The Sense of Things

I spend my working life with machines that have failed. Electronic control modules arrive on my bench dead or misbehaving, and my job is to determine what happened inside them and whether it can be undone. It is unusually good training for thinking about the future, because failure analysis forces you into a specific discipline: you can only reason from what the object retained. Whatever the module did not record, you do not get to know.

Almost nothing records anything. A module that has been in a truck for nine years, through nine winters and roughly a hundred thousand thermal cycles, arrives carrying no account of any of it. The information existed. It was simply never captured, and so a decade of physical history is gone, and I am left inferring a story from solder joints.

This is the ordinary condition of the built world, and I have come to believe it is the largest addressable inefficiency we have. Not a shortage of intelligence. A shortage of memory in the places where things actually happen.

In an earlier essay I argued for a particular architecture to fix this, and for what I called machina ex machina: infrastructure that lets machines produce the conditions for the next generation of machines. That essay was about how. This one is about what for. I want to describe, as concretely as I can, what human life looks like on the other side of a world that can describe itself — and I want to do it without the vocabulary that usually accompanies these predictions, because that vocabulary has been used to sell so much that it no longer carries meaning.

A companion essay to this one takes up the risks, which are serious and which I do not want to minimize by omission. But I think it is worth stating the positive case first, in detail and without embarrassment. People do not build difficult infrastructure because they have been warned. They build it because someone described something worth having.

The premise

I want to be precise about the level of capability I am assuming, because vagueness here is how these essays go wrong. When I refer to pervasive instrumentation below, I mean a physical environment with the following properties, all of which are achievable with engineering rather than breakthroughs:

• Measurement nodes cheap enough to deploy by the hundred in a facility, and by the thousand across a utility network, without a capital approval process. Call it the price of a decent smoke detector, falling.

• Operating life measured in years to a decade on a primary cell, or indefinitely on ambient harvested energy, with no scheduled human intervention.

• Substantial local memory: enough for a node to retain a high-resolution account of one physical thing over its entire service life, rather than transmitting instants and forgetting them.

• A local timeline that remains coherent without a network, and can be reconciled onto a shared timeline afterward, with honest uncertainty attached.

• Self-description in a machine-readable, cryptographically verifiable form: what this device is, what it measures, in what units, with what accuracy, attached to what, attested by whom.

• Composability. A measurement can be separated from the device that makes it, so a new physical quantity is a new probe rather than a new product program, and the long tail of measurements gets built by people close to each problem.

• Mesh-native operation with no dependence on a vendor cloud, so the system is useful when disconnected and does not die when a company is acquired.

• Deployment and recovery by machine — aerial or robotic — so that instrumenting a place is an errand rather than a construction project.

None of these individually is speculative. Every one exists today in some product, somewhere. What does not exist is all of them at once, in an open form, cheap. That combination is the thing I think is worth building, and the rest of this essay is about what it would mean.

Two limiting factors apply everywhere below, and I will not repeat them each time. The first is that measurement does not by itself cause action; someone or something still has to respond, and organizations are slow. The second is that physical deployment takes time — pipes, buildings, and grids turn over on decade cycles. I think the benefits I describe are large enough to survive both frictions, but they will arrive later than the technology allows.

I. The losing game

Water

Start with water, because it is the clearest case and because the numbers are almost comic.

A water utility treats water at real energetic cost, pressurizes it at further cost, and then loses a large fraction of it into the ground before it reaches anyone. Commonly cited figures for developed-world systems put the loss somewhere around a sixth of production, and considerably worse in older networks. The leaks are not exotic. They are joints and hairline fractures in buried pipe, and they announce themselves continuously through acoustic signature and pressure behavior to anyone in a position to listen.

Nobody is listening, because listening currently requires a technician with a correlator walking a line, and there are more miles of line than there will ever be technicians. So leaks are found when the street collapses, or during a survey that reaches any given segment once every several years.

Now populate that network with acoustic and pressure nodes at a density that only makes sense when a node costs very little and lasts a decade. Each one keeps its own high-rate history. Leak signatures emerge from correlation across neighboring nodes rather than from a single threshold, which means the system finds the small leak that has been running for eight months rather than the catastrophic one that already made the news.

