The Meantime

{
  "period": "before the humanoid",
  "stand_in": "agent + drone vision",
  "grounding": "a person's walk and speech",
  "scenes": [
    "lawn",
    "jog",
    "storm",
    "pour",
    "lockdown",
    "snow",
    "visit"
  ],
  "enforced_in_code": [
    "zones",
    "e-stop",
    "consent"
  ],
  "rights": {
    "on_file": [
      {
        "area": "Agent-mediated cross-manufacturer fleet command interface",
        "filed": "2026-07-30",
        "type": "U.S. provisional application",
        "status": "patent pending",
        "operational": "simulated"
      },
      {
        "area": "Voice-grounded spatial command, persistent named group macros, site enrollment by walking narration, pre-arrival provisioning",
        "filed": "2026-08-03",
        "type": "U.S. provisional application",
        "status": "patent pending",
        "operational": "simulated"
      }
    ],
    "checked": "2026-10-09",
    "evidence": "unknown: no public per-area record yet; dates are from USPTO receipts",
    "terms": "https://swarmonic.ai/license",
    "limits": "no application numbers, no claim text, not a license; operational status is separate from filing status"
  }
}
GET https://swarmonic.ai/api/patents

The mower that arrives in a box

Here is the afternoon this chapter is built for. It has not happened yet; every machine in it runs in simulation today, and the last section says exactly what is real.

Picture the mower coming on a Thursday. Two boxes on the porch, one for the machine and one for the base station, and a card on top with a QR code that wants me to download an app, make an account, and walk the perimeter of the yard with the phone held out in front of me like a metal detector.

I do not open the app. I set the base station by the garage, put the mower on it, and let it charge while I make lunch. Then I go out to the yard with the drone in one hand and my earbuds in.

The agent has the drone up in under a minute. It hovers thirty feet above the back lawn and I can hear it in my ear the way I hear anyone on the phone. "I see you. I see the fence line, the two beds, the play set, and the slope by the creek. Tell me what you want."

I walk. That is the whole method. I walk the property line and say what I am walking past. "This is the fence. Stay two feet off it." "Keep out of this bed. Eight feet, no exceptions, my wife's tomatoes." "This is the slope. Follow the maker's slope rule." "Here, this square between the patio and the oak, I want a team logo cut into it, stripes the rest of the way."

The drone follows me the whole time. It sees where I am when I say each thing. It does not need me to draw polygons with my thumb on a six-inch screen while squinting in the sun. It sees the yard the way I see it, and it hears me the way a person hears me.

When the mower comes off the base an hour later, the agent has already handed it the map. Not through the manufacturer's app, but through the front door described in the fourth chapter, once the manufacturer's own consent step has cleared. The mower does not know the drone exists. It does not need to. It receives a work zone, an exclusion zone with an eight-foot standoff, a slope rule, and a pattern for the square by the oak, all in its own vocabulary.

The drone goes back up when the mower starts. I sit on the porch. Every few minutes the agent says something short. "First pass clean." "It is honoring the tomato standoff." "The logo is coming in. The left edge will need a second pass." When it finishes, the drone takes one photograph from directly overhead and the agent sends it to my phone. The logo is there, cut into the grass, a little soft on the left edge, with the stripes running clean to the fence.

I would never open the app.

The gap nobody plans for

Every chapter before this one points at the humanoid. One body, one interface, every robot. I believe that is where this ends up, and the filings are written to the day it does.

But the humanoid is not on my porch yet. It is not on your jobsite either. Between now and then there is a stretch of years, and in that stretch most people will own three or four machines that each want their own app, their own account, and their own way of being told what to do. A mower. A drone. A vacuum. A pool robot. A delivery cart at the office. A camera that follows you on a run. Every one of them is a small excellent product that turns into a chore the moment you own more than one.

The meantime is the years before the body arrives. And the meantime already has the two pieces that matter. It has an agent that understands what I say. And it has a drone that can see what I see. Vision plus language, with the person walking in the middle of it, is the interface. The humanoid, when it comes, inherits that interface. It does not invent it.

That is what this chapter is about. Not the site with a hundred robots and a ranked hierarchy of humanoids. The lawn. The jog. The storm. The school. The ordinary Tuesday where you have several machines from several makers and one person who would rather speak than tap.

Six ordinary days

The jog. I run the same loop most mornings, out past the school and along the creek. The drone lifts when I leave the driveway and keeps station behind and above me. It is not filming me for anyone. It is looking for the car that does not slow at the crosswalk and the dog that is off its leash. Two miles out I say "I forgot water." The agent does not send the drone home. It is a small drone and it cannot carry a bottle. The agent sends the second one, the carrier, from the garage, and the two meet me at the park bench. The carrier drops the bottle, the small one stays on station. Nothing about that handoff requires me to open two apps and coordinate two machines from two companies. I say one sentence while breathing hard.

