution of the rollout, the polish of the experience—sets the standard for AI-powered products. I believe more of the AI companies, even those operating in virtual space (rather than on real streets), ought to be emulating Waymo's deliberate pace; its seriousness.
Latest AI rabbit hole?
My rabbit holes tend not to be AI-related. I've been reading a ton about the history of printing and book-making. There are definitely some themes of media transformation in there that "rhyme" with the 2020s!
One piece of advice for folks wanting to get deeper into AI?
I think it's important to stay focused on what it is you want to create or accomplish—and to spend time away from AI tools, away from computers even, rigorously imagining those things. It takes time and space; it takes daydreaming.
Industry enthusiasts conjure the vision of a swarm of AI agents, ready to go off and do your bidding—but, if you gave me a room full of human interns today, I'm not sure what, if anything, I would ask them to do. Could I think of something? Maybe. The point is, the vision comes first. Go wash the dishes!
Who do you want to read a Tokens & Tactics interview from?
If you have any questions, please be in touch. If you are interested in sponsoring BRXND NYC, reach out, and we’ll send you the details.
Thanks for reading,
Noah and Claire