The result is not a dashboard. It is a sixth of a city's treated water not disappearing, along with the energy used to treat and pump it, and a maintenance program that repairs pipe on a schedule rather than in an emergency at two in the morning with a road closure. In water-stressed regions this stops being an efficiency story and becomes a question of whether a city has to build another reservoir.

The reason this has not happened is almost entirely the cost and awkwardness of instrumentation. It is one of the largest, most tractable, least glamorous wins available to us.

It is worth pausing on why cost matters so much more than capability here, because it is the pattern underneath every section of this essay. The technology to find water leaks acoustically has existed for fifty years. What has never existed is a version cheap enough to leave in the ground permanently, everywhere, without anyone deciding it was worth it. Instrumentation has threshold effects: below a certain cost per point, you stop asking whether a measurement is justified and simply measure, and the questions you can ask change category. Sparse measurement answers questions you already had. Dense measurement reveals structure you did not know was there.

II. The grid we cannot see

Energy

The transmission side of the electrical grid is reasonably well observed. The distribution edge — the feeders, the pole-top transformers, the service drops — is close to dark. Utilities frequently learn a transformer has failed because customers call.

This was tolerable when the edge was a passive delivery system with predictable, well-characterized loads. It is becoming untenable now, because every heat pump, every EV charger, and every rooftop solar array applies stress at exactly the place nobody is watching, in patterns nobody modeled. Electrification is a plan to dramatically change the behavior of a system we do not currently measure.

Dense instrumentation at the edge changes what is possible in three ways. It makes failure predictive rather than reactive, since transformers announce their decline thermally and harmonically for months. It makes capacity real rather than assumed, so that the answer to whether a neighborhood can support forty more chargers comes from observation instead of a conservative model written in 1998 — which usually means more headroom exists than anyone will authorize using. And it makes local coordination possible, since a system that knows the actual state of a feeder can shift flexible loads by seconds and defer physical upgrades that would otherwise cost millions.

There is a further effect that I think is underappreciated. Distribution-level visibility is what makes local energy markets possible at all. A neighborhood where solar generation, storage, and flexible load can be coordinated against the real-time state of the feeder is a materially different economic object than one where all of that is invisible and therefore has to be managed by conservative rules. The infrastructure for that coordination is not the software; the software is the easy part. It is the measurement, and it does not exist.

The honest limiting factor here is regulatory rather than technical. Utilities earn returns on capital deployment, and a technology whose main effect is to defer capital deployment has an incentive problem, not an engineering one. But the pressure from electrification is going to force the issue, because the alternative is discovering the limits of the distribution system the hard way, repeatedly, in public.

III. The field is not uniform

Food and land

A field is not one thing. Soil moisture, compaction, and nutrient availability vary substantially across a single hectare, and the variation is stable enough to be worth knowing. Irrigation is nonetheless applied more or less uniformly, because measuring the variation at useful resolution costs more than the water saved.

Invert that cost and the practice changes. You water the parts that need water. In arid regions this is the difference between land that keeps producing and land that does not, and the water saved is water that stays in an aquifer being drawn down faster than it recharges.

The same logic runs through the rest of the food chain. Cold chain integrity is currently validated by a logger someone remembers to check, which means spoilage is discovered at the destination rather than prevented in transit. Grain storage failures announce themselves through temperature and moisture gradients days before the loss becomes total. Livestock health shows in movement and thermal signature well before it is visible.

What unifies these is that the food system loses an enormous fraction of what it produces to problems that were observable and unobserved. That is a rare category: waste that requires no behavioral change to recover, only attention.

IV. The building that stops arguing

The places we spend our lives

Most people reading this spend the overwhelming majority of their lives inside buildings that are performing worse than anyone realizes, in ways nobody can prove.

The archetypal case is a comfort complaint. Someone says the west side of the third floor is cold. Facilities checks the setpoint, finds it correct, and the complaint is logged as a preference. It recurs. Eventually someone discovers a damper actuator failed months ago and the system has been fighting itself since, burning energy to produce a result nobody wanted.

Instrumented at the resolution of the actual complaint — per zone, continuously, with each zone keeping its own history — the question becomes answerable in an afternoon, and mostly answers itself. The building tells you when it began, which is usually the most diagnostic fact available.

