Workflow Builder
The Workflow Engine is 80% built. It can match capabilities, orchestrate execution, score confidence, and route escalations. It can do all of this through an API that exactly zero business users will ever touch.
An engine without an interface is a secret. This is the interface.
The Job
When a business user needs to create, visualise, or adjust an AI-powered workflow — help them build a functional workchart in under 10 minutes, without writing code, without filing a ticket, without waiting for a developer.
| Trigger Event | Current Failure | Desired Progress |
|---|---|---|
| Ops Manager needs to visualise a workflow | Opens a spreadsheet. Draws boxes. Nothing connects to the actual engine. | Drags nodes on a canvas. Each node maps to a real engine capability. |
| Business Analyst wants to create a new workflow | Writes a requirements doc. Waits for a developer. Ships in 6 weeks. | Creates the workflow visually. Tests it. Ships in 10 minutes. |
| Process Owner needs to adjust confidence thresholds | Asks engineering to change a config value. Gets it next sprint. | Moves a slider. Threshold updates immediately. |
| Anyone needs to monitor execution | Checks Slack for updates. Searches logs. Asks someone. | Opens the dashboard. Sees every execution, every confidence score, every escalation. |
| New team member needs to understand a process | Reads a 40-page SOP that's already outdated | Opens the workchart. The process is the documentation. |
The job: "Let me see my automation, build my automation, and trust my automation — without engineering help."
The hidden objection: "Visual builders produce toy workflows that break in production." The answer: this builder doesn't generate workflows from scratch — it composes capabilities from the Workflow Engine's Skill Registry. Every node is a proven, tested, scored capability. The builder is a composition layer, not a code generator.
The Gap
| Layer | Status | What It Means |
|---|---|---|
| Workflow Engine backend | ~80% complete | Skill Registry, Capability Matcher, Orchestrator, Confidence Scorer — all functional |
| Workflow Builder UI | 0% complete | Zero visual interface. All engine power accessible only through API. |
| Business user access | None | Every workflow change requires a developer. Every threshold adjustment is a ticket. |
80% of the power. 0% of the access. That's not an engineering gap — it's a business bottleneck. The engine is a sports car with no steering wheel.
What It Looks Like
Think process.st for procedures. Zapier for integrations. But for orchestrating AI capabilities with confidence scoring and human escalation built in.
Canvas
| Element | What It Does | Maps To Engine |
|---|---|---|
| Skill Node | Drag a registered capability onto the canvas | Skill Registry entry |
| Connection Line | Draw dependencies between nodes | Orchestrator sequence |
| Confidence Gate | Set threshold — green/yellow/red zones | Confidence Scorer config |
| Escalation Branch | Route low-confidence outputs to a human | Escalation Router rule |
| Trigger Node | Define what starts the workflow | Task intake definition |
| Output Node | Define what the workflow produces | Work Receipt template |
Panels
| Panel | Purpose |
|---|---|
| Skill Library | Browse registered capabilities. Search. Filter by category. Drag to canvas. |
| Properties | Select any node. Configure inputs, outputs, thresholds. |
| Execution Monitor | Live view of running workflows. Confidence scores. Escalation status. |
| Audit Log | Historical view. Every execution. Every decision. Every score. |
Feature / Function / Outcome
| # | Feature | Function | Outcome | Status |
|---|---|---|---|---|
| 1 | Drag-and-drop canvas | Visual workflow composition from registered skills | Business users build workflows without code | Gap |
| 2 | Skill Library panel | Browse, search, and drag engine capabilities onto canvas | No hidden capabilities — everything the engine can do is visible | Gap |
| 3 | Connection drawing | Link nodes to define execution sequence and dependencies | Workflow logic is visual, not buried in config files | Gap |
| 4 | Confidence threshold UI | Slider controls for per-node and per-workflow confidence gates | Process owners adjust risk tolerance without engineering tickets | Gap |
| 5 | Escalation branch editor | Visual configuration of human escalation routes | Clear paths — when AI stops, which human starts | Gap |
| 6 | Execution monitor | Real-time dashboard of running workflows with live confidence scores | "What's happening right now?" answered in one glance | Gap |
| 7 | Audit log viewer | Historical record of every execution with filtering and search | Compliance and trust — every decision is visible | Gap |
| 8 | Workflow templates | Pre-built workflow patterns for common business operations | New users start from proven patterns, not blank canvases | Gap |
| 9 | Test mode | Run a workflow with sample data before going live | Catch errors before they reach production. Build confidence in the builder. | Gap |
| 10 | One-click publish | Deploy a workflow from canvas to live engine execution | Build → test → ship in minutes, not sprints | Gap |
Anti-Goals (v1)
What the Workflow Builder deliberately does NOT do in the first version.
