LabVIEW

cancel
Showing results for 
Search instead for 
Did you mean: 

LabVIEW and the AI coding revolution: are we being left behind?


@alexderjuengere wrote:

As a non-native speaker, I actually used AI to help me phrase those last three sentences, lol... ^^


Ai is good at some things, awesome at others and horrible on many things. Often they overlap. 

I just did a fun thing. A buddy recorded a song where he's ... imaginative, with his note placements compared to the instruments. I uploaded it to suno which analyzed the song and wrote a description of it and then i let it do a cover. It's pretty close with better singing.

G# - Award winning reference based OOP for LV, for free! - Qestit VIPM GitHub

Qestit Systems
Certified-LabVIEW-Developer
0 Kudos
Message 11 of 38
(3,753 Views)

My quick opinion: yes. The optimistic picture is that LV has honestly missed other big software revolutions, and we as LV developers have been fine. We missed the internet revolution for the most part. Completely missed mobile, but that was out of NI's hands and not needed anyway. But what is unique with AI is that it has appeal to the same developers that LV does; technical people that aren't necessarily programmers. Recently, I connected a power supply to my PC and asked Claude to create something to communicate with this device. Gave it links to the IVIfoundation.org site and waited. In a few minutes and some back and forth, it made a VISA layer in rust and wrappers in python to easily use the instrument. This is showing my age, but this was as magical as seeing LV just communicate with an old GPIB instrument with a GUI with my first LV (version 3.1) in the mid-90's. I feel like the "slop" descriptions thrown around about AI generated code is very similar to those of us who had to live with "LV isn't a real programming language" for many years. It funny hear many of the same arguments being thrown around like "It generates worse code than me." or "You don't know what this compiles to." In the day LV probably had both problems, but we lived with those issues because of the productivity gains or that it was frankly, more fun. LV felt like the future. 

 

Currently AI assisted programming is moving so quickly. Frankly, even the way I coded an example above with a chat is outdated. People have moved to agents. But agents are so two weeks ago, the bleeding edge is now loops. I'm just not sure how Nigel keeps up with this pace. Even if we get what was shown at NI Connect in late July it is very far behind what others are doing with AI. To be fair if we get what was shown at NI Connect, we are only 6-8 months behind something like Xcode. So maybe it is unfair to compare LV to the latest and greatest coding tools from the frontier model makers. One could argue both Xcode and LV are very far behind.  

 

While I agree that LV will be used into the foreseeable future because maybe our customers require it or for legacy reasons. Sitting down and opening up the latest LV after using Claude code or Codex, is like the feeling I had after sitting down with a plain text editor and a C compiler all those years ago after using LV. I *can* do it this the old way, maybe for reasons I should be doing this the old way. But I certainly don't like doing it this the old way.

Message 12 of 38
(3,700 Views)

I think what a lot of the "ai makes slop" people are missing is that yeah, if you say "Make me F@cebook" it'll churn out garbage.

 

But if you use it as an assistant (think quick drop on steroids) then it can be a giant productivity boost.

 

Some concrete examples where it would've helped save time and let me solve the "hard problems":

 

1- Creating documentation

 

2- The FTDI driver library is very rudimentary. All of the icons are identical, they use a homebrewed error handling system, references can't be passed to correctly handle flow control, etc. There are maybe 50-60 VI's. To make this usable, I had to spend a bunch of time creating VI Scripting tools to fix this bad code. After running the script, the functions had proper labels, actual error handling, correct flow control, correct library namespacing, etc. That was 8 hours or so I could've spent handling my actual problems.

 

3- I have some data from a customer in their semi-proprietary, undocumented XML format. I would really like to switch that to a better, more universal format. Manually parsing through their XML files and creating a converter is a real pain. AI is really good at processing text; I can use AI to make a Python script to convert between the janky XML solution and a standardized JSON solution, and it'll even create documentation (that I can manually verify) to help me understand that format. Note that I can use this to make a script I can rerun and debug; I'm not asking the AI to actually convert it for me. Just to make a conversion tool. I can't currently do this just inside LabVIEW.

 

4- I have a program that currently reads a handful of sensors. I got a request to add one more to it. This is simple enough, but it'll involve updating DAQmx calls and plotting functions, modifying the UI to show scaling parameters, updating init files (in a backwards-compatible way), etc. Nothing even challenging about it, but it's a few hours that I'd rather spend elsewhere.

 

There's a common complaint that AI can't solve new problems. My reply to that is... that's why it's so useful!

 

I want to spend MY time solving new problems, but I often have to spend a ton of time implementing known solutions to known problems. If AI can do that for me then I get to spend my time on the "interesting stuff".

 

Message 13 of 38
(3,668 Views)

@BertMcMahan:

I want to spend MY time solving new problems, but I often have to spend a ton of time implementing known solutions to known problems. If AI can do that for me then I get to spend my time on the "interesting stuff".