Extend this to air quality and the stakes rise. We learned during the pandemic that ventilation in most occupied spaces is unknown rather than adequate, and the evidence connecting carbon dioxide concentration to cognitive performance in classrooms and offices is strong enough to be uncomfortable, given that we do not measure it. Instrumentation here is not an optimization. It is finding out that a meaningful fraction of the population spends its working hours in conditions that measurably degrade their thinking, and then fixing it, which is usually a matter of running a fan differently.

There is also aging in place, which I want to handle carefully because it is where this technology most easily curdles into surveillance. The version I find defensible does not watch a person. It watches a house: whether the kitchen is being used, whether the temperature is being maintained, whether the water has run today. Those are properties of a building, they are enormously informative about whether an older adult is managing, and they can be arranged so that the resident controls who sees what. The distinction between instrumenting a place and monitoring a person is a design decision, it is easy to get wrong, and it is the subject of the companion essay.

V. The end of the three a.m. call

Work, and the people who maintain things

Discussions of automation usually skip the people who keep physical systems running, so I want to spend a section on them, because pervasive instrumentation changes their working life more than almost anyone's.

Maintenance work today is structured around ignorance. You perform preventive service on a calendar because you cannot see condition. You get called out at night because failures announce themselves at the last possible moment. You spend a large share of your time diagnosing — which is to say, reconstructing history that was never recorded — before you can begin the repair. And when you arrive, you are frequently working from documentation that describes the equipment as designed rather than as it now is after fifteen years of modification.

A well-instrumented environment inverts most of that. Condition is visible, so service happens when it is needed. Failures are anticipated, so work moves to daylight hours. Most importantly, the diagnostic phase collapses, because the equipment retained its own history and can simply be asked.

I want to be careful not to oversell this as a purely liberating change. Detailed instrumentation of physical work can just as easily become a productivity surveillance apparatus pointed at technicians, and there are firms that will build exactly that. But the underlying shift, if the data is oriented toward the equipment rather than the worker, moves skilled trades away from reactive scrambling and toward genuine engineering judgment. It also preserves knowledge that currently walks out the door when the person who has been in the building for twenty years retires, taking with them the fact that the third pump has always run hot and everyone knows to watch it.

VI. Everything is somewhere

Transport, logistics, and the things in motion

Global logistics is an enormous system that operates with startlingly little knowledge of its own contents. A container crossing an ocean is, from the perspective of everyone with an interest in it, a sealed box with a manifest and an estimated arrival date. What actually happened inside it — the temperature excursions, the shock events, the humidity, the four days it spent on a hot dock in Jebel Ali — is discovered at the destination if it is discovered at all.

The consequence is that the system is built around insurance rather than prevention. Cargo is damaged, a claim is filed, the loss is priced in, and the cycle repeats, because determining who caused the damage costs more than the cargo. Pharmaceutical shipments are discarded on suspicion of an excursion nobody can confirm or rule out. Produce is sold at a discount because its remaining shelf life is unknown and therefore assumed to be short.

Instrument the cargo rather than the vehicle — the crate, not the truck — with a device that keeps its own verified record and costs less than the loss it prevents, and a great deal of this resolves. Shelf life becomes a calculated quantity rather than a conservative guess, which means less food thrown away while it is still good. Damage becomes attributable, which changes behavior at the point where the damage is actually caused. And the shipment can report a problem in transit, while intervention is still possible, rather than at the destination when only the claim remains.

The same reasoning applies to the vehicles. A truck, a locomotive, a ship, and an aircraft are all machines whose failures are expensive, dangerous, and preceded by observable change. Aviation figured this out decades ago and has the safety record to show for it, but aviation could afford instrumentation that nothing else could. What changes now is that the cost of knowing drops below the cost of the cheapest asset worth knowing about, and the aviation approach becomes available to the ordinary world of trucks, buses, and municipal equipment.

The limiting factor here is commercial rather than technical, and it is a good illustration of a general pattern. In logistics the party who could deploy the instrumentation is frequently not the party who bears the loss, and the party who bears the loss has no way to compel the other. Verifiable identity helps more than it might appear: a record that a shipper cannot alter and a carrier cannot dispute makes it possible to write contracts against physical reality, which is what actually shifts the incentive.

VII. The building as a clinical instrument

Health care operations

I want to approach health carefully, because it is the domain where enthusiasm most often outruns evidence, and where the difference between instrumenting an environment and monitoring a person matters most.

