Trigger events, agentically.

The 2015 filing for Family 4 — On-Line Direct Response Platform, WO/2016/154153 — answers a different question than the earlier families. Families 1 and 2 answered "who" and "how valuable." Family 3 answered "how do we drive this." Family 4 answers "when." Its claim: the situation itself is the trigger. A preselected profile appearing at a specific location is the signal that an opportunity exists right now. You bind three pieces into one atomic event — identity (device ID), location (second specified location), and qualitative score (from Family 2) — and route the bundle to an actor. In 2015 the actor was an advertiser.

The architecture is cleaner than the industry it was commercialized against. Most targeted advertising in 2015 was scheduled or audience-retargeted — it fired when the advertiser decided to buy impressions, not when the real-world situation actually changed. The patent's mechanism is the opposite: the trigger is the world, not the calendar. Event-driven, not auction-scheduled. The exchange becomes a router for situations, not a clock for campaigns.

Eleven years later, the architecture is identical. What changed is the actor at the end of the pipeline. That single substitution reframes which parts of the mechanism matter most.

From advertiser to agent

In 2015, when the trigger fired, the opportunity went to an advertiser — a bidder at an exchange, a creative rendered on a device, a measurement pixel fired at the end. The advertiser was a human system operating in minutes-to-hours time. The damage an advertiser could do with a bad trigger was bounded: a wasted impression, an irritated user, a modest amount of money misallocated.

In 2026, the actor at the end of the pipeline is increasingly an agent — a software system that can send money, unlock a door, enroll a patient in a benefit, dispatch a field worker, authorize a procedure, initiate a refund. The time scale compresses to seconds. The damage a bad trigger can do expands dramatically. The same mechanism that routes "user enters store-adjacent geofence → show ad" now routes "patient leaves clinic grounds before follow-up → care-coordinator agent initiates outreach, possibly a pre-composed text, possibly a pharmacy call."

The patent's structure holds up beautifully. The problem it didn't need to solve in 2015 — because the actor was a human-run advertising system with built-in temporal limits — becomes the central problem in 2026: what does the trigger system do to keep the agent from acting badly, too often, irreversibly, or against the principal's intent?

The three things a 2026 build adds

Policy envelopes.Every trigger must be evaluated against a policy engine that governs what the agent is allowed to do under what circumstances. Policies encode consent (this user opted into proactive outreach), rate limits (no more than three outbound actions per day on this channel), sensitivity (health data permits fewer automations than commerce), and escalation (this class of trigger requires human confirmation before action). The policy envelope is not a feature of the mechanism — it's a first-class architectural layer between the trigger and the actor.

Reversibility guarantees.Actions that can be undone can be taken more readily; actions that can't must be gated more tightly. A refund is reversible; a publication is not. A soft notification is reversible; a settlement is borderline. The trigger-event architecture needs to carry a reversibility class with every action — and the agent infrastructure needs to honor it. In 2015 the ad impression was irreversible but cheap; the bet was that cheapness bounded the damage. In 2026 with expensive consequential actions, reversibility is the new cheapness.

Audit logs as first-class product.Every fired trigger, every policy evaluation, every agent action gets a reviewable record. Not as a compliance afterthought but as a core feature — the system's customers (principals, regulators, the users subject to the triggers) all need to be able to review what happened, why, and what reversal options exist. The 2015 patent described an audit log only implicitly; 2026 makes it as structural as the trigger engine.

What sits on the event bus

The implementation in 2026 looks different from the 2015 sketch. Instead of a dedicated ad exchange, the trigger system runs on a streaming event bus — Kafka, Redpanda, Tinybird, Temporal workflows — that any producer can emit to and any consumer can subscribe to, gated by policy. Instead of a single-actor advertiser, the subscribers are a fleet of agents with varied capabilities: send message, initiate payment, unlock door, enroll patient, dispatch technician. The opportunity-instantiator that the patent describes — the component that bundles profile + event + score into one atomic routable unit — is the key integration point. It's where the policy envelope evaluates, where the reversibility class attaches, and where the audit log captures the full provenance of the action.

There's a kind of poetic justice in this. The exchange-like architecture the 2015 patent was designed against was the "anything that makes the buyer happy with the number of firings per dollar" quantitative ad exchange. The architecture the patent described — profile-plus-event as atomic opportunity, score-gated routing, explicit actor addressing — produces exactly the structure a modern agentic system needs, for the opposite reason. It's not about maximizing quantity. It's about bounding quality: right action, right actor, right consent, right time, reversible if wrong.

Three non-ad applications that benefit

Chronic-care patient outreach.A diabetic patient skips blood-sugar logs for five days. Preselected patient profile + trigger (no-log-for-≥N-days AND location-≠-clinic) → care-coordinator agent dispatches a pre-composed text, possibly schedules a check-in call. Policy envelope: the patient opted in, outreach is capped at two per week, and the care coordinator can reverse the automation if inappropriate. The window to intervene was five days. Without the trigger system, it's five weeks until a human reviews the chart.

Agentic replenishment across contexts.A household's dog food runs out while they're traveling. Pantry IoT reports low; phone location is 500 miles from home. Trigger: replenishment-needed AND user-not-at-home → procurement agent negotiates delivery to the travel address, inside a pre-authorized spending envelope. Reversibility: within one hour of order placement, auto-cancel is available. Audit: every decision the agent made is in the log, in a form the household can review.

Workplace-safety escalation. Lone field workers — linemen, EMTs, delivery drivers — are hardest to protect because conventional alert flows are batch. Worker profile + trigger (location-enters-risk-polygon AND biometric-spike AND no-scheduled-task-here) → dispatcher agent checks in, escalates, or dispatches backup. Policy: the worker can suppress this class of check-in for the day if their situation is false-positive-generating. Reversibility: a soft check-in can be withdrawn; a dispatch of backup cannot.

What the licensing conversation looks like

For any counterparty building an agentic system — healthcare-care-coordination SaaS, commerce-replenishment platform, field-safety dispatcher, benefits-agent infrastructure — the Family 4 architecture is the spine. The claims cover the atomic bundling of identity + situation + score into a routable opportunity. The add-ons (policy envelope, reversibility class, audit log) are the productization; the core patent coverage is the mechanism underneath. The conversation I'm open to is structured as field-of-use licensing: healthcare, commerce, workplace safety, civic response, agentic financial services. Each is big enough to support its own licensee; none competes meaningfully with the others.

The 2015 patent was correct. The actor changed, and the stakes with it. The architecture that was a good bet for ads turns out to be a necessary structure for what comes after.