Using AI to speed up development

[GPGeorge]

I've heard that you must continually remind Chatty, et al, of your coding standards. Also, you must insist that they are used consistently. Does that correspond to your experience?

 

[RaptorRaptur]

well I know this isn't 'Access', but I use it a lot for refactoring T-SQL and converting T-SQL to Snowflake SQL, it's quite good once I learned a few gotcha's and now instruct ChatGPT to avoid those pitfalls.

One of the things I have enjoyed leveraging is the limitless amount of granularity you can include in telling ChatGPT's memory about how you want your SQL formatted. I have seen a lot of SSMS add-ons in my day, Redgate's SQL Prompt being the main one, but absolutely nothing - no tool in the world - compares with my ability to verbosely tell ChatGPT every minute detail about how I want it to format code. I could even tell it to produce a line break after x-number of characters in an IN clause specifically, or put every individual boolean expression in a where clause on the new line, etc.
That's one reason I use ChatGPT and not an AI tool without a memory, to me without a memory is worthless - my whole productivity gain IS about the memory

Do you have any idea how good ChatGBT would be with VBA and/or the Access way of handling SQL?

 

[GPGeorge]

It depends on a number of factors, but the short summary is that AI can speed up iterative, or incremental coding projects.

At least two of the basic factors to account for include good, detailed prompting and skeptical verification of output.

By that I mean specifying exactly what the circumstances and context are in which you and the AI are going to work. Be clear about your own role (Project Manager, lead developer, neophyte, whatever) and the AI's role (expert coder, code monkey, whatever). Lay out your design specifications in advance.

And never accept the initial attempt of your AI to provide you with code. Validate it, test it, dissect it and make sure it does what you want and need it to do.

It can speed up the process in some cases as you off-load the actual writing of the code. It can not substitute for due diligence on your part.

 

[RaptorRaptur]

It depends on a number of factors, but the short summary is that AI can speed up iterative, or incremental coding projects.

At least two of the basic factors to account for include good, detailed prompting and skeptical verification of output.

By that I mean specifying exactly what the circumstances and context are in which you and the AI are going to work. Be clear about your own role (Project Manager, lead developer, neophyte, whatever) and the AI's role (expert coder, code monkey, whatever). Lay out your design specifications in advance.

And never accept the initial attempt of your AI to provide you with code. Validate it, test it, dissect it and make sure it does what you want and need it to do.

It can speed up the process in some cases as you off-load the actual writing of the code. It can not substitute for due diligence on your part.

Thank you George,
I have had a lot of fun playing with AI over the last few days. Now starts the real work. I will take your recommendations to heart and spend some time doing just that

Regards

Alan

 

[Imb]

Thank you George,
I have had a lot of fun playing with AI over the last few days. Now starts the real work. I will take your recommendations to heart and spend some time doing just that

Regards

Alan

I believe more in generalization and re-use of code than automatic generation of "duplicate" code.

Imb.

 

[GPGeorge]

Help me understand how that principle, with which I agree, applies to the use of AI to create new code solutions to new problems.

 

[CJ_London]

agree with other comments - I have noticed that unless you reinforce that you are using Access VBA, AI's can slip back into providing code intended for excel. A fair number of functions exist in excel vba but not access vba - latest one it tried to foist on me is the excel Max function, and the nz function does not exist in Excel vba because it does not have the concept of 'null'

 

[NauticalGent]

I've heard that you must continually remind Chatty, et al, of your coding standards. Also, you must insist that they are used consistently. Does that correspond to your experience?

I had the same experience until I started using the paid version. Its almost scary how tailored and familiar it has become...

 

[The_Doc_Man]

I have noticed that unless you reinforce that you are using Access VBA, AI's can slip back into providing code intended for excel.

I agree with CJ's comments including his comments subsequent to my quote. I will put my own spin on them. VBA is the same for everything in Office that uses VBA because they all use the same VBA library (VBA Code >> Tools >> References). What is different is the main program's libraries and objects that are part of the environment; those can/will differ from one Office member to the next. But there is where Chatty and the other AIs have the problem. If you don't specify the context, they might assume Excel when you meant Access because BOTH of those are valid users of VBA and are part of Office. When you make references to objects in VBA, you sometimes have to prefix / preface your actions to establish a context first - like DAO.Recordset vs. ADODB.Recordset, or using a WITH clause on a specific object to establish an object context. No different with Chatty. You have to assure it knows the context.

 

[CJ_London]

You still need to know your code - I wanted to find a way to make the detail onPaint event conditional to reduce flicker every time a control was clicked on or the detail section was scrolled. AI told me it can't be done and suggested alternatives such as conditional formatting, UDF, etc.

