I'm returning to the embedded world after a 5-year hiatus and Expressif has really exploded onto the scene while Adruino, AVR, and microchip seem a bit more sidelined. Especially for cost-conscious projects, or regarding connectivity options (ie. wifi).
Does it make sense to use AVR/Microchip for new projects of low-medium complexity or would most people recommend ESP-family mcu boards for new-starts?
Initially hamstrung as a startup they had Low chip yields, poor software support, lack of designer fluency compared to more established stm, nxp, ti are western brands. Their chips are overpowered for many small embedded products, underpowered and more difficult to dev/deliver software compared to a raspberry pi or custom odm arm chip running posix linux & docker. Ai thinker wins best in class on processor power & power, but most electronics are continually plugged in (not battery) so that advantage is not huge deciding factor unless you are doing sustainability engineering and you realize how much more power+carbon offsets you required to approach carbon neutral.
Ai thinker is popular in shenzhen China ai market oriented contract design & manufacturing in China military-defense products & dystopian monitoring than the western chips. Those products, found only on the mainland are not ones that most ycombinator people are familiar with.
Ai thinker/espressif would design ref boards & interfaces, but of course they are not guaranteed of actual unit sales. Ti, nxp, even amtel have industry targeted chips that are certified for automotive, etc. Whereas ai thinker lacks the equivalent global sales force (and emb. Systems people have job security as long as they don't rock the boat and make their teams learn new ecosystems)
Also a lot of new embedded products in industry aren't new, the software is very old, much older than the the chip. The avg. Embedded systems devs learned design pre-iphone and its a 9-5 job, they are told what to do, and usually aren't starting from a greenfield since the devCost burden & supply side risks are higher changing chip lines (and dev environments), when viewed from a profitably lens.
Disclosure: My company specializes in esp32 & stm ag-tech.
Hobbyist? Prototyping? Possible manufacturing?
I love the ESP product family and they're in a ton of cheap IoT goods.
I think if you're talking about hobbyist stuff, there's no reason not to start with an ESP32.
FreeRtos makes scaling your codebase a dream compared to the usual AVR stuff, they're powerful, they can be low-ish power, there's a peripheral for everything, and the IO Matrix is genius (no more worrying about having specific pins open for specific peripherals like I2C).
But for production, I don't think ESP is up to the standards of *NXP and TI when it comes to power consumption (although I might be behind on that with the new RISC parts)
I also don't think the Wifi stack on the ESP-32 is quite as mature, and if you use off the shelf libraries there are a ton of pitfalls when it comes to reliability
But the tooling is great, and there's a ton of activity surrounding them so for me none of those are deal breakers.