Anyone else torn between sticking with the ESPHome API and moving stuff over to Matter?

I found myself in the exact same boat last winter when I upgraded my climate monitoring. I had a couple of old Sonoff sensors running native ESPHome profiles, which worked fine until my primary Home Assistant server went down for a whole weekend during a power flicker. All those sensors just went dark. That was the tipping point for me. I started integrating some Matter-enabled temperature sensors, specifically some inexpensive ones from Tuya, and configured them to bridge through Matter to a dedicated local hub on my network. I haven't gone all-in on removing the old ESPHome devices yet, but I run them side by side. The main thing I learned is that if you give Matter devices a local IP and ensure they aren't relying solely on cloud services, the redundancy is massive. They act like a reliable backup feed. It does mean I see the same reading twice sometimes, like my basement temperature coming from both the old native sensor and the new Matter one. I set up an automation rule in Home Assistant to always prioritize the Matter state if both are reporting, and it keeps the dashboard clean. For critical things like security and temperature tracking, I think the headache of managing two connections is a worthwhile trade-off for the peace of mind that the system won't completely fail if my main server hiccups.

I run a mix and I'd say don't pick a side at all. My door sensors and the garage contact sensors went Matter last spring so my wife can see them in her phone's own app, since she refuses to open Home Assistant and I got tired of being tech support when I'm at work. Everything else, especally the sensors that feed automations, stayed on the native ESPHome connection because updates show up faster there. A motion event hitting an automation thru Matter seemed to lag a beat or two compared to what I had before, so for anything timing sensitive I kept the original path. As for the duplicate device question, yes, the same sensor can show up twice, but they don't fight. You just end up with two entries and it's on you to ignore or hide one. I made that mistake once with a humidity sensor and had two graphs disagreeing with each other til I disabled the one I wasn't using.

My rule now is simple. Anything the household touches without me gets Matter. Anything that only feeds my automations stays as is. Slow rollout, one device type at a time, and you'll know within a week wether it earns its place.

One thing I noticed after testing a couple of devices on both connections at once is that the reporting behavior differs more than I expected. My battery powered door sensor updated nearly instantly over the native API but through the fabric it seemed to batch its reports to save power, which is normal behavior, just slower. So if you duplicate anything, expect the two entities to drift out of sync for a second or two during fast changes. I ended up giving the Matter versions clearly diffrent entity names so nobody in the house accidentally builds an automation on the wrong one.

Went through this same decision on my shop setup last year and the thing that settled it for me was flash space, not philosophy. Adding the Matter component to a plain ESP32 build pushed my compiled image from around 1.1MB to nearly filling the 4MB chip, and OTA updates got noticeably tighter to squeeze in. So now I only put Matter on boards with the bigger flash and leave the small ones on the native connection. My question for you is more practical: are you planning to keep one device serving both connections long term, or is the plan to eventually pick a lane per device? Running doubled entities forever sounds like a bookkeeping chore.

I only moved a handful of devices over, mostly stuff other people in the house touch, and left the automation sensors alone. The messy part for me was keeping the duplicates straight in HA, two entities per room and I coud never remember which was which. A cheap label maker actually fixed that. Each sensor got a physical tag with the room and which feed it was on, and after a couple months of the Matter side reporting fine I just deleted the native copies for those rooms and the confusion was gone. If you do run both for a while, tag the physical units early, it saves a lot of head scratching later.

I've been looking at this too, and I'm worried about how much time it takes just to get everything talking to each other in the first place. Setting up the original system was tricky enugh, and adding a whole other way for the stuff to communicate feels like making the initial setup ten times harder. I just want something that works without me having to constantly check two different places to make sure the numbers match up. If I add a few devices to both systems, how much extra work do I have to do in the main software to keep the connections clean?