Claude Opus Plans, Locals Execute
A Reddit discussion argues for a split workflow: use Claude Opus as the high-level planner, then hand the plan to a smaller local model for implementation and QA. The thread suggests this works best when the frontier model writes for the worker model, not for a human reader.
The strongest version of this workflow treats Opus like an architect and the local model like an executor. When the handoff is explicit, smaller models can punch above their weight; when it is vague, they drift, bloat tasks, or miss acceptance criteria.
- –Ask Opus to produce an implementation contract, not a narrative: goals, constraints, dependencies, file-by-file scope, and a hard definition of done
- –Force decomposition into atomic stories with sequencing, so the local model never has to infer project structure from prose
- –Include validation steps up front: tests to run, failure modes to check, and what “good enough” looks like for each story
- –Optimize the plan for a weaker reader: shorter bullets, fewer abstractions, and explicit assumptions beat elegant human-style writing
- –The Linear/backlog framing is useful because it turns a conversation into schedulable work, which is exactly where smaller models tend to succeed
DISCOVERED
45d ago
2026-04-18
PUBLISHED
45d ago
2026-04-18
RELEVANCE
AUTHOR
droning-on