The unglamorous version is the one I believe in. Hospitals lose extraordinary amounts of money and clinical time to problems that are purely operational: infusion pumps that cannot be located, sterilization loads whose parameters were not recorded, refrigerated pharmaceuticals whose excursions are unverified, operating rooms sitting idle because scheduling has no visibility into actual turnover time. These are inventory and environmental problems, they are solved by instrumentation rather than by medicine, and the beneficiary is a nurse who spends less of a shift hunting for equipment.

Environmental monitoring in clinical spaces is a second case with a clearer payoff than most people realize. Ventilation and filtration performance in patient areas is frequently assumed rather than measured, pressure differentials in isolation rooms drift, and the periodic validation that is supposed to catch this happens quarterly against a phenomenon that changes hourly. Continuous verification is the kind of thing that sounds like a compliance exercise and turns out to be an infection control intervention.

Then there is the home, which is where the ethical care is required. A great many older adults would prefer to remain in their houses, and the thing that most often ends that arrangement is not a medical event but the absence of any way for a family to know that things are still going well. The version I find defensible instruments the house rather than the person: whether the kitchen was used today, whether the heat is holding, whether water has run, whether the door opened. Those are properties of a building. They are highly informative about whether someone is managing. And crucially, they can be arranged so that the person living there controls what is shared and with whom, which converts the system from something done to them into something they operate.

The failure mode is obvious and I want to name it rather than gesture at it. The same hardware, pointed slightly differently and with the consent model removed, is an apparatus for monitoring a person continuously in their own home on behalf of somebody else. Whether a given deployment lands on one side of that line or the other is a design and governance decision, not a technical one, and the companion essay to this one is largely about how easily it goes the wrong way.

VIII. Instruments as the rate limiter

Science

Here is a claim I would defend strongly: across large areas of science, the binding constraint is not theory or computation. It is that we cannot afford enough instruments.

Ecology, hydrology, seismology, atmospheric chemistry, and oceanography are all fields where the phenomena are spatially heterogeneous and the sampling is sparse, because each measurement station is a capital project. We build models of continuous systems from data collected at a handful of points and then argue about the interpolation. Climate science in particular operates with heroic ingenuity on a network that would embarrass any laboratory scientist if it were proposed as an experimental design.

Drop the cost per station by two orders of magnitude and make deployment a flight rather than an expedition, and the sampling density changes by enough to alter what questions can be asked. Not better answers to existing questions. Different questions, of the kind that only become askable when you can see structure that was previously below your resolution.

There is a specific case I would point anyone at who doubts this. Our understanding of groundwater — a resource that a very large fraction of humanity depends on and that is being depleted faster than it recharges — rests on monitoring wells that are sparse, unevenly distributed, and in many regions read manually a few times a year. We are managing one of the planet's most important resources with a sampling regime that would not pass review in an undergraduate lab course, not because anyone is careless but because each well is expensive and there is no budget for a thousand of them. This is what a measurement cost constraint looks like when it collides with something that matters.

There is a second-order effect I find even more interesting. Science is currently limited by the reproducibility of instrumentation as much as by anything. When every measurement carries its own verifiable provenance — which sensor, which calibration, which conditions, on what timeline — a whole class of irreproducibility simply disappears. Not because researchers become more careful, but because the metadata that everyone knows they should record is recorded automatically, by the instrument, as a property of the measurement rather than a task for a graduate student.

IX. The first seventy-two hours

Disaster

Everything above assumes normal operation. The case that moves me most is the abnormal one.

In the first seventy-two hours after an earthquake, a flood, a wildfire, or a chemical release, the dominant problem is not resources. It is that nobody knows what is happening. Decisions that determine how many people die are made from fragmentary reports, in an environment where the existing infrastructure is often the thing that failed.

Instrumentation that can be deployed by air in minutes, that self-organizes, that keeps working when the network is gone, and that carries verifiable identity so that data from three agencies who have never coordinated can be trusted and combined, is a different capability than anything currently available to incident command. A wildfire perimeter covered in a moving lattice of temperature, wind, and particulate nodes. A structure covered in strain and tilt sensors before anyone is asked to walk inside. A flood plain seeded ahead of the crest.

The requirement that makes this work is the unglamorous one from the premise: self-description and independent timelines. A hundred anonymous sensors dropped from a drone, each with an unsynchronized clock and no record of what it is, is not situational awareness. It is a cleanup task. Everything depends on the boring layer.

