What It Means Agent-native architecture is a style of building software in which every part of the product is accessible to an agent through tool calls — including the agent's ability to modify the app from within itself. In the channel's material, the term describes products designed so that AI agents work on them natively, rather than the product being usable only by humans clicking through an interface. Linear: Rebuilt for Agents The clearest example comes from Linear. After staying deliberately quiet during the GPT-3 moment, Linear eventually decided to rebuild how the product works in a significant way [^5525946f-790a-5dbf-869f-eaf4933bd18b@163]. The result: Linear made its platform agent-native, meaning it works natively with AI agents rather than only with humans clicking through a UI [^5525946f-790a-5dbf-869f-eaf4933bd18b@140]. That investment paid off immediately — when OpenAI's Symphony launched, the main thing it hooked into was Linear [^5525946f-790a-5dbf-869f-eaf4933bd18b@126]. Claude Co-work: Buttons That Are Prompts Anthropic's Claude Co-work shows the same idea from the product's inside. Error messages revealed Claude Code internals, and the host described the app as a strong example of agent-native architecture: an agent wired to the UI so buttons just send prompts to the agent [^c21e925e-a374-5583-a551-628601b7d1b9@1316]. Instead of each button triggering hardcoded behavior, the interface's actions are handed to the agent as tool calls — and the agent can even change the app itself, as when a PR shipped from feedback given live on the stream [^c21e925e-a374-5583-a551-628601b7d1b9@5265]. Why the Pattern Matters - It lets external agents plug in first: a product built agent-natively becomes the natural home for new AI tools, as Linear was for Symphony. - It blurs the line between user interface and agent interface: UI actions become prompts, and agents become users. - It enables self-modification: because the agent can act on the product through the same tool calls, the app can improve itself from within. - Agent-native architecture = every part of the product reachable by an agent through tool calls, including self-modification. - Linear rebuilt significantly to become agent-native — and became the first platform agents like OpenAI's Symphony hooked into. - Claude Co-work wires its UI to the agent so buttons are just prompts to Claude Code.