模块 2 · The Offering Landscape: Who Plays Where
第 3 节,共 5 节 · 5 分钟 · 英语 · 只读
This lesson places two overlapping SaaS categories on the comparison matrix you built earlier: orchestration platforms and AI workflow automation suites. Both chain AI actions, but they differ in what they treat as the primary unit of design — a coordinated set of agents, or a trigger-action pipeline. By the end you should be able to name one shared capability and one distinguishing feature for each category.
Symphony's baseline is a control plane: it monitors a tracker board, spawns isolated agent runs, and reports proof of work back. Two SaaS categories overlap with that orchestration layer.
Orchestration platforms treat a coordinated set of agents as the primary unit. The design question is which agent handles which task, how work is routed between them, and how their outputs are combined. The supplied description of GitHub Copilot cloud agent is an example of this shape: you can create multiple custom agents that specialize in different types of tasks, and the platform automates branch creation, commit message writing, and pushing so developers can let the agents work in the background.
Workflow automation suites treat a trigger-action pipeline as the primary unit. The design question is what event starts the pipeline, which steps run in order, and what each step produces. The supplied description of GitHub Copilot for Linear is an example of this shape: assigning a Linear issue to Copilot triggers a fixed sequence — analyze the issue and open a draft pull request, work in an ephemeral environment, stream progress updates back to the Linear activity timeline, then request a pull request review when work is completed.
The boundary is the primary unit of design, not the presence of AI. Both categories use AI steps; they differ in whether the agent or the workflow is the thing being composed.
来源支持
来源支持
Two requests arrive. Request A: a customer-facing feature needs three specialized agents — one to research the repository, one to write the implementation, one to run tests — and their outputs must be combined before a pull request is opened. Request B: every time a documentation issue is filed, the same four steps should run in order: analyze the issue, open a draft pull request, run linters, request review. For each request, decide which category it maps to and state one shared capability and one distinguishing feature.
Start with the primary unit of design, because that is the boundary between the categories.
Request A: the designer's job is to decide which agent handles which task and how their outputs are combined. The three agents have specialized roles, and the coordination between them is the thing being designed. That is the orchestration-first shape. The supplied description of GitHub Copilot cloud agent supports this: you can create multiple custom agents that specialize in different types of tasks, and Copilot automates branch creation, commit message writing, and pushing so developers let the agents work in the background.
Request B: the designer's job is to fix the sequence of steps and the trigger that starts it. The steps are the same every time, and the trigger is the filing of a documentation issue. That is the workflow-first shape. The supplied description of GitHub Copilot for Linear supports this: assigning a Linear issue to Copilot triggers a fixed sequence — analyze the issue and open a draft pull request, work in an ephemeral environment, stream progress updates back to the Linear activity timeline, then request a pull request review when work is completed.
Shared capability for both: tracker-driven dispatch of AI work to isolated runs. Symphony's baseline does the same thing — it monitors a Linear board for work and spawns agents to handle the tasks, and the agents provide proof of work such as CI status and PR review feedback.
Distinguishing feature for Request A: multiple custom agents that specialize in different task types. Distinguishing feature for Request B: a fixed step sequence triggered by assigning an issue, with progress streamed back to the tracker.
来源支持
想一想
Pick one capability from the shared row of the matrix and describe a feature that would look similar but actually belong to the orchestration-first category rather than the workflow-first category. What in the feature description tells you which side of the boundary it falls on?