Goal
Set accurate expectations for the AI features before you rely on them.Prerequisites
- Familiarity with the StackShift AI agents
Workflow
1
Use agents for the tasks listed as fully live below.
2
Account for the current limitations before depending on an agent for production work.
3
Keep human operational judgment and direct logs in the loop.
What is fully live
- Stackie plus the Deploy, Database, Debug, Ops, and WordPress specialists.
- Automatic build-failure diagnosis (pattern matching for everyone; LLM analysis on paid plans).
- Durable threads, runs, events, checkpoints, handoffs, approvals, and tool execution records.
Current limitations
- LLM-backed build diagnosis requires an active paid plan.
- Destructive, costly, external, secret-bearing, and production-impacting actions require approval immediately before execution.
- A documentation answer cannot establish live resource state, and an operational tool name cannot establish a product capability.
- Pull requests require approval and are never merged or deployed automatically.
Operator-first by design
These boundaries are deliberate. When an agent summary disagrees with documentation, logs, status records, or tool output, trust the underlying evidence and verify that the relevant operation reached a successful terminal state.Expected result
You rely on the AI features with accurate expectations and know where the guardrails are.
Related guides
StackShift AI agents
Stackie and five direct-entry specialists perform evidence-based work through durable runs, tenant-scoped tools, and approvals bound to the exact action and target.
AI build diagnosis
Failed builds combine deterministic pattern evidence with governed model analysis and can hand bounded repair work to Deploy or Debug.