Hello guys,
I’ve been using various coders throughout the years whenever I need a custom indicator, but with the advent of AI and hearing all the praise, I’m wondering if I could just as well do it on my own with the help of an AI/LLM?
Anyone have any experience here?
Note: I am NOT a coder, but I’m not completely illiterate either.
Thanks in advance.
It really depends on what you want to do. Something easy, you can most likely do yourself. A lot of non-coders have successfully used AI to create indicators and strategies. However, I’ve also seen people that do so get stuck because they seem to not know how to ask the AI to fix things. Honestly, I think it’s worth it if you want to test many different ideas.
I’ve been using AI to code (not a coder before) with great success. I started to use it for AHK v2 (for UI automation). Then python TWS API (mainly for alerts). Dabbled slightly with Ninjascript.
I think it’s been almost 2 years since I started using AI for coding. They’ve gotten SO much better.
I use all of the free AIs and bounce back and forth between them. Never paid a cent.
Claude Sonnet 4.5 is my go to coding AI. It gives you a generous amount free messages. However, if your script is 700 plus lines of code you run up against the pay wall pretty fast. Also, start new conversations to get more access.
Gemini 2.5 Pro I think is one of the best at programming. If I can’t use or are having trouble with Claude I go here.
Grok is great for larger scripts and is very good all around.
ChatGPT just recently gave free access to its best model for free.
Perplexity is good for simpler questions when you don’t want to use up time with the other better AIs. For example if I need a large data set fomatted in a certain way, I go here.
DeepSeek and Qwen are supposed to be very good as well. I only use DeepSeek occasionally. I’m a little wary of the fact that they are based in China.
Hello guys,
Great. Thanks! I guess I don’t have anything to lose by trying it. 
FWIW, this indicator I’m needing now is very simple
If it’s simple code and is for your own personal use, then by all means use AI and learn the process. It’s always good to know at least some coding so you can see the flaws in AI generated code and correct it or clean it up. AI is not flawless. I’m no stranger to programming, but I still use AI in some parts of my work to speed up my work, but often have to make changes to adapt to my situation. And sometimes AI is flat out wrong. Happens more often than you might think.
If you’re writing code you’re going to share with others or sell to customers, I would strongly recommend hiring a decent programmer who knows both what they are doing and are familiar with the platform. Not every coder understands NT well enough to write reliable code.
I often use this example that AI is like a Formula 1 car - nice, fast, shiny. You’re going to get widely different results if your grandma is driving it vs Max Verstappen.
LLMs are great at implementing logic. What you have to do is convey the logic you’re trying to accomplish into words, accurately and completely.
One specific coding task I have found LLMs to be excellent at is RegEx. The syntax of RegEx is complicated to me. Although, if I just express in words the logic I want to match, LLMs are great at this.
In high school I remember drawing flow charts by hand for a precursor for a program. I think the best way to talk to an AI is to accurately explain the “flow chart” of the logic you’re trying to accomplish into words, language.
Exactly. That’s a great example @PCAutomation. AI does great with small, self-contained and well-defined functions. But it’s still very much lacking in putting together large complex solutions.
Your RegEx example is perfect. I recently asked AI to give me an email syntax parser/checker which requires RegEx. Took a few seconds instead of having to struggle through the syntax for a few minutes.
I too started programming by drawing flow charts and flow diagrams in a Structured Programming class in college in the late 80s. This teaches you how to think in terms of how computers work and helps write well structured and easily maintainable code, something that’s missing if you’re just jumping into programming with the help of AI and no other background in programming.
I have coded with NT for about 14 years - but would not call myself a programmer - i can end up doing most things, however.
I use ChatGPT free version and it’s 50/50. You have to be able to understand some things to be able to direct it. It gets things plain wrong and consistently uses SharpDX.Mathematics - which you can’t use in NT. I now have a saved template to direct it with every new chat - its memory does not persist across chats which to me seems a little pointless.
