
What I like best about Flux is that it combines design assistance with teaching in a way I haven’t found anywhere else. My background is in finance, not electrical engineering, and Flux has let me work through PCB designs I never thought I’d be able to develop on my own while actually learning why the design decisions matter.
I’ve already manufactured and successfully tested a four-layer protected power-switch board based on TI’s TPS25983 eFuse, and I’m close to sending out two more designs: one built around TI’s BQ25756ERRVR MPPT solar charge controller, and another using three TI TPS2HC16 automotive high-side switches tied into an onboard Arduino Nano BLE Sense Rev2.
The biggest value is how much of the development process I can work through in one place. I can upload datasheets, compare components, work through schematic and layout decisions, ask for routing guidance, estimate BOM costs, interpret manufacturing quotes, build test plans, and then come back with real-world test results and keep iterating. For a self-funded business, being able to do that much work before paying for fabrication, assembly, or outside engineering help is huge.
And when I don’t understand something, I can just ask. Instead of stopping to search through tutorials or documentation, Flux can explain the problem in the context of the board I’m actually working on. That combination of building and learning at the same time is what has made the biggest difference for me. It has completely changed what I thought I could realistically do myself. Review collected by and hosted on G2.com.
The biggest downside for me is the hosted interface. For high-precision manual PCB layout and routing, it is still less comfortable and less responsive than working in a traditional locally installed design tool.
The other issue is credit usage. Flux is incredibly powerful, but it can burn through credits quickly if you do not give the agent pretty tight guardrails. Sometimes that is entirely on me, but there have also been cases where the agent spends a long time thinking, comes back with essentially the same answer, or ultimately says it cannot complete the task. The feeling of accidentally blowing through your credit budget for the day and then having to stop using the agent until the limit resets is not great.
I have learned to be much more deliberate about how I structure requests and when I let the agent take a bigger swing at something, but I still think the usage model can make experimentation feel expensive, especially when the outcome is uncertain. Review collected by and hosted on G2.com.