Workflow Tier Classifier
Paste a product description or pitch deck excerpt and get back a plain-English verdict: are you decorating a model or owning a workflow? The classifier names your tier, identifies the missing layers, and tells you exactly what to build next.
When to trigger
- User says "workflow-tier-classifier", "classify my AI product", "am I decorating a model", "what tier is my deployment", "workflow tier"
- Quick self-assessment before a pitch or proposal
- Reviewing a competitor product or prospect's existing AI system
- As a public-facing lead-magnet entry point before a full
impl-arch-audit
The Five Tiers
| Tier |
What it means |
| Decoration |
Model access bolted onto a product. No deployment layer. The AI can be removed and the product mostly still works. |
| Wrapper |
Some structure around the model — a prompt, an API call, maybe a UI. But no workflow handoffs, data scoping, or authority limits. |
| Half-built |
Meaningful implementation (workflow design + data access at minimum), but missing authority enforcement, evaluation, audit trail, or recovery. Would fail a serious enterprise procurement review. |
| Workflow Owner |
Five of six layers present and structurally enforced. One gap remaining — typically evaluation rubric or recovery procedure. |
| Full Stack |
All six layers owned: workflow design, data access, authority enforcement, evaluation, audit trail, recovery & ownership. Enterprise-grade. |
The Six Layers (abbreviated)
- Workflow design — explicit handoff map between human and agent steps
- Data access — row/field-level scoping enforced in code, not prompt
- Authority — READ / WRITE / SPEND tiers declared and enforced at tool-call time
- Evaluation — company-specific rubric with drift tracking
- Audit trail — tamper-evident log with
reconstruct(run_id) API
- Recovery & ownership — compensating actions for every write; named owner for ongoing tuning
Workflow
Phase 1 — Accept input
Accept a free-form product description, pitch paragraph, feature list, or architecture summary. No structured input required. If the description is fewer than two sentences, ask one clarifying question: "What does the AI actually do when a user triggers it?"
Phase 2 — Score the six layers
Read the description and assign each layer a score:
- Present — layer is explicitly described and structurally implemented
- Implied — layer is mentioned or implied but not built into the product
- Absent — no evidence of this layer
Phase 3 — Apply tier
| Present count |
Tier |
| 0–1 |
Decoration |
| 2–3 |
Wrapper |
| 4 |
Half-built |
| 5 |
Workflow Owner |
| 6 |
Full Stack |
Adjust down one tier if Authority or Audit Trail is absent — these are binary enterprise requirements; their absence is more disqualifying than other gaps.
Phase 4 — Output
## Workflow Tier Verdict
**Tier: [TIER NAME]**
[One sentence plain-English summary of what that means for this specific product]
### Layer Assessment
| Layer | Status | Evidence |
|-------|--------|---------|
| Workflow design | Present/Implied/Absent | [quoted or inferred from description] |
| Data access | ... | ... |
| Authority | ... | ... |
| Evaluation | ... | ... |
| Audit trail | ... | ... |
| Recovery & ownership | ... | ... |
### What's Missing
[Bullet list of absent/implied layers with one sentence each on why they matter for this product's use case]
### Next Build
**Highest-leverage gap**: [layer name]
**What to build**: [one concrete artifact — e.g., "authority-tier config in agent.config.json", "structured audit log written on every tool call"]
**Effort**: [Weekend / Multi-sprint]
### Self-Assessment URL
Run `/impl-arch-audit` for the full scored audit with buyer-question analysis.
Rules
- Be honest about tier assignment. If a product description says "the AI respects user permissions" without explaining how, that's Implied, not Present.
- Authority and Audit Trail gaps are weighted heavier — call this out explicitly if either is absent.
- Never assign Full Stack without explicit evidence for all six layers. "Enterprise-ready" in a marketing description is not evidence.
- Keep the output skimmable. The tier name and the "Next Build" section are what most users will act on.
- If the description is a competitor product, apply the same rubric — do not soften the verdict.
Source
Nate's Newsletter, 2026-05-14 — "The Enterprise AI Deployment Layer: Why Model Access Isn't Enough"
Concept: The strategic question every AI builder should be able to answer — are you decorating a model or owning a workflow?
