Home Assistant AI Automation, Tesla FSD and Energy Monitoring with Phil Hawthorne – HGG691
Phil Hawthorne from the Home Assistant Podcast joins me for Home Gadget Geeks 691 to talk about where Home Assistant is headed, how AI is changing the way we build automations, Tesla Full Self-Driving, energy monitoring and some clever ways to make smart-home technology disappear into the house instead of taking it over.
Along the way, we discover that my own Home Assistant installation had fallen several releases behind. That leads to a useful discussion about Docker updates, breaking changes and what happens when a system keeps working well enough that you stop checking whether it is actually current. I fixed it after the show, and I’ll walk through that below.
We also get into a question I’ve been wrestling with across Jimbot, OpenClaw and Hermes: when should AI do the work, and when should AI simply write deterministic code that does the work from then on?
Chapters / Timestamps
[0:00] — Cold Open: Tesla FSD
[0:18] — HGG691 Introduction
[2:50] — Home Assistant Podcast
[4:13] — Grandfathered In
[5:06] — Home Assistant Revival
[5:50] — AI Makes YAML Usable
[8:30] — Official Integrations Arrive
[10:36] — Monthly Releases Return
[12:51] — Themed Maps Incoming
[15:17] — Docker Update Trouble
[21:22] — Dashboards Need Work
[23:27] — Automation Gets Messy
[31:31] — When To Use AI
[32:33] — Video Cutting Workflow
[37:42] — Community Day Announced
[40:47] — Local Meetups Worldwide
[43:09] — Tesla and Home Integration
[52:05] — Monitoring Every Watt
[57:40] — FSD Quirks and Updates
[1:04:40] — Smart Zones and Cooling
[1:12:41] — 3D Printed Sensor Planter
[1:21:23] — Home Assistant News Ahead
Guest Context
Phil Hawthorne joins me again from the Home Assistant Podcast. He has been covering Home Assistant for nine years and brings the perspective of someone who has watched the project grow from a developer-heavy YAML-first platform into a much broader smart-home ecosystem backed by the Open Home Foundation and Nabu Casa.
This Week on Home Gadget Geeks
- How AI is making Home Assistant and YAML more approachable
- Why community integrations can push manufacturers toward official Home Assistant support
- What is coming in Home Assistant 2026.10, including new OpenStreetMap-based mapping
- How my Home Assistant Container installation quietly fell several releases behind
- When AI should make decisions and when deterministic code should take over
- Phil’s Tesla Model YL, Home Assistant integration ideas and FSD experience
- Shelly CT clamps, the Home Assistant Energy Dashboard and the limits of optimizing every watt
- Smart climate zones, thermostat automation and The Stock Pot’s Apollo MTR-1 planter
Full Show Notes
The Home Assistant Podcast and Being Grandfathered In
The Home Assistant Podcast has been around for nine years, long enough that it began as the HASS Podcast. Phil explains that the original HASS name went all the way back to the early terminology around Home Assistant. These days, the much easier address is HomeAssistant.fm. The podcast has also been around long enough that Phil describes its use of the Home Assistant name as essentially being “grandfathered in,” predating organizations such as the Open Home Foundation and Nabu Casa.
Home Assistant Feels New Again
I asked Phil whether Home Assistant is seeing something of a resurgence. His answer is essentially that it never really went away, but there is clearly more happening around the project. Home Assistant has become much easier for ordinary users, manufacturers are taking the platform more seriously, and AI has removed another barrier that used to keep people away.
AI Makes YAML Usable
Home Assistant started as a very developer-focused project. Building automations meant learning YAML and getting comfortable working much closer to the configuration itself. Over the years, Home Assistant built increasingly capable graphical tools so users didn’t have to live in YAML.
Now AI has created an interesting full circle. As Phil puts it, AI can simply go back to YAML. The difference is that the human doesn’t have to be good at YAML anymore. An AI model can help identify devices, explain entities, write an automation, troubleshoot errors and make configuration that once felt intimidating much more approachable.
When Community Integrations Become Official
We also talk about what happens when community integrations become popular enough that manufacturers start paying attention. I had recently received a message through the community EcoFlow integration explaining that EcoFlow had contacted its developer about the way the integration was communicating with its products and was working toward a better API path.
Phil says these situations can go two ways: a manufacturer can see unexpected traffic and lock things down, or it can recognize that Home Assistant users are customers and find a way to support them. Phil points to LG as another example. He bought LG products partly because community Home Assistant support existed, and he says LG now has a developer working on official Home Assistant integration.
Home Assistant 2026.10 and New Maps
Phil previews the upcoming Home Assistant 2026.10 release and the normal monthly release cycle. One of the biggest topics is mapping. Users of older versions may have noticed an API-key warning appearing over their Home Assistant maps after changes involving the previous mapping provider.
