Are Home Assistant projects shifting from connecting things to extending Home Assistant itself? Data + a clicky-clicky explainer inside

, ,
1

I found a small signal worth pressure-testing:

9 of 33 Q1 2026 project launches in my sample primarily extend Home Assistant itself, compared with 1 of 20 in Q1 2025.

So: Are Home Assistant projects shifting from connecting things to extending Home Assistant itself?

This is far too small a sample to call a trend. But it made me curious, and I turned the sample into an interactive explainer so the data is easier to inspect:

Open the clickable explainer

I initially looked for evidence of an AI-driven wave of new projects. The forum posts could not tell me who used AI, so that hypothesis did not survive the data. What did show up was a possible change in what people are building: not only integrations for devices and services, but tools for operating, configuring, and extending Home Assistant itself.

I am sharing this early because this community knows the projects and context better than anyone. If I classified something incorrectly, missed context, or drew too much from the sample, please tell me.

The explainer includes the data, classifications, method, and source links. Or clone all of it from GitHub.

What I would especially like challenged:

  • Would you classify any of the 53 projects differently?
  • Does the distinction between connecting something and extending Home Assistant make sense?
  • What context does the forum data miss?
  • Am I reading too much into the difference between these two Q1 samples?

Have fun!

2

Nice explainer, but I’m not clear what point you’re making. Is it a bad thing? Does it matter?

My impression is that posts in the forum are moving quite quickly in the direction of connecting obscure devices (as opposed to making HA work) - but that may be a different issue.

What are you selling?

3

Define a Home Assistant Project.

Completely missing from your analysis are the multitude of DIY projects using ESPHome. In years past if a sensor or control didn’t exist “over the counter”, we made it ourselves. Today the appliance operators have taken over and the DIY crowd has all but vanished.

You also observed the difficulty in determining if a project build was aided by {he who must not be named}. This could be because mentioning {he who must not be named} was forbidden by the moderators, even threatening to expel forum members who violate the edict. The warning is still in the forum guidelines.

1 Like
4

There is no guideline against using AI to build things for HA as long as you mention it. There is a guideline against helping people using AI. People come here to be helped by people and AI often gives outdated or even plain wrong advice. If people want help from AI, they are free to ask an AI themselves, there’s no need for others doing it for them.

Plus, if you ever want AI to learn, they must learn from wht people wrote here, not from whatever a previous generation AI thought. Because that would make them dumber, not smarter. So hope the forum does not get flooded with AI even more then it already is.

Personally I significantly reduced my efforts to answer posts here because the fun went out of it. Way to often questions turned out to be requests to fix some bad automation that an AI dreamt up. It requires significant effort to spot that you’re ultimately trying to understand something that does not make sense in the first place. Fixing AI mistakes is not what gives me pleasure, so I simply don’t anymore most of the time.

This is why I strongly support the guideline that ALL things AI should be clearly marked as such, and AI is forbidden as a beans to answer other people’s questions.

5 Likes
5

I can’t tell the difference any more… I’ve failed the Turing test. :grin:

The thing about the Turing test is that it cuts both ways - if you can’t tell the AI from a person, you might as well treat it as human; but the corollary is that you might as well treat the person as AI.

Time to get off our high horses, maybe?

6

QUOTE:

:two::three: Don’t use ChatGPT

Don’t use ChatGPT, or similar tools, to generate answers that you provide. These tools often produce results that look good but are incomplete, misleading, or just plain wrong. If we believe that you’re doing that then we may suspend your account.

In my case I mentioned {he who must not be named} in ONE post where I clearly said that the code I posted was suggested by {he who must not be named} and received a threat of expulsion from a moderator within minutes. This was just a few weeks ago.

7

To generate answers. That is what is reacted firmly against. There are numerous posts about things people made using AI that are allowed. But if a user asks a question, and you post AI generated code as the answer to that question, tthen indeed you are in violation. Even if you mention it is an AI answer. The poster could have asked ChtGPT itself, no need for you to do so on their behalf.

And if you lack the skill to create the answer yourself, you cannot tell if AI did it right either. So the poster is the one wasting time testing the AI answer. Time you did not want to put in to thinking up and writing that answer.

If you ever tried to get an answer from an AI chatbot helpdesk of some company, then you must surely know that this is not what you want this forum to be like. And if you do: why come here, talk to ChatGPT instead. It will surely agree with you.

4 Likes
8

Nice explainer, but I’m not clear what point you’re making. Is it a bad thing? Does it matter?

My impression is that posts in the forum are moving quite quickly in the direction of connecting obscure devices (as opposed to making HA work) - but that may be a different issue.

What are you selling?
I am still forming my opinion on the “what does it mean?” and the “so what?” parts.

My personal take so far: if a bigger share of HA projects extend HA itself, that may be a sign of a maturing platform. A good thing.

Those “wait… is this a real thing?” use cases you mentioned are part of what drew me to HA in the first place. HA says I am on 379 devices nowadays, and I am still grateful when I find that someone else has already put their mind to a solution I can adopt when I need to do something about my own fair share of niche devices.

About what I am selling: great question! How happy are you with your dental insurance provider on a scale of … hello? Are you still on the phone? :sweat_smile:
No sale. No product, no newsletter, no service. I enjoyed looking through projects (that is the data), and then experiment with a way to visualize whats going on (that is the explainer). Only felt right to give it back to community.

1 Like
9

Completely missing from your analysis are the multitude of DIY projects using ESPHome. In years past if a sensor or control didn’t exist “over the counter”, we made it ourselves. Today the appliance operators have taken over and the DIY crowd has all but vanished.

By “Home Assistant project” I mean a launch announced in “Share your Projects.” The 53 records behind the headline are narrower still: integration launches found in a sample of 200 posts. To count as a launch, the opening post had to offer something usable or reproducible. I excluded requests, interest checks, reshares and later updates.

The frozen data and field definitions are in data/README.md. It is enough to inspect the classifications, but I would not call it a full codebook yet.

The initial sample does contain ESPHome and other hardware work. I coded 27 posts from Q1 2025 and 17 from Q1 2026 as hardware_or_physical_build . Examples include a garage remote, an air-quality monitor, an HVAC zone controller and ESP32 + CC1101 blinds.

Most of that hardware slice sits outside the headline analysis. I focused the main comparison on integrations because I could classify them more consistently. The data cannot tell me whether the DIY crowd is shrinking, or whether the flood of visible software launches is simply taking most of the attention while people keep building hardware.

IMHO that is an interesting question on its own. Given how varied the hardware projects are, it probably needs a larger sample and a longer timeframe.

I originally wanted to measure AI-assisted development, but the disclosure counts, to me, left more questions than they answered. None of the 20 confirmed integration launches from 2025 disclosed using a AI coding; five of the 33 from 2026 did. The remaining marked unknown because their posts do not say.

I had not considered how the forum’s AI rules are understood and enforced in practice. Public disclosure in a launch post is not a good proxy for actual use. I treat the disclosure count as a limitation, not a finding.

Reading another 1,000 posts would not have taught me how the forum’s history shapes either issue. Thanks for adding that context and taking the time to share your view!