Communication Is the Substrate, Not a Feature

The most consequential sentence in most loan files was typed into a chat app the system will never see.

Every operations leader recognises the pattern. The system of record holds the outcome: approved, rejected, escalated. The reasoning that produced the outcome lives somewhere else entirely, in a WhatsApp thread, an email chain, a phone call summarised from memory, a hallway conversation nobody wrote down. The work is on the system. The meaning of the work is scattered across channels the system cannot reach.

For human-run processes this was always costly. For processes where AI agents act between humans, it becomes disqualifying, and the reason is worth walking through carefully.

A Chain Is Only as Coherent as Its Context

Take the worked example from Anand's essay on this subject: a refund orchestration. An agent analyses a customer complaint. It triggers a logistics agent to validate the customer's claims. A human is notified with a summary and approves the refund. The customer confirms satisfaction. The CRM updates. A control algorithm reviews the transaction for fraud. An auditor reviews the overall system's performance.

Seven tasks, three kinds of actor: AI agents, a human decision-maker, and reviewers. The chain produces a good outcome only if each action stays consistent with the customer's intent and the organisation's intent at every step. And that consistency has exactly one carrier: the communication that surrounds each task.

So long as the communication stays within task context, each actor, human or machine, decides with the full basis for the decision in front of it. The moment those communications fragment into WhatsApp, email, or any off-context channel, the basis for the action taken at each point disappears. The logistics agent validates a claim without the nuance the complaint carried. The human approves from a summary missing the exchange that qualified it. The auditor reviews a sequence whose connective tissue was never captured. Every actor behaves correctly on the context it has, and the chain still fails, because the context each actor has is partial in a different way.

The Architectural Claim

This produces a conclusion that sounds mild and changes everything downstream: the communication architecture determines whether interactions happen within task context, and therefore whether each action stays consistent with the original intent. Communication belongs in the orchestration substrate itself. Treating it as a feature layered on top, a chat panel here, a comment box there, reproduces the fragmentation with better styling.

The demand this places on a platform is specific. Every interaction must be bound to a task, an actor, and an intent. A question about a case lives on the case. A clarification travels with the work to whoever, or whatever, holds it next. A decision's rationale is recorded where the decision was made, readable by the reviewer, the auditor, and the next agent in the chain.

What This Shaped in the Platform

This argument, made well before it was fashionable, is the design rationale behind the way Ontoz structures its workspaces: the surfaces where business users work, where support teams resolve queries, and where oversight monitors the whole, all built so that two-way communication happens inside task context rather than beside it.

It also explains a design choice in the Lending Labs LOS that otherwise looks like a channel gimmick: WhatsApp document collection is wired into the workflow, so a borrower's uploads and confirmations land on the application itself. The channel is the customer's choice. The context binding is the architecture's.

The test for any orchestration platform, before the AI features and after the demo: when work changes hands, does the meaning travel with it? A platform that cannot answer yes is a place where every handoff quietly starts over.