What it is good for is getting the idea mapped out - eg you want a flashing text box at the top right of the chart - it can give you the scaffolding on which to build. If you don’t know ninjascript or C# it’s not good. I dabbled with Claude which is much better but i find it more difficult to copy and paste for some reason?? Still exploring the AI universe
Hello everyone,
Thanks for the great replies.
Well, yesterday I created my first custom indicator using Claude and honestly I think it’s quite amazing. 
Up until now I’ve exclusively used Upwork for such projects. If it’s a small and less important indicator, I’ve used cheaper coders trying to break into UpWork and the results have been mixed. For more important stuff, I’ve used my go-to coder who’s more expensive, but delivers great results.
The indicator I created yesterday would be categorized as a less important indicator.
Anyway, it took me around 1 1/2 hours with quite a few iterations to achieve the desirable result. Only twice did I actually get errors when compiling and Claude fixed it immediately. The main time consuming part was re-booting Ninja for each new iteration of the indicator.
Claude immediately understood my ‘needs’, so no communication errors (which quickly happens with a hired coder) or misunderstandings.
No waiting for a reply from the coder when de-bugging.
And if I wanted to implement a new feature Claude was only too happy to do it for me offering improved solutions even. This is where a coder might get annoyed or start asking for more $$$.
I will definitely use this more for similar stuff. Tonight, I plan on getting Claude to improve an indicator I wasn’t quite happy with through UpWork. I’m sure it will be handled smoothly.
I imagine it will be great for VBA scripts and Excel formulas, too. Also an area where I’ve used hired work in the past.
Claude and ChatGPT are very good for designing and modifying indicators. One word of caution you have to be exact in your requirements and if it goes down the rabbit hole it can waste a large no of hours on fruitless quests to ‘solve’ something. Is better if you have an understanding of NinjaScript as it still OFTEN gets things totally wrong or adds library references that NT8 can’t use eg Math - I have now added this to my .md file as it did it so frequently. But overall i am delighted and it has increased my productivity ten fold.
Well, since I made this post I went really deep into this and have around 500 hours of coding with Claude Desktop now - and mostly Claude Code for the last few weeks. Building Ninjatrader indicators is working very well, but interestingly, there can be some time spent debugging fairly trivial (I would think) stuff.
Yet, at the same time, Claude created an indicator for me that my earlier go-to coder quoted me $500 and a solid day of work to accomplish. Amazing.
My main project is a very complex analytical engine. I’m not a coder, but it’s progressing along very well, although slower than I had hoped.
Oh, in addition to designing a few Ninjatrader indicators I actually redesigned my entire indicator suite from scratch in optimized form cleaning out every minor bug and annoyance from earlier.
I can’t imagine how freelance or small programmers can survive this, but maybe many people (like me a few months ago) don’t know what’s actually possible with AI these days.
PS: I’ve been using ChatGPT as a reviewer for my main project and he keeps finding stuff Claude misses. Interestingly, though, the few times I’ve tried to have ChatGPT build something he seems to actually struggle? Once I actually had Claude review Chat’s work and Claude ended up doing it instead after ChatGPT had failed 3 times in a row.
That’s a way for freelancers and programmers to survive. 
What I see often is people that have no clue about coding will say things like this. Look on x/twitter for a few minutes and you will see every vibe coder post something like that. Although I do see AI reducing jobs, it also amplifies the abilities of those that actually know what they are doing. I still think it’s cool that people are building things.
The goal in software has always been to build with bigger bricks. AI is the latest in a long series of bigger bricks. Definitely a game changer. Some will get pushed aside and others will rise up, adapt and ride this, the latest wave
So, you’re saying I have no clue about coding?
I may not know how to write code very well, but I definitely know a thing or two about coding. Most importantly, I’ve worked with coders through multiple projects. Most of them smaller jobs through UpWork and similar, but two larger projects as well.
Claude with my guidance definitely outperforms all of them. Zero complaining. If I want to start from scratch he’s only happy to help out. More creative. No waiting for a reply on e-mail or phone.
