Pillar · AI in Luxury AV & Automation
Where AI is real. Where it is vaporware. And how we specify it.
By Mike Restrepo, Owner · Restrepo Innovations
A client asked me last month whether their new home would have AI. I told them it already does. and that most of what gets sold as AI in this industry has nothing to do with what is actually working in their walls. That gap is the problem this pillar exists to address.
For most of the last decade, “AI in the home” meant a glowing puck on a kitchen counter that could set a timer most of the time. The promise was always bigger than the experience. That changed in the last 18 months. Voice models, vision models, and on-device inference matured enough to close the gap between the demo and the daily use. The systems we install today actually do what people thought smart homes were supposed to do back in 2016.
But the hype machinery did not slow down when the technology caught up. If anything, it accelerated. So this pillar is specific. Below is what AI does well in 2026, what it does not, and the specification rule we use on every Restrepo project.
What AI Actually Does Well in 2026
One. AI cameras. Camera systems used to do one thing: record video and bury you in motion alerts. Within a week, most clients muted the app. which defeated the point. The cameras we specify today distinguish people from vehicles, recognize specific familiar faces and ignore them, flag unknown persons in real time, identify license plates of regular visitors, and detect packages being dropped or removed. The signal-to-noise ratio finally tips in your favor. We run all of this on a segmented Ubiquiti VLAN with no direct path to the public internet. Privacy is a design decision, not a disclaimer.
Two. Adaptive lighting. The system monitors exterior light conditions and adjusts interior fixture levels to maintain a target lux reading at a given surface. That is not marketing language. it is a closed-loop photometric control system. A lighting system that responds to the actual conditions in the room is fundamentally different from one that follows a clock. We program this through Crestron’s native intelligence layer paired with Lutron dimming hardware. The AI label gets applied loosely here, but the function is real and it shows up in every room every day.
Three. Predictive maintenance and energy logic. The system watches the rack. amplifier temperatures, processor uptime, network port errors, drive health on the NVR. When a metric trends the wrong way, we get a service ticket before you experience an outage. Most clients only notice this when something does not break on a holiday weekend. Modern automation engines also layer in occupancy patterns, calendar context, weather, and solar production. The thermostat adjusts thirty minutes before you usually arrive home, not the moment your phone crosses a geofence.
What It Does Not Do. Named Products, Specific Failures
One. Habit-learning algorithms. Every platform has promised a home that learns your behavior. The reality: human routines are variable in ways that short-term observation cannot reliably capture. A system that learns you dim the living room at 8pm will start doing it on the nights you have dinner guests and want full brightness. The correction mechanism becomes a battle. Machine learning applied to home automation is most useful as a data analysis tool for the integrator. not as a live decision-making engine inside the home.
Two. Josh.ai in daily use. We installed it. We supported it. We fielded the calls when it did not work correctly. Josh.ai’s cloud NLP layer introduces latency between 500 milliseconds and 1.2 seconds under normal conditions. up to 4 seconds under load. Firmware updates pushed regressions into previously stable installations. The callback rate from Josh.ai installations was higher than any other component category in our residential work. It is a more thoughtfully designed product than consumer voice alternatives. But protecting our clients’ investment means recommending what works reliably every day, not what performs best in a controlled demo. We no longer specify it as a standard recommendation.
Three. AI as an autonomous behavioral agent inside the home. The idea that a home should modify its own behavior based on observed patterns is where the design philosophy breaks down. Autonomous modification of home behavior based on pattern inference is a source of unpredictability in environments where predictability is what clients are paying for. A home automation program built from a genuine understanding of how the client lives will outperform any adaptive algorithm, every time. Understanding the client is not a machine’s job.
Our Specification Rule
The framework is not “AI versus no AI.” It is this:
AI as an analytical and sensing tool, yes. AI as an autonomous decision-maker inside a client’s home, no.
That means: AI cameras and analytics on a private VLAN. Adaptive lighting built on Lutron hardware and Crestron programming logic. Predictive maintenance monitoring the rack and flagging anomalies for human review. Voice control that triggers the Crestron system. not one that bypasses it. Local-first processing wherever the technology supports it, with cloud only for large language models that cannot run on-premise.
Voice should trigger your system, not control it directly. The moment you expect a consumer voice platform to manage a multi-zone automation sequence by itself, you have exceeded its design intent. And for voice that matters. the keypad does not mishear you. It does not depend on your accent, the background noise, or whether a cloud service is responding. Press the button, the scene runs. Every time.
Where This Leaves Us
The version of the smart home I was selling in 2016 finally exists in 2026. The trick now is staying disciplined about what we deploy and what we hold off on. Not every demo is ready for a real client’s real life.
We specify AI where it earns its place. in cameras, in energy intelligence, in maintenance prediction, in meeting room transcription, in adaptive lighting built on real hardware. We do not specify it where it creates unpredictability or service exposure. That is the rule. It is not complicated. It is just honest.
AI by Pillar
Every Restrepo article on AI in luxury AV, organized by where it lives in your project.
AI in Control Systems. View pillar →
AI in Lighting. View pillar →
AI in Networking. View pillar →
AI in Conference Rooms. View pillar →
AI in Surveillance & Access Control. View pillar →
- Ubiquiti UniFi Access & Protect in a Luxury Estate
- Bergen County Estate Home Automation (Gate & Intercom)
More on AI surveillance and access. license plate recognition, familiar-face flagging, AI forensic search, biometric access. publishing through 2026 as part of the Restrepo content velocity plan.
AI in Home Theater. View pillar →
More AI-and-cinema content. auto-framing, object-based audio processing, predictive rack maintenance, 4K/8K AI upscaling. coming through 2026.
Specifying AI in a luxury residence or commercial project? We engineer what is real and decline what is not.
Start Your Project