1---2name: workflow-tier-classifier3description: Paste a product description or pitch and get back a five-tier verdict (Decoration → Full Stack) with the missing implementation layers called out. Lightweight lead-magnet classifier and self-assessment tool. Use on "workflow-tier-classifier", "am I decorating a model", "classify my AI product", "what tier is my deployment".4---56# Workflow Tier Classifier78Paste a product description or pitch deck excerpt and get back a plain-English verdict: are you **decorating a model** or **owning a workflow**? The classifier names your tier, identifies the missing layers, and tells you exactly what to build next.910## When to trigger1112- User says "workflow-tier-classifier", "classify my AI product", "am I decorating a model", "what tier is my deployment", "workflow tier"13- Quick self-assessment before a pitch or proposal14- Reviewing a competitor product or prospect's existing AI system15- As a public-facing lead-magnet entry point before a full `impl-arch-audit`1617## The Five Tiers1819| Tier | What it means |20|------|--------------|21| **Decoration** | Model access bolted onto a product. No deployment layer. The AI can be removed and the product mostly still works. |22| **Wrapper** | Some structure around the model — a prompt, an API call, maybe a UI. But no workflow handoffs, data scoping, or authority limits. |23| **Half-built** | Meaningful implementation (workflow design + data access at minimum), but missing authority enforcement, evaluation, audit trail, or recovery. Would fail a serious enterprise procurement review. |24| **Workflow Owner** | Five of six layers present and structurally enforced. One gap remaining — typically evaluation rubric or recovery procedure. |25| **Full Stack** | All six layers owned: workflow design, data access, authority enforcement, evaluation, audit trail, recovery & ownership. Enterprise-grade. |2627## The Six Layers (abbreviated)28291. **Workflow design** — explicit handoff map between human and agent steps302. **Data access** — row/field-level scoping enforced in code, not prompt313. **Authority** — READ / WRITE / SPEND tiers declared and enforced at tool-call time324. **Evaluation** — company-specific rubric with drift tracking335. **Audit trail** — tamper-evident log with `reconstruct(run_id)` API346. **Recovery & ownership** — compensating actions for every write; named owner for ongoing tuning3536## Workflow3738### Phase 1 — Accept input3940Accept a free-form product description, pitch paragraph, feature list, or architecture summary. No structured input required. If the description is fewer than two sentences, ask one clarifying question: "What does the AI actually do when a user triggers it?"4142### Phase 2 — Score the six layers4344Read the description and assign each layer a score:45- **Present** — layer is explicitly described and structurally implemented46- **Implied** — layer is mentioned or implied but not built into the product47- **Absent** — no evidence of this layer4849### Phase 3 — Apply tier5051| Present count | Tier |52|--------------|------|53| 0–1 | Decoration |54| 2–3 | Wrapper |55| 4 | Half-built |56| 5 | Workflow Owner |57| 6 | Full Stack |5859Adjust down one tier if Authority or Audit Trail is absent — these are binary enterprise requirements; their absence is more disqualifying than other gaps.6061### Phase 4 — Output6263```64## Workflow Tier Verdict6566**Tier: [TIER NAME]**6768[One sentence plain-English summary of what that means for this specific product]6970### Layer Assessment7172| Layer | Status | Evidence |73|-------|--------|---------|74| Workflow design | Present/Implied/Absent | [quoted or inferred from description] |75| Data access | ... | ... |76| Authority | ... | ... |77| Evaluation | ... | ... |78| Audit trail | ... | ... |79| Recovery & ownership | ... | ... |8081### What's Missing8283[Bullet list of absent/implied layers with one sentence each on why they matter for this product's use case]8485### Next Build8687**Highest-leverage gap**: [layer name]88**What to build**: [one concrete artifact — e.g., "authority-tier config in agent.config.json", "structured audit log written on every tool call"]89**Effort**: [Weekend / Multi-sprint]9091### Self-Assessment URL92Run `/impl-arch-audit` for the full scored audit with buyer-question analysis.93```9495## Rules9697- Be honest about tier assignment. If a product description says "the AI respects user permissions" without explaining how, that's Implied, not Present.98- Authority and Audit Trail gaps are weighted heavier — call this out explicitly if either is absent.99- Never assign Full Stack without explicit evidence for all six layers. "Enterprise-ready" in a marketing description is not evidence.100- Keep the output skimmable. The tier name and the "Next Build" section are what most users will act on.101- If the description is a competitor product, apply the same rubric — do not soften the verdict.102103## Source104105Nate's Newsletter, 2026-05-14 — "The Enterprise AI Deployment Layer: Why Model Access Isn't Enough"106Concept: The strategic question every AI builder should be able to answer — are you decorating a model or owning a workflow?
m2ai-portfolio