| Anti-Goal | Why Not |
|---|---|
| Code editing | The builder composes capabilities. If you need code, you're building a skill, not a workflow. |
| Skill marketplace | Skills are registered in the engine. The builder consumes them. Marketplace is a v3 problem. |
| Version control | v1 ships with save/load. Git-style branching and diff is complexity the builder doesn't need yet. |
| Collaborative editing | One builder, one workflow. Multi-user editing is a v2 problem after single-user proves the model. |
| Mobile-first design | Workflow building is a desktop activity. Canvas manipulation needs screen real estate. |
Saying no to these is the discipline. Each one is a real feature that real users will request. None of them is necessary to prove the core hypothesis: business users can create functional workflows in under 10 minutes.
Business Dev
The Builder is how the Engine reaches business users. Without it, the Engine sells to developers only.
| Layer | Decision | Current Hypothesis | Validation Signal |
|---|---|---|---|
| ICP | Who builds workflows? | Ops Managers and Business Analysts at Stackmates ventures | 3 non-technical users create a functional workflow without developer help |
| Offer | What does this enable? | "Build AI-powered workflows visually. Ship in 10 minutes, not 6 weeks." | Time-to-workflow under 10 minutes for a standard business process |
| Channel | How does adoption happen? | Internal-first — every venture's workflows built through the Builder | New workflows created through Builder, not through API calls or tickets |
| Conversion | What proves it works? | A workflow built in the Builder executes successfully in the Engine | First user-built workflow produces a valid work receipt with confidence score |
| Retention | Why does it stick? | The workchart IS the documentation. Leaving means losing process visibility. | Weekly active builders > 60% of workflow owners |
| Expansion | How does it compound? | Templates from one venture seed workflows in another | PrettyMint workflow template adopted by HowzUs without modification |
Commissioning
| Component | Schema | API | UI | Tests | Status |
|---|---|---|---|---|---|
| Drag-and-drop canvas | Pending | Pending | Pending | Pending | 0% |
| Skill Library panel | Pending | Pending | Pending | Pending | 0% |
| Connection drawing | Pending | Pending | Pending | Pending | 0% |
| Confidence threshold UI | Pending | Pending | Pending | Pending | 0% |
| Escalation branch editor | Pending | Pending | Pending | Pending | 0% |
| Execution monitor | Pending | Pending | Pending | Pending | 0% |
| Audit log viewer | Pending | Pending | Pending | Pending | 0% |
| Workflow templates | Pending | Pending | Pending | Pending | 0% |
| Test mode | Pending | Pending | Pending | Pending | 0% |
| One-click publish | Pending | Pending | Pending | Pending | 0% |
Status: 0% built. The Workflow Engine backend is ~80% complete but entirely inaccessible without this interface. This is the highest-leverage UI build in the Mycelium network — it unlocks everything the engine already does.