X. The end of disposability

Materials, repair, and what we throw away

This is the domain I know best, and the one where I think the effect is most underestimated.

We discard an extraordinary quantity of functional or repairable hardware because determining whether it can be repaired costs more than replacement. That calculation is driven almost entirely by missing information. A returned electronic module carries no record of what it experienced, so diagnosis starts from zero, and skilled diagnostic labor is expensive enough to exceed the value of the device.

Now suppose the module retained its own operating history: thermal cycles, supply excursions, fault events with timestamps, the conditions under which it stopped working. Diagnosis stops being an investigation and becomes a lookup. The economics of repair invert for an enormous category of goods, and remanufacturing — which is dramatically better than recycling, because it preserves the embodied energy and labor in the object rather than melting it down — becomes viable at scale.

The same information changes design. Manufacturers currently learn about field failures through warranty claims, which are heavily filtered, delayed by years, and stripped of context. Population-scale field history would close a feedback loop that has never really been closed, and products would get better in the specific ways reality demands rather than the ways a test lab predicts.

I find this compelling partly because it requires no sacrifice. It is not asking anyone to consume less. It is asking that objects be able to explain themselves, so that the effort already invested in them is not thrown away for lack of a record.

XI. The parts that assemble

Toward systems that build themselves

Everything so far has been an application of instrumentation. This section is about what happens when instrumentation becomes a construction material.

Consider what a node in the premise actually is: a compute element with memory, a radio, a verified identity, a coherent local timeline, and a physical interface that accepts modules describing themselves. Deployed by the thousand, those are not just sensors. They are a substrate with the properties you would need to assemble larger systems out of smaller ones.

The first step is aggregation, and it is already sensible today. A group of nodes observing the same physical process can share their observations directly with each other rather than each reporting upward. What emerges is a description of a system rather than a collection of readings — the behavior of a pump loop, an entire feeder, a whole structure — computed at the edge by the things doing the observing. The unit of analysis stops being the sensor and becomes the system, without anyone assembling that view centrally.

The second step is specification. A dense measurement network observing its own environment produces exactly the data required to specify what should be built next: which quantity is missing, where coverage is thin, which failure was caught too late and what additional channel would have caught it earlier. That analysis is not exotic; it is straightforward given the data, and impossible without it. The network describes the gap in its own coverage in a form precise enough to act on.

The third step is where it becomes properly interesting. Once placement is done by machine, and once probes describe themselves well enough to be selected automatically, and once the specification for what is missing is generated by the network, the loop closes: a sensing system that identifies its own blind spots, specifies the instrument that resolves them, and dispatches something to place it. A person still authorizes each step, and should. But the analysis, the specification, the provenance, and the physical act are all performed by the system.

There is a fourth step that I think is the genuinely interesting one, and it depends entirely on the memory in the premise rather than on any of the mechanical capability.

A network of nodes that each retain years of local history is, collectively, a distributed record of how a physical environment has actually behaved. Not a model of it — a record, held in the places the behavior happened, by the devices that observed it. That distinction matters because a record of this kind has a property that centralized models lack: it degrades gracefully and it is hard to lose. There is no single copy, no single vendor, no single database whose corruption erases the history of a building. The environment's memory is distributed through the environment, the way structural knowledge in an organization is distributed through its people.

Once that record exists, systems can be built out of it rather than merely reporting into it. A new node placed in an instrumented facility can be told by its neighbors what normal looks like here, and can begin operating with priors instead of starting from zero — which shortens commissioning from months to hours and is roughly what a new employee gets from their colleagues on the first day. A replacement device inherits the history of the one it replaced, so continuity survives maintenance rather than being broken by it. And a system that has been observing a process for years can specify, with real evidence, the tolerances a replacement component must meet, because it knows what the process actually does rather than what the original specification assumed.

That last capability is the seed of something I find genuinely striking. An environment with a long, verified, distributed record of its own behavior contains, in a machine-readable form, most of the specification for its own next revision. Not the creative part. Not the decision about what the facility is for. But the enormous, tedious, error-prone middle of engineering — what the actual loads are, what the actual duty cycles are, where the actual margins were wrong, which assumptions the physical world declined to honor — is exactly what such a record contains and exactly what engineers currently reconstruct by hand, badly, from incomplete evidence, on every project.

