How Noda Is Driving LINE Automation to Zero Cost With Claude Code

Official LINE has long relied on auto-messaging tools that cost tens of thousands of yen per month, and operating multiple accounts could drive annual costs into the millions. Kazushi Noda decided to replace that stack with an in-house tool built around Claude, using Anthropic's Claude Code to create LINE Harness.

The finished product positions itself as a full replacement and upgrade over paid tools, with both subscription and operating costs set to zero. It even includes the ability to spin up an official LINE account quickly, opening the door for individuals and small businesses that previously gave up because of cost. The interviewer says Harness ended up covering a LINE build-out that would otherwise have required an order worth around 1 million yen.

Noda runs 11 parallel Claude Code instances on his PC using a screen-splitting setup. Whether it is implementing Harness, preparing workshops, or designing intake sheets, he inserts new tasks sideways the moment they occur to him and processes them in parallel. In his words, it is an operating model built to achieve "11x productivity."

He also lays out a concrete prompt-design technique. Beyond the official documentation, he has Claude read through competitor tool explainers, compare both sides, and surface features that exist in the official spec but are missing from competing products. That gives Claude enough context to suggest both differentiation points and feature priorities for the in-house tool. The approach is grounded in a simple philosophy: when instructing AI, give it the best possible source material.

He also shares an exercise he says every new workshop participant should do first: "Using your past ChatGPT conversation history, classify what AI can do entirely (green), what AI can also do (yellow), and what only you can do (red), using emoji." The goal is to reveal the questions people keep repeating unconsciously.

Why give it away for free? Noda says there is something inherently satisfying about increasing liquidity itself. He connects the idea to the shift from capped access at movie theaters and record stores to effectively unlimited access through services like Netflix, Apple Music, and Spotify. His core motivation is to use AI to dissolve constraints such as LINE tool pricing and clunky admin interfaces into something more fluid.

What had been nagging at him for years, he says, was a sense of exhaustion with the so-called "dopamine economy" of virality, advertising, and excessive entertainment consumption. Even its seeming opposite, like saunas, camping, and minimalism, eventually becomes repetitive. What remains is the layer of "making, understanding, and passing things on." He calls this new mode, where distribution itself becomes the reward, the "post-dopamine economy."

As AI makes professional-grade production available to everyone and social platforms fill with AI slop, Noda predicts a countertrend: a return to a world where a small number of people create and distribute genuinely valuable things. His goal with LINE Harness is to build that "place" ahead of the curve.

Near the end of the discussion, Noda points to two emerging technology currents: "human in the loop" and "agent-to-agent." The former describes a pipeline where AI handles planning through execution and asks humans for approval only when needed. He also references Adoness's task-management idea of not using "TODO" but "DOD (Definition of Done)," where the goal is framed around external conditions such as "the other party has replied."

On the latter, he explicitly mentions Google's A2A protocol. He imagines a world where each party's AI agent acts as a translator, helping with the most painful parts of multi-person projects, such as profit allocation and consensus-building. Noda says this could make it possible to bring in "difficult but highly capable people" through their agents. He also argues that video itself will become more multi-layered: instead of uploading one version and stopping there, subtitles, visuals, and difficulty could be transformed per viewer, changing the structure of content distribution.

The final long-range vision he mentions is a "world government." If AI continues pushing toward greater liquidity, he argues, even borders as constraints may gradually dissolve, moving the world closer to a model of regional specialization, though he adds that realization is still far off.