Well i have been using Homeassistant and Esphome for years now but the sad and for now final conclusion is, that one can not trust in device integrations built on the HomeAssistant infrastructre unless one is willing to modiufy/adapt perfectly running systems about four times a year. HA is a great framework but in recent years has been undergoing many "improvements" that might be understandable for an amateur project but not for devices you have to rely upon.
From now i will use EspHome only for playtasks and unimportant stuff until it proofs to be a reliable and backwards compatible platform und not full of design flaws. For reliable and important devices the HomeAssistant QA is not meeting my expectations and i better invest my time in native ESP-IDF development - bullet proof and not subject to "breaking changes" being more or less untested and breaking my nerves without improving anything. So ESphome - Thank you and good bye !!!!
e.g: Modbus integration, Custom integrations, class definitions, naming conventions, lack of well defined interfaces, .....
Do you have any more specific examples of how it's worse vs when you first started? Are there any particular bugs that affect your experience? What improvements have broken your workflow? You said:
Modbus integration, Custom integrations, class definitions, naming conventions, lack of well defined interfaces
But you did not give any actual examples of how those things don't meet your expectations. And since they don't, why can't you just revert back to an older version that you like better?
And are the issues ACTUALLY with Home Assistant/ESPHome or is it just a matter of migrating to the newer frameworks and/or fixing other issues like WiFi?
Expectations i did not have any but HA offered a framework that allowed to implement custom integrations with reasonable low effort. All that implementations were working well but became unfunctional as fundamental classes had been redesigned or sometimes just renamed without a need. I am in computer business since CPM and DOS 2.0 and of course development is rapid and ongoing 
Revert back would be a good idea maybe - to HA 2023 ? Can you tell me how to ?!
Issues ? No problem to switch lights on and off for sure.
A Modbus integration ? 2023 not existing - 2026 existing but not functional.
I am not crying for help - just crying
Some developments have taken a good path (Arduino omg) and many many programmers do a really good job. But at the end changes are made disregarding any backwards compatibilties and by dropping custom integrations without offering functional alternatives is not wise.
I could list tons of functions no longer supported, deprecated, dropped, reasons given or not but i am tired of App.register_component no longer supported messages.
You want to implement a módbus device ? Oh your device is echoing ? Oh an echoing RS485 RTU device ? Hmmmm - maybe in 2027 ?
Even AI is fully confused btw. - There are no ESPHome standards as things change like the weather nowadays. If one named a pin "direction_pin" it has to stay like that even if the next guy likes "rts_pin" more. Set conventions and stay with them for a long term. Extend classes but try to make them upwards compatible. Improve code but test before releasing it.
That's not saying much...
Not saying much either
For me it became easier to rely on espidf documentation (5.5.4 right now), develop code that is bullet proof and not in danger when the next weekly "improvements" arrive. Did you not read the "release documentation" ? Not even the newspaper !
Well, AI has problems because it is trained on old data. Things change fast in the HA world.
I think you and I both have something in common with the AI, and were trained on OLD data and struggle with the new ways...
I have no sections in my dashboard. I have just a couple of automations built by the UI that I have not ripped apart and changed yet. I use device_id in the bare bare minimum of places. The new stuff doesn't always work for me. And for me that's OK, I figure it out when I have to and if I don't have to, I don't. And I just let the project move at the speed it needs to. Just don't ask me about Add-ons/Apps, that is likely to fluff my fur.
You got fur to fluff, mine went 20 years ago old timer (yes us AARP types or older).
P.S. I love it when Claude (AI) responds with "I see where you are going now and you are correct" I feel I'm back in the school system training my new batch of student helpers all over again
I understand your concern, However, I think that is just the nature of a small (compared to global SaaS), community built program - backwards compatibility is simply not a huge thing for the developers trying to add more features. However, isn't that technically your responsibility as a custom integration maintainer to read the release notes before updating?
As a side note, personally, at least in ESPHome, I have been noticing huge improvements when it comes to naming conventions. For example, using one SPI driver for multiple screens, unified audio pipeline...But i understand that it's taken a long time to get there.
As for rolling back, I'm not sure how far back you can go without breaking something, but you could try to restore an old backup (if you have any) or running something like ha core update --version 2025.2.5
And just out of curiosity, what's not working with the Modbus integration? The only device I have using modbus is our generator, which is covered by Genmon.
I for example are unable to use my eq3 bluetooth TRVs. With all ESPHome versions in 2026 they are not available or become unavailable very soon. With one from 2025 they are reliably connected by BTproxies.
You realise EsPHome is repackaged Platform IO, which in turn is repackaged bits from various idfs such as Arduino, Espressif and others. Some is even gung-ho open source, vibed on the fly.
The fact that it works at all is a small miracle.
Back to machine language may be desirable for full control, but along comes a fab designer that changes the chip at the layer and transistor level, and your carefully tuned timing loops all of a sudden don't work no more.
We rely on modularity, compilers, building blocks, and even AI helper bots to break up our problems to manageable bits and assemble them in unique ways to solve our pressing problems. You trust that each of them works robustly, as documented, the APIs are well defined, and hope the final result is functional.
For high reliability issues like life support and multigenerational space missions, you better pore over the code a bit more closely. You don't always get to update your code at whim.
For the rest, dragging a few building blocks around the screen is all they want, and they are satisfied with.
Feedback is good and drives us to improve, endlessly.
There is no perfect code - you are kidding yourself if you believe this. Pick your tools wisely, and wield them well in anger!
Just use them with the version that does work. Help others by telling them the versions that works .
Can you file a bug report in ESP-IDF or ESPHome? I'm sure the reason that no one has fixed it is because not many have your specific use case!
I thought several times about reporting it, but was unsure where.
They had optimized a lot in ESPHome lastly, so i really want to use it.
But with unavailable TRVs my heating automations dont work. Fortunately its summer now
There are myriad issues with esphome. I have several devices that worked fine, then after an upgrade, they magically didn't anymore.
I feel OP's pain.
Hey, if you want to report it you can do so here: Issues · esphome/esphome · GitHub
Did you get any compile depreciated errors? Can you be a bit more specific? Not working can mean a lot of different things depending on who you ask 
Nope. Compiled and installed fine, just doesn't work anymore.
It's all hacked together from 4 or 5 different contributors, so nobody "owns" it.
Things like this are what prevent HA from becoming mainstream, and keeps it a "nerd toy".
Too little specifics but lots of shade. Blaming ESPHome may be a long bow to draw.
You've thrown down the gauntlet here, and the suspicion that the tradesman is blaming the tools for shoddy workmanship.
Turn on detailed debugging, post your yaml code, compile log and a small segment of the following run log, (nicely formatted please), and you can put your money where your mouth is.
Otherwise some may think you are simply trolling.
[Put the link here to a new thread so we don't get too distracted here with debugging something that may not be ESPHome related at all]
I'm afraid I have to agree with IOT7712 here. If you don't give any specifics, no one can help.
As for HA becoming mainstream, why would it? Most people are perfectly happy with their little cloud based Alexa setups, and then HA steps in for those who want a cut above.