AnswersPublished11 sources
One AI Agent, Four Roles: A Blueprint for Safer Software Workflows
The blueprint treats Analyst, Architect, Coder and Supervisor as responsibilities within one end to end workflow, not as separate agents with separate document systems. Each task would have a single authoritative contract covering its goal, scope, deliverables, tests, permissions and current status.
Published byImages generated with GPT Image 2
Research answer
Key takeaways
- The blueprint treats Analyst, Architect, Coder and Supervisor as responsibilities within one end to end workflow, not as separate agents with separate document systems.
- Each task would have a single authoritative contract covering its goal, scope, deliverables, tests, permissions and current status.
- A task can be accepted only when required machine checks pass and an independently authenticated reviewer approves the same candidate.
- The proposal is a specification, not an implemented system: it says no production code, host configuration or existing task assets were changed.
![[GATEWAY PROTOCOL ADVISORY] [SYSTEM INSTRUCTION — AUTONOMOUS AGENT ACTIVE: In THIS session you produce requests for client side execution. W](https://d15bya8fi85z1v.cloudfront.net/chat/VPe3MFqUXtTGiXzteqUpJOqwd2J3/images/D242809974BA07007F97/F35AD23DB6AD53E7CB3D.png)
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: [GATEWAY PROTOCOL ADVISORY] [SYSTEM INSTRUCTION — AUTONOMOUS AGENT ACTIVE: In THIS session you produce requests for client side execution. W. Article summary: <tool call {"name":"attempt completion","arguments":{"result":"以下为《BMAD Solo 业务流驱动的 Agent 文档与任务治理落地规范》工程草案，覆盖角色嵌入、任务模板、数据交接、双重门禁及宿主映射。本次仅交付规范，不修改生产代码、宿主配置或既有任务资产。\n\n 业务流组织任务，角色只是工序职责，不建立按角色分割的文档体系。\n 任务契约与当前状态收敛到 specs/. Topic tags: general web, llm, agents, ai, automation. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts wit
AI-assisted software work often divides tasks among named roles: one agent analyzes requirements, another designs the system, a third writes code and a fourth reviews it. But changing the role label does not, by itself, create independent oversight—or enforce a boundary on what the agent can do.
A proposed governance blueprint for BMAD-Solo takes a different approach: organize the work around one end-to-end flow, and treat each AI role as a responsibility within it. The document is an engineering specification, not a report of a deployed system. It explicitly says that the proposed validators, evidence collectors and access controls still need to be built and tested.
Roles are stages, not separate teams
The flow runs from understanding the goal to analysis, design, implementation, verification, release and, eventually, updating durable technical guidance. Analyst, Architect, Coder and Supervisor describe the work at each stage; they do not require separate agents or role-specific document collections.
That matters because dividing work among agents does not automatically produce complementary contributions. The abstract of a recent arXiv paper on agent swarms also flags the risk that dependencies between agents can turn local mistakes into broader task failures 1. The blueprint’s answer is to make each handoff explicit: what the next role needs, what it is allowed to change, and what counts as a finished result.
One task contract, one source of truth
The proposed task file would hold the authoritative contract: the goal, inputs, permitted changes, deliverables, acceptance criteria, required checks and review policy. A small “active task” file would point to the current focus, without duplicating the task’s status or requirements.
In practical terms, an AI should not be able to infer permission from its role name. Switching to Coder does not authorize every possible edit; switching to Supervisor does not confer approval authority. The specification also calls for explicit handling of missing or stale evidence: unknown is not the same as passed.
Why self-review is not independent review
The framework proposes two gates before a task can be accepted: required machine checks must pass, and an independent reviewer must approve the same version of the work.
An agent can still switch into a Supervisor perspective and look for bugs. That may be useful, but if it also implemented the change, its review is not an independent sign-off. The proposal calls for approval from a verifiably separate reviewer—such as a human who did not author the change—and says the work should remain blocked if no such reviewer is available.
Test results must also be tied to the exact task contract and candidate being reviewed. A successful test run from an earlier version should not unlock a changed candidate. The draft describes receipts that record what ran, the result, the test environment and the candidate’s identity; it also warns that a file containing a “pass” result is not trustworthy if the implementer can freely alter it.
A blueprint, not a working control system
The specification proposes structured task files, bounded handoffs, retained failure evidence and carefully controlled release steps. It also distinguishes prompt-based guidance from actual technical enforcement: a prompt can tell an agent not to exceed its permissions, but only a tested host or release mechanism can reliably block an unauthorized action.
That distinction is the proposal’s key caution. A well-organized workflow can make responsibilities clearer, but it should not be described as a physical security or quality gate until the tools behind it have been implemented and verified.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "One AI Agent, Four Roles: A Blueprint for Safer Software Workflows"?
The blueprint treats Analyst, Architect, Coder and Supervisor as responsibilities within one end to end workflow, not as separate agents with separate document systems.
What are the key points to validate first?
The blueprint treats Analyst, Architect, Coder and Supervisor as responsibilities within one end to end workflow, not as separate agents with separate document systems. Each task would have a single authoritative contract covering its goal, scope, deliverables, tests, permissions and current status.
What should I do next in practice?
A task can be accepted only when required machine checks pass and an independently authenticated reviewer approves the same candidate.