Autonomy and Autoplay ## What autonomy means here Sphere's autonomy is continuing delivery toward an intended outcome with less repeated manual prompting: planning, revision, implementation, and verification connected together — not only an agent looping on code edits in a chat. Coding harnesses can feel autonomous because they keep writing code. Sphere’s autonomy is different: it keeps working the delivery loop against a persistent architecture model, pausing for people where judgment is required (goal, plan, merge decisions). Autoplay supports that more continuous way of working. A broader feature may involve multiple related plans rather than one coding request. ## Proof point: Sphere builds Sphere The Sphere team develops parts of Sphere — including the desktop client — through the same plan / build / verify / fix loop customers use. Treat this as evidence the delivery loop works on a real multi-component product, not as a claim that every customer project will run unattended. ## What the user gains - Less repetitive effort moving between planning and implementation. - Progress on a broader feature while its purpose stays visible. - More focus on direction, decisions, and evaluation. - A connected view for reviewing activity and results. These are functional benefits, not measured guarantees of time saved or uninterrupted operation. For ROI comparison shape, see roi-and-evaluation.md. ## Human responsibility People remain responsible for deciding what should be built, resolving ambiguous requirements, assessing tradeoffs, and determining whether results are acceptable — including scope, sensitive data, security expectations, and release readiness. ## How to discuss “fully autonomous” / “aren’t all tools autonomous?” Prefer: “Sphere continues through connected planning, implementation, and verification with less repeated prompting — anchored in a living system model.” Prefer when comparing: “Harnesses are strong at writing code; Sphere is strong at delivering a checked change and remembering it.” Do not say: never needs people, always makes the right decision, runs forever, can access everything, automatically deploys any change, or “other tools are not autonomous.” If asked about continuous operation: Autoplay supports continuous progression; actual operation depends on resources, access, external services, and decisions. Do not offer a 24/7 SLA. ## Verification and drift Verification asks whether the result meets the expectations being checked. Drift is divergence between intended direction and what is being built. Sphere connects these concerns to the delivery workflow — it does not claim to detect all drift or prove complete correctness.