Raya: Outbound Customer SMS Platform

Role Product Manager, AI and Sales Experience · Freedom Forever Stack Anthropic Claude SDK · Twilio (SMS + MMS) · Lightspeed ERP / SaaS event triggers Coverage Freedom Forever launches Raya (PR Newswire)

Context

Raya is Freedom Forever's brand for its third-party-facing AI: an amalgamation of agents serving both customers and internal sales reps. I worked on Raya directly in two ways. Day to day I expanded its skill set, and my headline contribution was building the outbound customer-SMS platform described below.

Problem

In discovery I found the real driver. Freedom Forever had let go the 40+ person project support team that gave white-glove service to at-risk customers, the team responsible for a 3% lift in project realization. Meanwhile Raya already gave internal sales reps a deep set of tools, MCP integrations and issue-resolution skills, to handle those same customer problems. The opportunity was clear: deliver that white-glove experience directly to 150,000+ eligible customers through agents that reused the rep-side tooling, preserving the realization lift and improving issue resolution instead of dropping customers back onto blunt, automated messaging.

What I built

An outbound SMS platform on Twilio, sitting on top of the existing agent orchestrator. Per use case you configure a distinct agent profile: its personality, its skills, and the content it is allowed to send. The platform triggers off events in the SaaS and ERP layer, so when a defined condition fires it calls the orchestrator and hands the chosen agent the customer's full prior conversation history. Context carries even if a different agent or channel contacted that customer before. The platform enforces do-not-call compliance and runs SMS campaigns with configurable follow-up triggers.

How I built it

I built this platform end to end with Claude Code, as a non-engineer. The strategy that made a solo build practical was to build into the system that already existed rather than around it: reuse the platform's existing functions and integrations wherever they were already there, and write new agent skills that matched the patterns of the established codebase. Keeping the surface area small enough for one person to own was a deliberate design choice, not an accident.

The hard part

The difficulty was policy, not the model. Contact windows, frequency caps, opt-out, and "never make this feel like harassment" were product decisions dressed up as technical ones. The highest-stakes piece was the cross-agent context handoff: unifying conversation history so a customer never receives a contextless message from a system that should already know them.

Outcomes

The platform was in active development when Freedom Forever entered shutdown in April 2026. Raya itself was announced publicly (link above).

What I learned

The hard problem in customer-facing LLM products is not the model, it is the policy. Every interesting decision (when to reach out, when not to, how much context to carry, how to phrase things) is a product decision. And unifying context across a fleet of agents matters more than any single agent's cleverness.