Phil explains that Home Assistant has been working with OpenStreetMap on a replacement and redesign. He describes vector maps, improved zooming, redesigned zones and support for map themes. The Home Assistant community already takes dashboard customization pretty seriously, so giving everyone control over the look of the map should take that to another level.
My Home Assistant Was Months Behind
All that talk about new releases led straight into troubleshooting my own system. I checked my version during the show: 2026.5. Phil was previewing the upcoming 2026.10 release, and my installation was several releases behind.
The reason became clear pretty quickly. I’m running Home Assistant Container — Home Assistant Core running in Docker — rather than Home Assistant OS. Because the container itself wasn’t being actively refreshed, I had fallen behind even though the system continued working normally.
Phil points out the real cost of letting this go too long. Instead of dealing with one month’s changes, you can suddenly be jumping across several months of potentially breaking changes.
What I Changed After the Show
I went back and fixed this after recording. My Home Assistant Container deployment was moved into a cleaner Compose-based workflow and upgraded from 2026.5.0 to 2026.9.4, with backup and rollback protection in place.
I’m intentionally keeping upgrades manual. What I didn’t want was another situation where Home Assistant quietly falls months behind without me knowing. So I added a Home Assistant update check to the Jimbot status workflow.
- Running version
- Current stable version
- Whether an update is available
At the time I completed the work, Jimbot reported Running: 2026.9.4, Current stable: 2026.9.4, and Update available: NO. That closes the loop on something we discovered live during this episode. The larger lesson applies well beyond Home Assistant: a system that keeps working hasn’t necessarily stayed current.
Dashboards and Automation Traces
I asked Phil what he’d still like to see improve in Home Assistant. Automation tracing is high on his list. Home Assistant holds a lot of information about what happened, what triggered an automation and what came next. The challenge is presenting that information so users can quickly understand why something happened.
Dashboards are another area where both of us have had mixed experiences. I’ve used AI to create Home Assistant dashboards with wildly different results. Sometimes a simpler model produces something minimalist and useful. A more capable model can occasionally decide to solve several problems I never asked it to solve. You get an impressive dashboard. It just may not be the dashboard you wanted.
When AI Tries to Be Too Helpful
Phil shares a great example from his own Home Assistant setup. He had existing automations controlling blinds. One blind needed to account for a nearby sliding door while another didn’t. He asked an AI agent to add calendar-based behavior around some construction work.
Instead of carefully modifying the existing logic, the agent started rewriting things. Before long, the wrong blind was reacting to the sliding door and overlapping calendar events had created increasingly complicated logic. Eventually Phil had to stop the agent and fix the automation himself.
That gets at an important lesson with AI agents: the ability to make a change is not proof that the change is correct. Human supervision still matters, particularly when an agent has permission to alter a working system.
When Should You Use AI?
Phil makes a distinction that fits closely with what I’ve been learning while building my own AI workflows. If a process should produce the same output every time from the same input, AI may be most useful for writing the code that performs the process. Once that code works, let the deterministic code run. Don’t keep asking an LLM to reinvent the answer.
But some jobs have changing inputs and require interpretation. A podcast transcript is a great example. Every episode is different, and finding the important moments, summarizing the conversation or identifying useful clips requires judgment. That’s where AI makes sense.
Use AI where judgment adds value. Use deterministic code where consistency matters.
Building My AI Video Workflow
The current workflow uses ChatGPT to help identify promising segments, Hermes to work locally with the media and FFmpeg, Google Drive to move artifacts between systems, and Hermes again on the YouTube management side.
Could I manually make some of these videos faster today? Absolutely. But that’s not entirely the point. I’m learning which pieces AI does well, which pieces need deterministic tooling, what needs human review and where the handoffs break. Phil mentions services such as Riverside and other dedicated AI editing tools that already automate some of this. Those may be faster. For me, building it is part of the hobby.
Open Home Foundation Community Day
Open Home Foundation Community Day is coming on Saturday, November 7, 2026. Rather than trying to create one giant event that works across every time zone, Community Day is built around local, in-person gatherings.
Phil is hosting the Melbourne event, and he mentions events being organized in Brisbane and Adelaide, with Sydney also potentially joining in. People can find an existing event or use the Open Home Foundation‘s materials to organize one locally.
Phil Bought a Tesla Model YL
Phil recently picked up his first EV, a Tesla Model YL six-seater, and one of the things that attracted him was its potential integration with Home Assistant. He hasn’t automated it yet because he’s learning how his household uses the car first. Once he knows the patterns, he’ll know what is worth automating.
Climate control is high on the list. On a normal school day, Home Assistant could precondition the car before the family leaves. On a school holiday, it could skip the routine. For an unexpected trip, Phil likes the idea of a physical button near the door that simply says, in effect, “We’re leaving soon — get the car ready.” Eventually battery level, electricity prices and solar production could all become part of those decisions.
