What is included
Six operations covered in the founding package. Each agent is specified, versioned, and auditable. No black-box SaaS; the prompts, tools, and outputs are visible to the operator.
Inbound triage and draft replies
Classifies incoming tickets or emails by intent, drafts a reply in the operator's voice, and routes escalations. Queued for approval before send.
Lead qualification and enrichment
Scores inbound leads against ICP rules, enriches with firmographic data, writes the first outreach in the rep's voice, drops into the CRM.
Content drafting and review
Produces first drafts for blog, social, newsletter and internal updates from a shared vault. Editor approves before publish.
Order and fulfilment ops
Watches orders, flags anomalies, reconciles against stock and delivery systems, auto-drafts customer notifications when SLAs slip.
Approval queue and audit log
Human in the loop is default. Every agent action is logged with actor, timestamp, input, output. Matches the Loomgraph governance pattern.
Weekly operator review
Thirty minutes, same weekday each week. Review agent outputs, retire what did not work, tune what did. No set-and-forget.
First 90 days
Concrete milestones. Each phase ends with a working output the operator can point at, not a slide deck.
Audit and first agent live
- Inbox and CRM access walkthrough
- Identify the three highest-volume ops
- Deploy one agent end to end
- Approval queue wired to Slack or email
Expand to three agents
- Two more agents from the priority list
- Shared vault for prompts and tools
- Audit log endpoint exposed to the operator
- First weekly review ran with team
Review and compound
- Retire or rewrite agents that missed target
- Measure hours saved and errors caught
- Pilot a fourth agent on a new surface
- Case study draft for cohort rights
Proof this is not greenfield
The patterns below are shipping today on ElektraOS, MealShift and Loomgraph. The SMB package is the same runtime, scoped down.