I found a way - onPaint now only fires when I need it to.

 

[GPGeorge]

You still need to know your code - I wanted to find a way to make the detail onPaint event conditional to reduce flicker every time a control was clicked on or the detail section was scrolled. AI told me it can't be done and suggested alternatives such as conditional formatting, UDF, etc.

I found a way - onPaint now only fires when I need it to.

I've had similar experiences with Claude incorrectly telling me something can, or can't be done. Hence the need for a cycle of prompting, testing and reprompting.

One example was a newer feature in the PowerApps environment that Claude hadn't yet encountered, the ability to execute Stored Procedures directly.

He wanted to use Power Automate, which works and which was the earlier implementation. He only relented when I pointed him to documentation from Microsoft on the new feature.

As someone I respect reminds me quite often, when working with assisted intelligence, one of you has to be the adult in the room.

 

[NauticalGent]

I've heard that you must continually remind Chatty, et al, of your coding standards. Also, you must insist that they are used consistently. Does that correspond to your experience?

Here is an example of what I was talking about. I was bored and asked "Chatty" a non-specific question. It helped wrote a PS script a few months back whose primary purpose is a FE Updater. I kept adding features and in the process, shared a few things about the application; the Workload Management Tool (WMT).

Truth be told, I am not 100% clear on what exactly it is suggesting, but what I am impressed with is it's ability to "remember" details from past conversations - conversations that were in different threads. It has even adapted "talking" to me in the same vernacular I use when talking to it:

I also asked it what "memories" it has on file that it uses when responding to me.

 

[AccessBlaster]

Even with the free model ChatGPT remembers quite a bit about our conversations. It's kind of creepy.

 

[NauticalGent]

Even with the free model ChatGPT remembers quite a bit about our conversations. It's kind of creepy.

And...you would like someone like me who HATES big brother would avoid it. I guess the convenience outweighs the hazard.

I have been assimilated...

 

[DaveMD]

From my experience, generative AI is a useful tool to assist with software development. If there is a well-defined contained feature or piece of functionality that you are working on developing, AI can be very useful. It can also help with brainstorming ideas. However, it does not replace the need for human expertise and critical thinking and evaluation.

I think the biggest weakness of AI in its current form is that it is overconfident to a fault. I have never had an AI indicate uncertainty in its responses even when there are clear inaccuracies. I think the technology companies are too focused on making their AI models appear “intelligent” but instead exhibit foolishness that is not apparent to an untrained eye. I think AI companies should focus on having their AI models make clear when there is uncertainty in the accuracy of a response. It is much worse to confidently provide nonsense than to communicate uncertainty and/or ask for more clarification or context before providing a response.

 

[RonPaii]

I just finished an AI session trying to solve a problem with a poorly documented API. It repeatedly gave me solutions using object properties that don't exist. It did finely admitted that it could not reference any examples that would solve the problem.

 

[Imb]

Help me understand how that principle, with which I agree, applies to the use of AI to create new code solutions to new problems.

In earlier times I worked as a chemist in the research of semiconductors. The static memories still had to be "invented".
When we had a problem or wanted to learn something, we did a "literature survey". Most of the findings were not relevant, and sometimes you said: "Hm, interesting". Not that this was the solution, but more a direction to investigate.
Today you use AI, and get in a few seconds tons of answers. But also here: most are not relevant, many wrong answers, and sometimes an "Hm, interesting".

Was this the solution? No, only a step in that direction.
Then the period of testing comes. Does the new step increases the problems? Then discard it, an try something else.
When it did not increase the existing problems, you probably can be in the right direction, and ready to take the next step. This continues until "the penny drops", and you found "a" solution for your problem.

So the development goes slowly, step by step (or sometimes by accident). That is the reason that big inventions or improvements are done done on several places on earth in about the same time.

So I see "the creation of new code solutions to new problems" as an evolutionary, "longlasting" path, where new code solutions are developped for old problems, together with the use old code for new problems, and all intermixed.

The advantage of AI is that your "literature survey" is almost spontanuous. But the critical mind of the developper must be even more developped, because the garbage content of AI-results is stupendous.

 

[Isaac]

Do you have any idea how good ChatGBT would be with VBA and/or the Access way of handling SQL?

Not off hand, but I'll guess it would be pretty good and easily enough tested.

 

[Isaac]

I've heard that you must continually remind Chatty, et al, of your coding standards. Also, you must insist that they are used consistently. Does that correspond to your experience?

No, once I've told it something in no uncertain terms and maybe even thrown in a phrase like "update your memory", it seems to stick pretty well.
I always use ChatGPT while logged in and I am on the $20/mo plan, and I make use of Projects too

 

[NauticalGent]

...and I make use of Projects too

Say more...