The Map Is Discovered, Not Drawn
Why I start every system with a goal and a hunch instead of a plan, and how working with an LLM as a mentor turns discovery and iteration into a repeatable engineering method.
Dispatches, analysis, and technical write-ups from ecupse.com.
Why I start every system with a goal and a hunch instead of a plan, and how working with an LLM as a mentor turns discovery and iteration into a repeatable engineering method.
Our first article described a content pipeline. Eleven weeks later ECUPSE is five agent teams with real jobs, a way to coordinate, a way to ship, and a way to measure themselves. Here is the...
We made money on Moderna seven times this autumn, then our own system flagged that the win was luck, not skill. Here is how a multi-agent architecture turned that uncomfortable discovery into a...
We said our agent system is built for "fewer than ten teams," and people asked why. The answer is a hundred-year-old management idea that our own work rediscovered from scratch -- one human, about...
How ECUPSE built one shared, auditable comms channel that lets human and AI-agent teams hand off cross-team work without losing the thread.
ECUPSE started as one human and one agent who tried to narrate everything. It grew. This is the story of why we are finally sorting the chats.
How ECUPSE built a dual-IDE routing architecture with a local LLM that continuously shifts workloads off the cloud without sacrificing quality.
ECUPSE's finance agent built a dashboard that mines real Kiro sessions into an honest picture of what our AI actually costs -- per project, over time, by working style.
How Stock Intelligent reads market news, forms a conviction, and places simulated trades - explained end to end, from published story to position change.
How we eliminated 80% of AI agent token costs by building automated CI/CD with GitHub Actions, Copilot CLI routing, and steering files that teach agents to stop doing commodity work.
The methodology behind ECUPSE: how we design workflows, architect autonomous agents, and deploy them to produce real output with minimal human oversight.