Monitoring Every Watt
Phil is working on better home-energy monitoring using Shelly hardware with current-transformer clamps that can measure individual electrical circuits. His Home Assistant Energy Dashboard still has a chunk of consumption that is effectively unknown, and he wants to know where it went. The oven and air conditioner are two of the loads he wants to isolate.
Once those circuits are visible, the questions become much more useful: How much power is the air conditioner really using? Would changing its temperature by a degree make a meaningful difference? Could that usage move to a cheaper time of day? What happens once solar enters the picture? Home Assistant gives him a way to turn energy monitoring into something actionable rather than simply another graph.
When Energy Monitoring Goes Too Far
After installing solar and watching my energy consumption more closely, I started shutting things down. The gaming PC didn’t need to run. The storage server didn’t need to run. Eventually I had dramatically reduced the power usage in my studio — and then I looked around and realized I owned all this computer equipment and most of it was turned off.
Phil went through something similar with his Synology NAS. He tried shutting it down overnight after realizing how much power it consumed over 24 hours. That saved power, but it also interfered with things that were supposed to happen overnight — drive health checks, downloads and other background work. Measurement is valuable. Optimization is valuable. But the goal shouldn’t be to optimize the usefulness out of everything you own.
Tesla FSD: Impressive, But Not Ready to Go Unattended
Later we return to the Tesla, and this is where the cold open for HGG691 comes from. I ask Phil how much he currently trusts Tesla Full Self-Driving. He has been impressed. On a country road, he watched the Tesla perform an overtaking maneuver that he personally might have been hesitant to make.
But would he send the car somewhere unattended? Not yet. “It’s close,” he says, but there are still situations where intervention matters. That includes one particularly memorable example: Phil says the Tesla drove through a red light. He adds that in hindsight he probably should have intervened sooner because he was still responsible for the vehicle.
Phil also describes lane-positioning behavior with his Model YL in Australia where the vehicle rides closer to the center of the road than he would prefer. He says it isn’t actually crossing the center line, but it gets close enough to the road markers that he notices it. It is impressive technology that still reminds you why the driver needs to stay involved.
Smart Zones Beat Thermostat Wars
Then we get into perhaps the most dangerous home-automation problem of all: changing someone else’s thermostat setting. Phil’s advice is simple: tread carefully. He’s had automations override manually selected temperatures, resulting in the memorable complaint, “Stop gaslighting me. I want it at that temperature.”
The better solution is often smarter zoning. Phil uses zones so bedrooms and living areas can be heated or cooled differently depending on occupancy and time of day. That got me thinking about the efficient Wi-Fi-connected window air conditioner we’ve been using around our kitchen and dining area. Rather than trying to make one thermostat solve every comfort problem in the house, targeted cooling combined with occupancy and presence can be a much better answer.
A 3D-Printed Presence Sensor Planter
Phil finishes with one of the more creative smart-home projects I’ve seen recently. Dylan from The Stock Pot has created a 3D-printed planter designed around the Apollo MTR-1 presence sensor.
Instead of another obvious sensor hanging on a wall, the technology becomes part of something that looks like normal home décor. The planter also includes an LED strip that can become another notification surface inside the house: dishwasher finished, laundry waiting, trash night or another Home Assistant status alert. Phil explains that the MTR-1’s millimeter-wave sensing can operate through the plastic housing, allowing the sensor itself to remain hidden.
What’s Next for Home Assistant?
Phil expects the final months of 2026 to stay busy. Community Day is one of the major events on the calendar, monthly Home Assistant releases continue, and he hints that there may be more coming from Nabu Casa before the end of the year. No spoilers from Phil — just enough of a hint to make us watch what happens next.
FAQ
Why didn’t Home Assistant tell me my Docker install was out of date?
Home Assistant Container does not manage the underlying Docker image for you the way Home Assistant OS manages its own update path. In my case, the container kept working, so I did not notice that the image itself had fallen several releases behind.
When should AI be used for Home Assistant automation?
Phil’s rule of thumb is useful: use AI where the input changes and judgment is valuable, such as summarizing transcripts or helping design an automation. When the same input should always produce the same output, use AI to help write deterministic code and then let the code do the repeatable work.
Can Tesla data be used in Home Assistant automations?
Yes. Phil is interested in using vehicle state, battery level and location as Home Assistant inputs for climate preconditioning, charging decisions and household routines. He is intentionally learning the family’s real usage patterns before automating them.
What are CT clamps used for in home energy monitoring?
Current-transformer clamps measure electrical current on individual circuits without putting a smart plug between the appliance and the outlet. Phil plans to use them on large dedicated circuits such as the oven and air conditioner so Home Assistant can account for more of his home’s total energy use.
What is the 3D-printed presence sensor planter?
It is a planter design from The Stock Pot built around the Apollo MTR-1 mmWave presence sensor. The sensor is concealed inside the planter, while an LED strip can provide Home Assistant notifications for things such as laundry, dishes or trash night.