What I specifically said was freelancers and small programmers.
Please tell me why I should use a freelance programmer now as Claude literally does it better and faster and friendlier?
Claude Code literally produced a complex indicator on the first go in less than 20 minutes. One that my main programmer who I’ve been happy to pay $$$ quoted me $500 for and I was uncertain about the value, so I said no.
Maybe what I said earlier was a bit exaggerated. It could just be my bias from working on complex projects across large teams, including projects with budgets in the 8 to 9 digit USD range. From my perspective, knowing only the basics doesn’t really mean someone understands coding. If the LLM were taken away, many people wouldn’t know how to proceed on their own. That said, I think Claude is awesome. I use it myself. I also think it’s great that non-technical people are using these tools to build things and learn more about coding. I don’t really care whether you work with freelancers or small programmers. If Claude is working well for you, then that’s what you should use.
My points were the following. First, you don’t actually know why ChatGPT was struggling while Claude wasn’t. Someone who understands what they’re doing can usually guide ChatGPT toward the solution they want. For example, a freelancer or experienced programmer could probably get the result you’re looking for more easily, simply because they know how to work with the tool and understand context, whereas you are not a coder. This isn’t necessarily a limitation of the LLM itself. At this point, I think both models are very capable. The real limitation is the user.
Second, because the user is the limiting factor, an LLM will amplify the abilities of someone who understands the context and knows how to reach the desired solution. For example, an experienced programmer will often think about things like time and space complexity, which directly affect how the code should be written, while someone without that background probably wouldn’t consider those factors. As a result, the LLM will produce different quality outputs depending on the level of understanding of the person guiding it.
Right. And when you’re saying you’re working with complex projects in the 8 to 9 digit USD range that is vastly different from coding up an indicator in Ninjatrader or writing up code in VBA.
I never said AI will replace programmers as a whole as these agents definitely need someone with domain experience and knowledge to guide them.
The only reason I’m able to have my larger project move along (450 hours sunk into it so far) is because I’m good at big picure thinking and covering every need from A to Z - taking my time to spec out and plan things very carefully in advance with no technical debt (if I can trust my agenst that is) moving forward. Testing, testing ,testing. And documenting very thoroughly.
I’m basically replacing an old system which was programmed for me which is basically spaghetti code with zero documentation and a lot of compiled code. Wise from that experience and failure I’m programming everything with independent modules and layers where nothing is ‘entangled’ or intertwined unless necessary.
I don’t doubt that a good programmer can develop better than me with his domain knowledge using AI tools, but for me, I don’t see the need any longer with my use case. I also have my own domain knowledge in trading and it’s beautiful to now have to spend time explaining to programmers (and getting objections) what I want to do. It’s kind of amazing, but I feel like Opus 4.5/4.6 very quickly “understands” what I need and even come with suggestions to improve it - like I’m sure the average freelancer wouldn’t.
… seems like it to me, at least at the freelancer and small programmer level.
Probably just me, but literal zero technical debt does not exist.
Anyways, I’m glad it’s working for you so keep at it.
I think what you are attempting is worthwhile! It’s what all traders should be doing to make themselves more profitable. Jump in, get wet and figure it out.
Friendlier…?! Could not agree more! If S overflow were a football I would punt it on first down. All over the place, downvotes for perfectly legitimate questions.
Like Mr Roboto, I used to be in the software world of big spend, big code. Now, as a trader, software is falls into the category of overhead, an expense. AI can perform a well focused task faster than any coder ever will - that is just the “John Henry” reality of things. And for me, that has been a great thing.
If you are a trader and you are getting the results you need then proceed and think no more of it. Why ask for opinions? Ultimately, P&L is the only opinion which matters.
If you are a software vendor, then at some level I think you really need to understand what AI is producing. Knowing the code will allow for more thorough testing and fewer field issues.
FWIW
one of my concerns as well - a coder that has some experience w/ NT