Sorting a boundary task: a multi-day coding workflow
Example
Givens: a person wants to hand Muse a multi-day coding workflow — writing code, running it, and iterating until a feature is finished. The sources state that Muse Spark supports tool-use and that Muse can compile code the agent writes, and they also state that coding workflows are an area of current performance gaps. Question: does this task fall on the capable side or the unreliable side, and on what evidence?
Walkthrough
Step 1 — Check whether the source names the capability. The sources describe Muse Spark as supporting tool-use and describe the VM as having enough storage, CPU, and memory to do real work "like compiling code the agent writes." So code execution is inside the described scope. Step 2 — Check whether the source names a gap for this task shape. The same sources state that Meta continues to invest in areas with current performance gaps, "such as long-horizon agentic systems and coding workflows." A multi-day workflow is long-horizon, so the gap applies directly. Step 3 — Combine the two findings rather than picking one. The capability is real and the gap is real; they describe different aspects of the same task. Step 4 — State the result with the boundary intact. The task is not out of scope, but it sits on the unreliable side of the boundary because the source names long-horizon coding workflows as a current gap. The defensible conclusion is that Muse can work on this task while the person reviews its activity, not that it will finish the feature reliably.
Source support
- Introducing Muse Spark: Scaling Towards Personal Superintelligence(ai.meta.com · approved excerpt)
- How We Built Safety Into Muse | Meta AI Research(research.meta.ai · approved excerpt)
- How Muse handles your privacy, safety and security | Meta Help Centre (en-gb)(meta.com · approved excerpt)