
I don't think I am smart enough.
I am also not sure how the autogen that would check the work would be implemented.
I think I would need a whole forum of autogen experts, a great documentation set like Ignition has along with the 50 hours plus of training videos, and a paid support line for live help.
Thanks for sharing this, this is so cool
Here's an inspirational quote that is intended to dissuade this line of reasoning:
'Whether you think you can, or you think you can't--you're right" ~ Henry Ford
Henry Ford's statement is wrong, and I know that for sure because I thought I could some things I couldn't. lol But thanks
It means (or at least thatâs how I interpret it) that if you think you canât, you wonât even try, and if you donât try, you for sure canât do it.
Wether you actually succeed if you do try is almost irrelevant. I donât remember who said âI didnât fail to make it work, I merely found a thousand ways it doesnâtâ.
Or something like that.
Regarding Ignition, I think " pturmel " is better than ChatGPT
At the rate that autogen teams are being developed per Mathew Berman's videos, maybe I will be able to automate some JSON soon personally.
I would imagine the new LLMs are working much better?
I read somewhere things are slowing down because everyone is (trying) to block (free) access to their data.
Censorship also been quite the focus...
Altho it seems some improvments were made in the o1 model for coding and math. but dont think this one is free to use, so cant tell you if its any better
I was a little alarmed at the pace of Manus the other day, but then it seemed like it was very API driven with some webpage generation.
Hard to tell what is cutting edge, and what is just fake.
Still watching what Alex is doing.
Today I wanted to batch create a few hundred getBit expression UDT members to parse out bitmapped alarms from DNP3 counter objects... String encoding the JSON, escaping all the "s, building a giant concatenation, and reversing the process with the output ended up saving me time vs ChatGPT with output that was easier to test. When Excel can get the job done, sometimes you need to be a freak in the sheets.
When I started reading your post, that's where my brain went before I even got to your mention of it. I still use excel for quickly assembling stuff like this because I know there won't be any unexpectedness in the output.
Just came across this post looking for what is the state of using AI to generate ignition projects. Given that itâs almost 2026, whatâs the current state? Do we have a âCursorâ for automation controls engineers yet?
The short answer is no.
Early in the new year IA is planning to release a first party MCP module, free for experimentation here on the forums. The initial version of that module will provide a designer workspace where you can author your own MCP tools/resource/prompts in a manner similar to WebDev, but it also contains all the pieces "under the hood" to allow modules to contribute T/R/P as well.
That, to me, is a missing fundamental piece of the architecture - you need something deterministic that has specific domain knowledge baked in (e.g. the exact JSON schema for Perspective components) before you can get to a reliable, general-purpose "turn a description in prose into a useful Perspective project" level of automation.
One thing that remains to be seen is whether MCP is actually the answer 'long term'. It's a very underspecified protocol when you look closely at it, and modern cloud LLMs are burning so much money rapidly consuming training data and getting better (incrementally) at ever more niche domains given the right framework.
Suffice to say that IA is fully aware of the opportunities here (at least for SCADA/HMI; not touching PLC code with a ten foot pole), but we're also very aware that AI is also on its trajectory through the hype cycle right now, and we'd rather move responsibly later than rush into something early that burns goodwill.
Thanks for the response Paul! Excited to play with the MCP Module when itâs released!
