AI Sales Hub
Multi-product AI sales and marketing agents with human approval on every external action
AI agents research, analyse, update an internal CRM and draft outreach. Nothing leaves the system without human approval and an evidence-based readiness gate.
Status: Started Nov 2026. Sprint 3 delivered a durable agent runner, a metered AI gateway and the first Lead Discovery agent. · Period: Nov 2026 – present · Source: Private repository.
Screenshots are not published yet. The architecture is shown below.
01Overview
An agentic system designed around safety from the start. Agents may act internally, but every external action requires a person to approve it after passing an evidence-based readiness gate. AI usage goes through a metered gateway.
02The problem
Running sales and marketing for several products as a solo builder, without letting autonomous agents send unchecked messages.
03Who it is for
- Product owners running multi-product sales and marketing
04My role
Architecture and implementation.
05Implemented features
- Durable agent runner.
- Metered AI gateway.
- Lead Discovery agent.
- Product-scoped authorisation and an audit log.
06Engineering challenges and approaches
Autonomy with accountability
The domain layer encodes the readiness policy and sales gate as pure rules, so no code path can send an external action without approval.
07Outcomes and current status
- Core platform and the first agent implemented and tested.
No usage or revenue numbers are published here unless I can verify them.
08Roadmap (not yet shipped)
- More specialised agents.
- Analytics and cost reporting.