Review and repair
Bug fixing, pull-request review, local change review, and generated review guidelines.
Agentic engineering · VS Code
AI Engineer Micy
A multi-agent engineering participant that brings specialist quality, security, reuse, testing, planning, and Azure DevOps workflows into the developer’s working environment.
4specialist review lenses coordinated by a shared orchestrator
8+engineering workflows spanning fixes, review, analysis, planning, setup, and S360
2enterprise authentication paths: Microsoft SSO and Azure DevOps PAT
1developer surface: a conversational participant inside VS Code
System architecture
Specialists coordinate; developers stay in control.
Micy receives intent through VS Code, builds repository and work-item context, routes the task across relevant specialist agents, then synthesizes actionable output for review or execution.
Workflow coverage
From diagnosis to governed delivery.
The participant turns specialist analysis into concrete developer workflows rather than presenting a generic chat surface.
Understand and plan
Repository analysis and task planning turn broad intent into evidence-grounded execution steps.
Enterprise operations
Setup automation, S360 scenarios, Microsoft identity, and Azure DevOps integration connect engineering to operational context.
Technology foundation
Native to the enterprise developer workflow.
Micy combines the VS Code extension platform with typed orchestration, specialist-agent coordination, Microsoft identity, and Azure DevOps context.
VS CodeChat participant and extension host
TypeScriptTyped extension implementation
Semantic KernelMulti-agent orchestration
Azure DevOpsPRs, work items, and S360
Microsoft IdentitySSO and enterprise access
Node.jsExtension runtime and SDKs
Solution fit
Coordinated specialists beat one generalist pass.
A single model reviewing a change returns one blended opinion, in a window away from the code, with no view of the work item behind it. Micy separates the concerns into explicit lenses, runs them where the developer already works, and keeps them attached to the enterprise system of record.
Review that goes past style
Quality, security, reuse, and testing run as four distinct lenses under a shared orchestrator, so a security finding is never averaged away by a formatting comment in a single blended response.
Engineering attached to the system of record
Azure DevOps pull requests, work items, and S360 scenarios are inputs to the analysis rather than context a developer has to paste in by hand, and access runs on Microsoft identity.
Advice that turns into executed work
Eight-plus structured workflows cover bug fixing, pull-request and local change review, repository analysis, task planning, and setup automation — inside VS Code, in the repository already open.
| Capability | Typical market approach | AI Engineer Micy |
|---|---|---|
| Review depth | One generalist model pass produces a single blended opinion. | Four specialist lenses — quality, security, reuse, testing — coordinated by a shared orchestrator, each judged on its own criteria. |
| Where the developer works | A separate portal or web app the developer has to context-switch into. | A conversational participant inside VS Code, operating on the repository already open. |
| Enterprise context | Code only; the assistant cannot see the work item, the pull request, or the compliance scenario. | Azure DevOps pull requests, work items, and S360 scenarios are first-class inputs. |
| Identity and access | Individual API keys held by individual developers. | Microsoft SSO or Azure DevOps PAT, so access follows the identity the organization already governs. |
| Output shape | Free-text suggestions the developer still has to translate into work. | Eight-plus structured workflows spanning fixes, review, analysis, planning, and setup. |
| Tunability | Prompt behaviour is fixed inside the vendor’s product. | Typed Semantic Kernel orchestration, so specialist prompts and review policy can be recalibrated per team. |
Comparison note: the market column describes the default posture of general-purpose coding assistants rather than any named vendor product. Workflow and capability counts are supported by the repository documentation; no productivity or quality improvement is claimed without a controlled benchmark.
Related systems
Measurement note: repository documentation supports workflow and capability counts shown here. No numeric productivity or quality improvement is claimed without a controlled benchmark.