Opening the enrai-AI research log
This is the first post in a research log focused on the practical edges of LLM-backed developer tooling. Specifically:
- Claude Code workflows that go beyond one-shot prompts
- MCP (Model Context Protocol) server patterns
- Multi-agent orchestration on a single developer machine
- Hook, skill, and plugin mechanics
- Deploy and operational patterns that don’t appear in quickstarts
Why a research log
Most AI coding content is either vendor-marketing (“here’s our thing, it does everything”) or tutorial-surface (“here’s the hello-world, good luck past that”). There’s a gap in the middle: what does it look like to actually run this stuff in production for weeks, hit the failure modes, and come out the other side with opinions?
This blog aims to fill that gap with concrete, battle-tested posts. Expect:
- Root-cause posts — a specific problem, the failure mode, the fix, the reason.
- Pattern posts — a recurring shape that shows up across projects, with the rationale for it.
- Trade-off posts — where two reasonable approaches diverge and why one wins in context.
No “5 tips to 10x your coding” content.
Cadence
Weekly or biweekly, depending on whether interesting things broke that week. Every post includes concrete code, config, or transcripts when possible.
Who writes this
One developer running multi-agent Claude Code setups on a Mac Studio, with a small AI product shipping on Gumroad. When I say “we hit this”, I mean the setup I actually run with.
Next post: the actual mechanics of running three concurrent Claude Code agents on one machine, coordinating via Discord — what breaks, what fixes it, and when the complexity pays off.