 


@BertMcMahan

I completely agree with you. I'd like to delegate at least the repetitive tasks to AI, so I can dedicate more time to architecture, solving new problems, and high-level design.

0 Kudos
Message 14 of 38
(3,642 Views)

Last time I was down at NI Week or NI Connect or whatever we're calling it..  Nigel was generating code.  This was about 3 years ago.  Give it a prompt, and it spat out working LabVIEW block diagrams.. nothing complex.. but that was three years ago..  I sat down with the demonstrator, and we gave it simple prompts and it wrote the block diagrams..  it wouldn't take anything crazy complex, but for simple tasks..

 

My guess is, the hold up is how to make it efficient so it don't become a token monster and insanely expensive for NI to operate.. or we're waiting for NI to invest in hardware..

 

Bright side, (as said by others).. it's code guaranteed to be written by a human..

 

 

0 Kudos
Message 15 of 38
(3,543 Views)

We have craft beer, soon we will have craft code 🙂

0 Kudos
Message 16 of 38
(3,533 Views)

@wldgoose wrote:

Bright side, (as said by others).. it's code guaranteed to be written by a human..


 

I often have to manually analyze patent applications

 

AI is highly effective for summarizing patent applications, providing a solid starting point for diving into the details. However, domain expertise remains essential to guide the AI, ask the right questions, and prevent hallucinations that could lead you astray. Ultimately, legal counsel is indispensable to counter-check the results, as all outcomes must be legally binding

Message 17 of 38
(3,477 Views)

Emerging AI/agent bridges: Community developers have already built LabVIEW MCP Server Toolkits (e.g., implementations by Jan Göbel/jgoebel and others, available via VIPM and GitHub). These implement the Model Context Protocol (MCP) — an open standard that lets LLMs (such as Claude Desktop) discover and call LabVIEW VIs as “tools.”

  • You can register VIs so an external AI agent can execute them, pass parameters, receive results (including images or front-panel captures for visual reasoning), and potentially drive higher-level scripting/automation.
  • This is essentially the “LabVIEW MCP server exposing VI Scripting as tools” idea proposed in the forum post — and it already exists as third-party open-source work (requires recent LabVIEW versions and some dependencies like HTTP server and JSON toolkits).
  • It does not give the LLM direct “understand the entire binary VI” superpowers, but it lets agents act on LabVIEW programmatically from outside the IDE, which is a major step toward closing the gap.

That said, being able to use AI effectively for architecture and docs but then being “on your own” for actual code review, refactoring, bug tracing, or generation is frustrating — especially when younger engineers expect AI to be a first-class collaborator.

 

It is a legitimate risk in certain segments, particularly among new or generalist software engineers who have grown up with text-first AI tools and may default to Python + hardware libraries or other ecosystems unless LabVIEW demonstrates clear productivity advantages including AI leverage. In LabVIEW’s traditional strongholds (test & measurement, industrial automation, HIL, real-time/control, hardware-abstraction-heavy work), the dataflow model + seamless hardware integration still deliver massive value that is hard to replicate purely in text, even with AI. Many experienced users argue that a well-structured LabVIEW diagram is often more readable and maintainable than equivalent text for parallel/dataflow problems.

 

 

Message 18 of 38
(3,448 Views)

Hi All,

I just wanted to toss in my “noob” thoughts…

I am not a coder…

I am not a LV programmer…

I use LV all the time…

I like graphical (my A_D_H_D/dyslexia/visual-perceptual learning challenges favor pictures over words)…

 

I think that there is very high value in a noob like me typing “build me a “Hello World” vi and AI returning just that.

Now add seamlessly integrating TestStand, DIAdem, SystemLink, Flexlogger, etc. 

 

I have kicked Nigel’s tires - not much luck…

I have walked down the “Python node” path…

I have not done the “DLL stash” road…

 

But…

 

I think NI should be pouring in every resource to get AI to or above the level of AI & text code…

 

I am going to let out NI Rep know I will be a lab rat…

NI needs to make this happen…

0 Kudos
Message 19 of 38
(1,291 Views)

@NI_LV_82_User wrote:

I think that there is very high value in a noob like me typing “build me a “Hello World” vi and AI returning just that.

Now add seamlessly integrating TestStand, DIAdem, SystemLink, Flexlogger, etc. 


For most such questions i think the Example Finder is golden! As for AI i had good help of ChatGPT last week in solving an issue and i could paste a screen dump of the code and it interpreted it pretty well. (It got some things wrong, but after corrections from me i got a good suggestion that actually solved my issue)

G# - Award winning reference based OOP for LV, for free! - Qestit VIPM GitHub

Qestit Systems
Certified-LabVIEW-Developer
0 Kudos
Message 20 of 38
(1,041 Views)