Two years before ChatGPT, the team at Verizon's all-digital carrier deployed the architecture that became the model for everything Agile Catalyst now ships. Pillar of the joint MuShuHaRi + S+3 Agile thesis.
Verizon's answer to a generational shift in mobile distribution was Visible — a digital-only postpaid carrier with no stores, no call centers, and no commissioned sales force. It was meant to ship at a fraction of the cost-to-serve of an incumbent. By 2020, the math wasn't yet working.
The incumbent telecom cost-to-serve baseline is structural. Stores, agents, retention teams, billing call centers, IVR systems, churn-recovery campaigns — all built around a customer who calls for help. A digital-only sub-brand has none of those line items, and none of the human levers an incumbent uses to recover a difficult interaction.
Visible's early operating reality was that customers still needed help. The first-generation answer was a chat system staffed by humans. That chat ran into the predictable wall: it scaled linearly with subscribers, the agents needed product expertise the LLM-free tooling of 2019 couldn't reliably supply, and the cost per interaction slid toward the very benchmark Visible was built to escape.
Three things had to happen simultaneously, or the brand thesis didn't hold:
What the team built in response — through 2020 — was an operating model for AI-assisted human work that pre-dated the language-model era. It was not, in any meaningful sense, a chatbot. And one specific application of it was concrete enough to patent: a machine-learning system that catches the subscribers who are struggling to port their number or activate their device, and fixes the root cause before the failure becomes a support contact.
ChatGPT would not be released until late 2022. The transformer architecture existed; commercial LLM tooling did not. What this team built had to be assembled from supervised classifiers, retrieval systems, knowledge graphs, and human-AI interface design — not from a single foundation-model call.
The operating model does not replace the human agent. It augments the agent at every step of the interaction, against a continuously-updated knowledge model of the product, the customer, and the resolution playbook.
The architecture had five working layers. Each one is recognizable, in shape, in MuShuHaRi's Operational Discipline stage and in S+3 Agile's Scale Pillar. The order matters — every layer is only as strong as the one beneath it.
No autonomous response. No customer-facing chatbot replacing the human. No black-box decisioning. Every interaction had a named human accountable, and every AI action was reviewable, reversible, and explainable to a Verizon compliance team.
The operating model was the methodology. One specific application of it was novel enough to be granted a US patent: the system that governs onboarding, number porting, and device activation — the highest-volume, highest-failure moment in a digital carrier's customer lifecycle.
Porting a number and activating a device is where digital-only subscribers get stuck, retry, and call for help — and every failed attempt burns device and network resources while generating exactly the support contacts Visible was built to avoid. The patented system scores an onboarding risk value for each subscriber from signals like elapsed time, previous carrier, error codes, and support-contact history; uses that score to flag the subscribers who are about to fail; identifies the root cause; and automatically remediates — correcting data-entry errors, reconfiguring network devices, opening tickets, or reaching the customer with targeted guidance — before the failure turns into a cost.
Risk assessment and resolution for end device porting and activation. Filed October 2021, granted August 2024. First-named inventor Himanshu Niranjani, on a six-inventor Verizon team; assigned to Verizon Patent and Licensing Inc. The broader human-AI operating model is the methodology this invention sits within and informed — it is not itself the patent.
The cost-to-serve curve bent first. The engineering-productivity curve followed within two quarters. Both metrics held — they were not a launch-window spike — and both became the structural advantage that allowed Visible to keep its price point against a market that was contracting around it.
The cost-to-serve number was the metric the brand was measured on, but it was not the most consequential outcome. The engineering team stayed small — a precondition to the unit-economics thesis ever holding. A digital sub-brand that needs the team size of an incumbent has lost the argument before it starts.
The architecture compounded. Every interaction improved the system: the retrieval surface got sharper, the knowledge graph got denser, the human agent got faster. This is the operating definition of a learning system, and it distinguishes a real AI capability from an AI demo.
And the regulatory posture held. Because the human was in the loop and the AI was reviewable, compliance review at every layer of Verizon's parent organization could be passed. The architecture did not require regulators to accept a black box. AI that survives an audit — the difference between something shipped and something shelved.
When Agile Catalyst formalized the two frameworks years later, the Visible architecture was the reference implementation. Every principle in the books has a working precedent in the Visible build.
The five-layer model is what DouJou ships as AI infrastructure. The cadence under it is whatS+3 Agile licenses as a delivery framework. The maturity progression is what MuShuHaRi formalizes as a maturity framework. The investor lens is howOneX.Ventures evaluates portfolio companies. Visible was where it was first proven; the federation is how it now scales.