Digital Employees · Catalog
1000+ agent templates, generated and tested — not hand-built
The Role Factory composes agent templates from versioned packs, tests every variant, and grows the catalog through quality gates from 20 to 1000+.
The formula
Agent Template = Role + Industry Pack + Skill Packs + Language Pack + Tool Permissions
+ Knowledge Scope + KPI Pack + Eval Pack + Persona Policy
Example: SDR role + Real Estate pack + WhatsApp outreach and calendar booking skills + Tamil pack + CRM, WhatsApp and Calendar permissions + project brochures and price list scope + site-visit-booked KPI + objection-handling evals + synthetic avatar with AI disclosure = Real Estate Tamil Site-Visit Booking Agent.
A variant is generated and tested, not hand-built. Each pack is versioned independently, and the template pins pack versions.
Taxonomy
| Level | Examples | Owner |
|---|---|---|
| Family | Revenue, Marketing, CX, Finance/Admin, HR/Talent, Legal/Compliance (assistive only), Operations/Procurement, Creative/Media, Engineering/App-builder, Industry-specific | One3 product |
| Role | SDR, Review Responder, Collections Assistant, SEO Analyst… seeded from the 140-role catalog | Family lead |
| Variant | Role × Industry × Language × Channel | Role Factory (generated) |
| Tenant instance | A rented variant configured with a client's data, memory and policies | Client (data) + One3 (template) |
Pack types
| Pack | Contains | Versioning |
|---|---|---|
| Skill Pack | Actions used, prompts and skill graphs, tool-use examples | semver; breaking change = major |
| Industry Pack | Objects, vocabulary, playbooks, compliance rules, KPIs | semver |
| Language Pack | Language and code-mixing prompts, tone, Indic evals, speech choices | semver |
| KPI Pack | Outcome definitions, attribution defaults, guardrail defaults | semver |
| Eval Pack | Golden tasks, red-team cases, simulator scenarios, pass thresholds | semver; must not weaken without review |
| Persona Policy | Avatar class, disclosure text per channel and jurisdiction, voice rules | semver; legal review on change |
Certification levels
| Level | Meaning | Requirements |
|---|---|---|
| Listed | Available for sandbox trial | Template spec complete, Eval Pack passes minimum, persona policy set, owner named |
| Verified | Available for production rental | Eval Coverage Score ≥ threshold, enough design-partner shadow decisions with agreement ≥ threshold, zero critical incidents, ledger integration tested |
| Certified Outcome | Eligible for performance bonus or outcome pricing | Verified + holdout-validated lift in several tenants of the target segment + dispute rate below threshold over a defined period |
No invented thresholds
The sample sizes and thresholds for Verified and Certified Outcome are decisions to be set from design-partner data. The Verified badge needs a template-level Eval Coverage Score of at least 85 on its allowed skills, across at least two design-partner tenants, or a "Simulator-verified" label until live data exists.
Catalog growth gates
| Stage | Catalog size *(target)* | Gate to move on |
|---|---|---|
| G1 (Year 1) | 20 Listed, 6 Verified | ≥ 3 revenue squads live; statements for all renting tenants; 0 consent incidents; positive return evidence for ≥ 3 agents in ≥ 3 tenants each |
| G2 (Year 2) | 100 | Generated variants pass evals without manual prompt edits for most variants; support load per template sustainable |
| G3 (Year 3) | 300 | Partner-built templates pass certification; dispute and incident rates within thresholds; replacement engine proven |
| G4 (Years 4–5) | 1000+ | Industry and language packs at scale; Certified Outcome in top families across industries |
Catalog quality rules
Four rules that keep the catalog useful.
- 1
No template without its safety kit
Every published template has an Eval Pack, a persona policy, a named owner and kill criteria.
- 2
Variants earn their level
A variant inherits its parent's level only if its own Eval Pack passes; otherwise it drops to Listed.
- 3
No catalog bloat
Templates with no rentals for a defined period are archived.
- 4
Assistive roles stay assistive
Legal/compliance, HR and credit-related roles have an A2 ceiling. Humans decide.
The first 20 rentable agents
Planned for the agency and SMB beachhead in Year 1. Six are planned to be Verified at launch.
AI SDR — Inbound Lead Responder (Verified)
Speed-to-lead and meetings booked. CRM, WhatsApp, email, calendar. Ceiling A3.
Support Triage & Reply Assistant (Verified)
Auto-resolved without reopen; first response time. Desk, Knowledge, WhatsApp. Ceiling A3.
Content & SEO Assistant (Verified)
Briefs and drafts approved. Publishing stays human. Ceiling A2.
Collections Reminder Assistant (Verified)
DSO reduction. Reminders only; R4 never. Ceiling A3.
Meeting-to-CRM Updater (Verified)
Accurate field updates from meetings. Ceiling A3.
Knowledge Curator (Verified)
Gap closure and answer coverage. Ceiling A2.
Review Response, Social Scheduler, Ads Budget Pacer
Local business growth roles. Ads pacing acts only within caps.
Creative Variant Generator, Website Chat Greeter
Synthetic avatars, always disclosed as AI.
Lead Nurture, Quote Follow-up, Win-back
Nurture and conversion roles with consent and frequency evals.
Appointment Booking, Client Reporting, Cart Recovery
No-show reduction, on-time agency reports, recovered orders (holdout).
Market Research, Invoice & Payment-link, Lead Enrichment
Cited research, same-day invoices, clean CRM data.
Questions about the Role Factory
Are 1000 agents available?
No. 1000+ is a directional target for Years 4–5. Year 1 plans 20 Listed and 6 Verified templates.
Can agencies build their own variants?
Yes. Agency private variants stay agency IP and are not shared with other agencies.
Can partners publish templates?
From Year 2, after certification: eval pass, security review, disclosure compliance and documentation.
What happens to a template that underperforms?
Kill criteria send it to retrain, restrict or retire — for example negative template margin, high supervision ratio, high dispute rate or poor rental retention.
Why can't HR or legal agents act on their own?
Those roles are assistive only, with an A2 ceiling. People decide hiring, legal and credit outcomes.
Suggest a role
Tell us which role your agency or business needs first. Demand data shapes the catalog mix.