Metrics
| Metric | Target | Why It Matters |
|---|---|---|
| Time-to-workflow | < 10 minutes | North star — if it takes longer, the builder is too complex |
| Non-technical builder rate | > 70% of workflows created by non-engineers | The whole point — business users build, not developers |
| First-attempt success rate | > 60% of workflows execute correctly on first publish | Usability — builders trust what they build |
| Template adoption rate | > 40% of new workflows start from a template | Leverage — proven patterns compound |
| Execution monitor daily views | > 80% of workflow owners check daily | Trust — people use what they can see |
| Workflow count growth | +5 workflows/month per active venture | Platform compounding — more workflows, more value |
Risks + Kill Signal
| Risk | Mitigation |
|---|---|
| Canvas is too complex — users can't figure out the interface | Start with templates, not blank canvas. First experience is "modify a working workflow," not "build from nothing." |
| Workflows built visually don't match engine capabilities | Builder nodes map 1:1 to Skill Registry entries. No phantom capabilities. What you see is what the engine can do. |
| Process.st / Zapier already solve this | They orchestrate tools, not AI capabilities. They don't score confidence or route escalations. Different design center. |
| Builder becomes a bottleneck — engine evolves faster than UI | Skill Library auto-syncs with Skill Registry. New engine capabilities appear in Builder without UI changes. |
| Non-technical users create bad workflows | Test mode catches errors before production. Confidence gates catch bad outputs after. Two safety nets, not zero. |
Kill signal: If average workflow creation takes more than 30 minutes after onboarding, the builder is too complex. The entire value proposition is speed — business users shipping automation faster than they can file a Jira ticket. If the builder can't beat the ticket queue, it's not a tool. It's another interface nobody asked for. Kill and simplify.
Mycelium Capability
This is the visual workflow editor — the interface layer that makes the Workflow Engine accessible to anyone who can drag a box and draw a line.
| Venture | Uses The Builder For | Priority |
|---|---|---|
| Stackmates | CRM workflows, proposal pipelines, client reporting automation | Primary — first consumer |
| Dreamineering | Content pipeline workflows, research-to-article automation | High |
| HowzUs | Property assessment flows, compliance check sequences | Medium |
| PrettyMint | Product listing workflows, inventory management automation | Medium |
| BerleyTrails | Trail data processing, recommendation generation flows | Medium |
| TouchForFun | Session planning workflows, prompt generation sequences | Lower |
| BetterPractice | Protocol recommendation flows, progress tracking automation | Lower |
Engine without Builder = power without access. Builder without Engine = interface without intelligence. They ship together or they ship nothing.
Dependency
The Builder depends entirely on the Workflow Engine. Every node in the canvas maps to a Skill Registry entry. Every connection maps to an Orchestrator sequence. Every confidence gate maps to a Confidence Scorer threshold. The Builder doesn't generate logic — it composes engine capabilities visually.
WORKFLOW ENGINE (backend) WORKFLOW BUILDER (frontend)
───────────────────────── ────────────────────────────
Skill Registry ← reads ← Skill Library panel
Capability Matcher ← feeds ← Task-to-skill suggestions
Orchestrator ← maps ← Canvas connections
Confidence Scorer ← sets ← Threshold slider UI
Escalation Router ← config ← Escalation branch editor
Audit Trail ← shows ← Audit log viewer
Work Receipts ← shows ← Execution monitor
Build sequence: Engine first (in progress). Builder second (this PRD). Together they form the complete automation capability for the Mycelium network.
Context
- Phygital Mycelium — The capability network this builder belongs to
- Workflow Engine — The backend brain this builder makes accessible
- ETL Data Tool — Upstream: engine needs clean data, builder needs to show it
- Data Interface — Data access layer the engine queries through
- Commissioning State Machine — Progressive maturity model for tracking builder readiness
- Sales CRM & RFP — First workflow consumer: CRM is the proving ground
- Mushroom Caps — The ventures that build workflows through this interface
- Jobs To Be Done — Demand-side thinking: what progress, not what features
- Flow Engineering — Maps that produce code artifacts
- Standards — Where proven workflow patterns graduate to