Once You Start Pedaling, You Can Go Far

Yoshida: tōku works alongside executives and leadership teams, going all the way to pinning down—in words—the value that will keep their organization needed by society into the future. We start from articulating and redefining things like purpose, mission, vision, and values, and carry that through the whole branding domain: design, expression, language.

Our name comes from our phrase "farsight and dialogue." Left alone, people get pulled by their own biases and day-to-day gravity toward what's near. The far view is something you can't see on your own—so you see it through dialogue. Our name tōku, "far" in Japanese, means exactly that. So my daily work depends on wide-ranging information gathering—what the world's most forward-looking people are thinking—and on continuously structuring and accumulating my own thinking.

We're a genuinely small firm: two people, recently founded. Our hypothesis is that smallness—lightness, fast decisions, agility—lets us deliver value the big players can't. But agility is the flattering word for it; our human capital is literally two. The question that never leaves us is how to use AI to compensate for that, keeping the lightness while maximizing output.

Yoshida: I was an early adopter of ChatGPT, but for about three years I stayed purely conversational. ChatGPT for a long time, then Gemini, and this March I finally created a Claude account—but I had what I'd call terminal-screen phobia (laughs). I assumed Claude Code meant living in a terminal, so I started with Claude Cowork, which promised I wouldn't have to touch one. I collected plausible-looking tips from social media and tried them, but I'd burn through tokens instantly, never get the output I imagined, and realize I was still basically giving conversational instructions. I felt I didn't understand the rules of the game at any fundamental level.

And unless I could really use AI, the "small team, maximum output" idea would never happen. A firm like ours has to build AI literacy at double or triple the level of a big-company businessperson just to break even. That takes a different order of learning intensity—and moving from chat tools to agent-style AI, I didn't even know what I didn't know. Resolving that kind of not-knowing on your own is possible, but it costs enormous energy and often ends up being the long way around.

Right now anything about AI gets attention, so plausible information pours in endlessly. Sorting the gems from the gravel and self-studying is possible in theory—but my client work is busy. Whenever I tried to pay that sorting cost on the side, the work that actually pays this month won. "I'll get to AI when things calm down"—and six months would pass. The person who says they'll get to AI once things calm down never actually calms down (laughs).

Yoshida: I found Wolkin through a Facebook post by a fellow business owner who was taking the AI coaching. I commented "what is this?" and he connected us.

The biggest factor was that the plan was made to order. Having failed at pecking away at generic best practices, I knew the crux is the connection between what's out there and my own circumstances—the way I actually want to use these tools. Wolkin looked like they would custom-build exactly that bridge.

Then there was unlimited questions—someone to be my harness for resolving the "I don't know what I don't know" problem. And honestly: they didn't seem scary (laughs). No harshness, no "how do you not know this." The character of the Wolkin team was part of the decision.

Yoshida: The clearest performance gains in my actual work came once we reached Claude Code. But would starting with Claude Code have been better? Not at all. Working through Cursor and Dify in the sessions first, I understood the mechanics—what vector search is, what it means to handle meaning in many dimensions. That gave me a foundation for meta-application: "it works this way, therefore doing X should produce Y." I'm the type whose later self-learning compounds much better when I understand the reasoning.

Frankly, the first two sessions were pretty unglamorous (laughs). Flashier exercises would probably score higher on immediate satisfaction, so I remember that they had the courage to build the curriculum starting from the plain parts.

That judgment came from the two interview sessions before the program started. Some people want the katsu curry brought out immediately; I'm the type who wants the tasting menu, built course by course. They clearly compose the meal differently for each client—so my advice to anyone starting is to use those first two hearing-and-proposal sessions for everything they're worth. I talked their ears off about my work and even what type of learner I am.

Yoshida: I tried to say, as much as I could, what I wanted to become through AI. I do accompaniment work myself, so I know: when client and coach share the goal, the coaching gets dramatically easier.

There's an obvious contradiction—"I don't know what I don't know, so how can I articulate a goal?" But that's exactly when you shouldn't be embarrassed. Don't worry about looking like someone who knows nothing about AI; over-expect, like it's a magic wand, and just say what you want regardless of feasibility. This is where Wolkin's not-being-scary pays off. With someone who talks down to you, "I don't want to look stupid" kicks in and you stop talking. With them, there's air in the room to expose your AI-poor self. Say everything, and they'll tell you what AI is genuinely good and bad at—the over-expectations get corrected in the actual sessions. After that you can lean on them.

Yoshida: It changed a great deal. The biggest use is consolidating, organizing, and managing scattered information. My work isn't point-by-point performance—engagements run two months at the shortest, a year or two at the longest. "What did that person say several meetings ago," the history of a decision—that accumulation matters. And every client is in a different industry, with different vocabulary. Take the single word "concept": its granularity and definition differ completely from one company to the next. Run a project using the same word with different meanings underneath, and it will inevitably meander—that's exactly why accumulation matters. "Where did that idea from the first meeting go?" Done by a human, that requires a talent for remembering. Done by AI with the right structure in place, it just works. Being able to lift all of that out of my brain's capacity was huge.

It now goes all the way to structured output—simple documents. And that "continuously structuring and accumulating my own thinking" I mentioned at the start isn't a side task for us; it's the product itself. So investing in AI wasn't really about efficiency—it was reworking the foundation of what we sell. If I dress it up a little: the share of my tasks that only a human can do has completely changed. It's healthy.

Yoshida: Not one moment in particular—but I actually appreciated that Wolkin shared, openly, that they are themselves mid-trial-and-error. When I asked "what should I do here?" or "is this possible?", they almost never answered with a flat "it's like this." Instead: there are many approaches, we used to do it that way, we've currently arrived at this, and we don't know where it goes next.

They did give me answers, of course. But more useful than any answer was the reasoning behind it—that's what feeds independence. At goal-setting they told me: the goal is for you to run your own learning loop and graduate. Not to get me hooked on Wolkin and unable to live without them—not "come again," but "you don't need to come back, you're okay now." The exchanges where that philosophy showed through are the ones I remember.

Yoshida: The lump in my throat—that "someday I have to do this" going round and round, always postponed—finally moved. It was my biggest open issue of the year.

AI is a road without end, but a bicycle is heaviest at the start. They stood on the pedals with me through the heaviest, highest-torque part—and now the wheel is turning and momentum is doing its work. It's not scary anymore, and it's not a chore.

Before, I could never protect a special time slot for "touching AI," so I went in circles. Committing to the sessions worked like going to a personal trainer: it got the bicycle up to speed. From here, I don't need to carve out separate AI time—the self-learning loop runs inside regular work and daily life. To go up another level I'd probably need dedicated hours, but I can keep pedaling this moving bicycle a long way. So for anyone who feels they can't find the time—anyone facing that heavy first stretch of pedaling—I'd strongly recommend it.