That is what I mean by machines being used to build themselves, and I want to be clear about how modest the claim is. There is no artificial general intelligence in that loop. There is a network that knows what it is measuring, knows what it is not measuring, and can act on the difference. Every element is engineering. What makes it feel like something new is that the composition is a system with a genuine, if narrow, capacity for self-extension.

The reason I keep returning to this idea is that it resolves a bottleneck that otherwise looks intractable. The single largest obstacle to instrumenting the world is not any technical property of instruments. It is that somebody has to decide where each one goes, specify it, procure it, install it, commission it, and maintain it, and there are not enough somebodies. Every step of that chain is human labor that scales linearly with the number of measurement points, which is why measurement density has stayed roughly flat for decades while the cost of the sensing element collapsed. A system that can specify and place its own extensions is the only structure I can see that breaks that linearity.

Extend the same logic to robots and the argument I have made elsewhere follows directly. A humanoid robot arriving in an unfamiliar building has only the present instant, observed from its own vantage. A building that has been keeping its own records can tell it what it needs to know. The environment stops being an obstacle to be perceived and becomes a collaborator that has already done most of the work. I think this is a sequencing claim worth taking seriously: large-scale robotics is gated less on robot capability than on whether the world can explain itself, and the second problem is being worked on by far fewer people.

XII. What a world that pays attention is like

The texture of it

I want to end with something harder to quantify, which is how any of this actually feels to live with.

There is a version of this future that I want to argue against first, because it is the one usually sold. It involves screens everywhere, notifications about your building, an app for your water heater, and a general expectation that people will spend attention interacting with their instrumented environment. That vision is both unappealing and wrong. The correct outcome of pervasive instrumentation is less attention spent, not more. A system that requires you to look at it has failed at the thing it was for.

The honest answer is that it mostly feels like less friction and fewer surprises. Your water bill does not spike because a slab leak ran for four months. Your power stays on through a heat wave because the utility knew which transformers were marginal. The road is not closed for emergency repair. Your flight is not delayed by an unexpected mechanical. The bridge you drive over is being watched continuously rather than photographed every two years. Food costs less because less of it spoiled between the field and the shelf. Your parents stay in their house longer, on their own terms, because their house can quietly indicate that things are fine.

None of that is dramatic. Good infrastructure never is. The nineteenth-century public health revolution — clean water, sewers, food inspection — did more for human life expectancy than almost any medical intervention, and it is invisible to the people it saves. Nobody feels the cholera they did not get.

What I find genuinely exciting is a different thing, which is the epistemics. We currently make enormous decisions about the physical world — where to build, what to replace, how to allocate water, which failures to tolerate — on the basis of sparse sampling and confident modeling. A world with pervasive instrumentation is one where a large class of arguments becomes settleable by evidence. Not everything, and not the values questions, which are the important ones anyway. But the factual substrate underneath those arguments would be dramatically firmer than it is now, and firmer ground makes for better disagreements.

That is the future I am working toward, and I want to state plainly that I do not think it is guaranteed, or even that it is more likely than not. The same properties that make this technology valuable make it dangerous in specific ways that deserve their own essay, and I have written one. Ubiquitous sensing is ubiquitous observation. An identity layer that lets a probe prove itself can just as easily let a vendor own facts about your property. A world that pays attention is a world that can be made to pay attention to you.

But I think it would be a mistake to let that stop us from describing what is worth building. Almost nothing here requires a scientific breakthrough. It requires a great deal of careful engineering across radio, power, cryptography, mechanical design, and data — done by people willing to work on unfashionable problems whose payoff is diffuse and slow.

I also want to be honest that the timeline is long and the work is unglamorous in a specific way that deters people. There is no demo. You cannot show a venture partner a compelling forty-second video of a leak that did not happen. The payoff arrives as an absence — of failures, of waste, of emergencies — distributed across a great many people who will never know which of the bad days they did not have. The people who built municipal water treatment are not famous, and the reason is precisely that they succeeded.

That is a fair description of everyone who has ever built infrastructure. It is not glamorous work and it does not produce a demo. It produces a world that keeps its promises slightly better than the one before, and then another one after that.

I would like to help build it, and I would like company.

Connecting the digital to the physical.

Contact

Links

wadesthomas1@gmail.com

+12155019211

© 2026. All rights reserved.