Designing inside a shared brain: lessons from an agentic workflow
What working inside an agent-driven workflow taught me about context, speed and what we trade for it.
Understand how our agentic workflow works, and what it means for design
Every ticket makes the shared brain a little smarter.
In March, I went deeper into how the agentic workflow on my team actually works. Not just using it, but understanding the thinking behind it.
At its simplest, the idea is this: pull knowledge from every source into one place, a single repository, and organize it into structured context files. When an AI agent picks up a task, it can find what it needs quickly instead of guessing.
A few principles shape how it runs:
- Humans guide and confirm. People set direction and sign off on the process. The agent does the execution.
- Every ticket feeds the knowledge base. Each piece of work is meant to improve the product, so it makes sense for that work to improve the shared knowledge too.
- Structure matters as much as content. Context only helps if the agent can find it fast, so how it's organized matters as much as what's in it.
Over time, the repository becomes a kind of shared brain. The more the team contributes, the better the agent gets at helping all of us.
What worked
- Speed. Nothing about the system surprised me more than how much faster work moved.
- Context that compounds. Because every ticket adds to the knowledge base, the system gets more useful the more we use it.
- A more cohesive team. Shared context means people across teams start from the same understanding.
What didn't
- Craft takes a back seat. When the agent handles execution, learning the craft of engineering gets pushed aside in favor of speed. That's fine in an intense sprint, but it's not a good long-term trade.
- The picture is still filling in. The knowledge base leans toward engineering today. Bringing in more design context, like research and the reasoning behind decisions, is the next opportunity.
Every ticket makes the shared brain a little smarter.
As someone who sits between design and engineering, I care about shared understanding, and a shared brain is that idea built into a tool. I can see real benefits in design adding its own knowledge: research, decisions and the reasons behind them. That's exactly the why engineers so often don't get.
It also changed how I work outside my job. I'm building the same habit in my Obsidian vault. Every learning, engineering or design, goes in and gets tagged, so when I connect Claude to it, the agent can reference all of my thinking. A personal shared brain, with one contributor.