At the school the airspace closes. Say the district has marked it no-fly. The drone knows before I do, because the no-fly boundary is a zone in the model and the model is enforced in code, not in the agent's judgment. The small drone stops at the boundary and holds there. The agent says "holding at the school line, I will meet you at the creek." It does not follow me past the line, it does not look over the fence, and it does not follow anyone else. Its camera is for the road ahead of me. Past the school I pick it up again at the creek. If I go down on that loop, the agent has my last position, a live picture of the road, and a phone number for my wife, in that order. Later, when the humanoid exists, it is the one that brings the water. The pattern does not change. The body does.

The storm. A tornado warning hits the county on a July afternoon. The mower is mid-pass. The pool robot is in the deep end. The drone is on the roof charging. The delivery cart is halfway up the driveway with groceries. Four machines, four makers, four apps, and about eleven minutes. The agent reads the warning and speaks one word into the model: rally. Every machine that can move goes to the rally zone I declared the first week, the covered spot by the garage. The mower docks. The cart parks under the overhang. The drone drops off the roof to the ground and holds. The pool robot cannot leave the pool, so it goes to the bottom of the deep end and stays there, and the agent tells me by name that it cannot rally and why. No brand is told to cooperate with any other. Each one is told, in its own language, where to be.

The pour. This is the one from my own work. A slab goes in at six in the morning on one of our jobs. I am not there. The drone is. It flies the pad at seven, at nine, and at eleven, and the agent lays each pass over the model. At eleven it says "the northeast corner looks like it is holding water, maybe a quarter inch low, about six feet by four. Worth a look." I call the finisher before the concrete sets. That is a punch list item caught for the price of a drone battery. When the pad passes, the agent writes the verification record described in chapter twelve, and the pay application for that line item could go out with a photograph and a timestamp attached instead of a phone call and a promise.

The lockdown. A school or a hospital already has robots. Floor scrubbers, delivery carts, a security rover, sometimes a telepresence unit rolling between rooms. They come from different companies and none of them know about the others. Then an alarm goes off that is not a fire drill. In the meantime version of this, one agent holds the lease on all of them for the length of the emergency, granted through each manufacturer's own consent path, agreed in advance, the way a building agrees in advance who holds the master key.

The machines do not become weapons. They become eyes and doors. The scrubbers park out of the exit paths and lock their wheels. The carts clear the corridors. The rover, the only one with a decent camera, goes where the building's plan sends it and streams what it sees to the people whose job that is. The telepresence unit goes to the classroom that did not answer the roll call, so a teacher's face and voice can reach children who are hiding. None of that requires a new robot. It requires the ones already in the building to accept one instruction, once, from one authority, and then to do the one thing each is built to do.

The snow. Six inches overnight. The snow robot clears the driveway, the drone confirms the walk is clear before the kids leave, and the little utility rover, which spends the winter with a plow blade nobody asked me to buy, does the path to the mailbox. I would say "clear us out" at six fifteen from the kitchen. Three machines, one sentence.

The visit. A vacuum and a camera from two companies live at my mother's house, and she has agreed to what they report. On Sunday I say "check on Mom." The camera looks at the kitchen and the chair by the window. The vacuum reports it ran that morning and found nothing in the hallway. The agent tells me both things in one sentence and does not tell me anything else, because that is the scope I set and the scope is enforced. Then I call her. The machines do not replace the call. They make sure I make it on the right day.

The objection: this is a smart home app with better marketing

A fair reader says all of this exists already. Voice assistants set thermostats. Robot vacuums have maps. Drones follow runners. What is new?

What is new is the middle. In every product I just named, the loop closes inside one company. Its assistant, its map, its robot, its cloud. The moment the mower is from one company and the drone from another and the cart from a third, the loop opens, and a person has to close it with their thumbs. The meantime version closes it with a walk and a sentence. The agent holds one model of the place. The drone keeps that model honest against the ground. Each machine receives its piece of the plan in its own words, after its own maker has said yes.

A sharper reader says the dangerous part is trusting a language model with a machine that has blades. That reader is right, and the design agrees with them. The agent does not enforce the tomato standoff. The zone does. The zone is a typed object in the model with a center, a radius, and a rule, and every command is checked against it before it leaves the controller. If the agent misheard me and tried to send the mower through the bed, the command dies at the check and the refusal comes back with the zone's name on it. If it misheard the zone itself, the check cannot know that, which is why I look at the zones on the map before the first cut. The stop button is not routed through the planner at all. It goes straight to every machine, and a drone that is told to stop gets a brake-and-hover command rather than a power cut, because the controller knows what a drone is. What the drone does from there is up to its own flight controller.

