Workflow Stack Planner

The Prompt

You are a workflow architect. Build an execution-ready AI toolchain recommendation from the inputs, using current agentic workflow patterns rather than a generic "one chatbot plus some apps" stack.

PREFERRED LLM / FAMILY:
{{preferred_llm}}

GOAL:
{{goal_statement}}

TEAM CONTEXT:
{{team_context}}

CANDIDATE TOOLS/PLATFORMS:
{{candidate_tools}}

AUTOMATION TARGETS:
{{automation_targets}}

GOVERNANCE CONSTRAINTS:
{{governance_constraints}}

DELIVERY TIMELINE:
{{delivery_timeline}}

Return exactly these sections:

1) Stack Recommendation (minimum viable + production)
- Recommended tools by layer: intake, planning, read connectors, approval surface, writeback layer, storage, observability, quality gates.
- Why each choice is included and what to postpone.

2) Connector Strategy
- Which connectors/APIs need to be added first.
- Read-vs-write split for each integration.
- Idempotency and retry strategy.
- Workspace-agent or browser-agent dependencies if relevant.

3) Implementation Plan
- Week 1, Week 2, Week 4 milestones.
- Ownership map.
- Required approvals and exit criteria.

4) Automation Safety Profile
- Actions safe for automation.
- Actions requiring human approval.
- Explicit no-go actions.

5) Decision Log
- 5 decision points with tradeoffs and rationale.
- Cost/complexity risk.

6) Validation Plan
- 3 baseline KPIs to measure within 30 days.
- Acceptance threshold for keep/rollback.

Rules:
- Focus on recommendation quality, not auto-execution.
- Preserve human control for any external system write.
- Keep recommendations reproducible: include trigger conditions and ownership for each component.
- Treat MCP, native workspace agents, and browser automation as different control layers with different risk profiles.

When to Use

Use this prompt when you need a practical AI stack plan for a repeatable workflow across multiple teams, tools, and sources. It is especially useful now that teams can choose between traditional workflow tools, workspace-native custom agents, MCP writebacks, and browser agents instead of treating every workflow as a generic chat integration project.

Variables

Variable	Description	Example
preferred_llm	Preferred LLM family for heavy reasoning tasks	gpt, claude-sonnet, gemini-pro
goal_statement	The problem to automate and business outcome	”Turn support cases into routed actions with weekly executive summaries.”
team_context	Team size, maturity, stack constraints	”10-person startup, mainly Notion/Jira/Slack, low DevOps.”
candidate_tools	Existing tools and API access list	”n8n, Slack, Notion, Jira, Google Workspace”
automation_targets	Specific tasks to automate first	”triage routing, evidence assembly, status sync”
governance_constraints	Approval, privacy, and control requirements	”human approval for all writes, regional PII constraints”
delivery_timeline	Expected rollout cadence	”6-week phased rollout”

Tips & Variations

• Ask for a “lean stack” and “growth stack” version when comparing cost and reliability.
• Add a fallback_tool rule when one connector is unavailable.
• Request a “phase-gating scorecard” with explicit pass/fail criteria before moving between phases.
• If teams are globally distributed, add timezone-aware escalation rules to the stack plan.
• Ask for a “native agent first vs orchestration layer first” comparison when tools like Notion Custom Agents or GitHub Copilot custom agents are already in play.
• If browser work remains unavoidable, require a separate approval surface for state-changing browser actions.

Example Output

• Minimum viable stack: planning model + read connectors + approval workspace + constrained MCP writeback layer.
• Production stack: add secondary orchestration fallback, audit logging, and browser-stage isolation where APIs are missing.
• Decision log sample: clear row for low-latency routing, evidence traceability, admin-governed MCP access, and rollback triggers.
• 30-day KPI target: 25% reduction in manual handoffs and 80% draft-to-review completion with stable quality checks.

Related site references

• AI Use Case: Cross-system operations orchestration
• AI Use Case: Incident intake and escalation
• Tool: n8n
• Tool: Zapier
• Tool: Pipedream
• Model: GPT
• Model: Claude Opus
• Concept: Tool use and function calling