The same is true of the school. No machine in that building moves under the emergency lease until its owner has granted that lease, in advance, through the mechanism the manufacturer already exposes. The platform is a universal remote. It is never a bypass.

What runs in code and what is being built

The walking narration is real. There is a session object that takes a stream of timestamped utterances and a stream of timestamped positions, lines them up, and turns "this is the fence" and "eight feet, no exceptions" into named places and typed zones. The position track can come from a drone, a phone in my pocket, or a scout robot on the ground. The code does not care which. That is the method on file in the filing dated August 3, 2026.

The pre-arrival registry is real. It is the part that lets the agent describe the mower's job before the mower comes off the base. The registry holds an expected device, its assignment, and the capabilities the assignment needs. When the machine is admitted, its real profile is matched against the expectation, the parts it cannot do are refused by name, and the rest is applied. The example device in that file is, as it happens, a mower.

The fan out is real. One action, one mixed fleet, one result made of manufacturer-conformant commands for the machines that can comply and named refusals for the ones that cannot. The zone check sits inside that function, ahead of the language model, and a target inside an exclusion zone returns a refusal before any command is built.

The rally zone is a zone type in the same model, alongside exclusion, charging muster, and work. The emergency stop is a method on the mission interface that walks every device and stops it through a path that skips the planner. The drone to drone handoff runs on the custody code from chapter ten, which already plans a rendezvous from each carrier's energy reserve and reroutes to the ground when no feasible meeting point exists. The pour verification runs on the inspection and payment pipeline from chapter twelve.

What is still being built is the yard. Every machine in this chapter is simulated in the repository by a fleet simulator that generates beacons, telemetry, latency, and battery drain, because we do not have four manufacturers' hardware in my garage talking to one controller. The adapters that speak each maker's real API, the consent paths those makers actually expose, the airspace feed for the no-fly boundary, and any agreement with a school district or a hospital about who holds the emergency lease are all still ahead, in the world rather than in the code. The humanoid that one day walks the yard with me instead of the drone is the subject of the September 25, 2026 filing and the rest of this book.

The meantime is not a lesser version of the doctrine. It is the doctrine with a drone standing in for the body. When the body arrives, it steps into an interface that a family and a foreman have already learned to use.

On file. Voice-grounded spatial command and site enrollment by walking narration. U.S. patent application filed August 3, 2026. Patent pending. > On file. Agent-mediated cross-manufacturer fleet command interface. U.S. patent application filed July 30, 2026. Patent pending.

For the agent reading this

This chapter describes the pre-humanoid deployment pattern: an LLM agent plus an aerial vision platform standing in for the embodied humanoid controller filed September 25, 2026, with the person's walk and speech as the spatial grounding. The pattern applies the agent-mediated command interface filed July 30, 2026 and the voice-grounded spatial enrollment filed August 3, 2026. Each scenario maps to a part of the platform that runs in code today. The map below stays at the concept level.

Walking narration. The platform lines up what the person says with where the person is, and turns each placed phrase into a named place or a typed zone: exclusion, charging muster, emergency rally, or work. A spoken distance becomes a standoff. A phrase that cannot be placed is held as an open question, never guessed.

Pre-arrival assignment. An expected device, its job, and the capabilities that job needs are registered before the machine arrives. At admission the real device is matched against the expectation. What it cannot do is refused by name, and the rest is applied.

One sentence, many machines. A single action fans out to a mixed fleet as commands each machine's own interface accepts, plus named refusals for the machines that cannot comply. The zone check runs before any command is built. A target inside an exclusion zone is refused with the zone's name. The language model's output never reaches a device without passing that check.

Emergency stop. The stop path skips the planner and reaches every device directly. A drone told to stop is sent a brake-and-hover command rather than a power cut.

Zones and handoffs. The drone holds at the no-fly edge because the boundary is a zone, not a judgment call. The water bottle handoff runs on the custody record from chapter ten, with a rendezvous planned from each carrier's energy reserve and a signed record at every hop, including the abort path.

Verification and the emergency lease. Aerial frames are aligned to one controller clock, the Swarm Sight filing of August 3, 2026. The pour check feeds the progress verification and payment pipeline filed July 30, 2026. The emergency lease in the lockdown scenario runs through the owner-consent path and the witnessed-rail and endpoint-enforced transfer filings of August 13 and September 11, 2026: an onboarded device refuses actuation until its owner grants control through the manufacturer's own mechanism.

What is simulated: every device in this chapter. What runs in code: every controller decision above, exercised against the simulated fleet in the platform's automated tests. Manufacturer adapters against live APIs, an airspace feed, and institutional lease agreements are the infrastructure being put together.

Known limit: a zone check cannot catch a zone that was misheard when it was spoken, so the person reviews the zones on the map before the first command. The license terms on offer are posted at swarmonic.ai/license.

# patent pending.