Code Building Code

  • As a DBA and developer, I personally don't need an AI to help me write my code. However, I could really use an AI to monitor my servers, maybe even tweak the resources and fix minor issues automatically, so I can then focus more of my undivided attention on writing code.

    "Do not seek to follow in the footsteps of the wise. Instead, seek what they sought." - Matsuo Basho

  • I think AI driven dev is one of those marketing hype jingles going along with the often unrealistic promise of AI.  "just tell it what you want and it will figure out what to do".  Again a dramatically oversimplified concept, but still I suspect the pressure will come from elsewhere (probably not from within IT).

    The thing to be careful about is that we have an intelligent, articulated, argument as to why it's not a feasible concept.  From the accounting point of view, we tend to be viewed as "expensive", so "don't need" isn't going to save a job.  "doesn't work" or "can't be done" will, but that will require a lot more articulation.

    ----------------------------------------------------------------------------------
    Your lack of planning does not constitute an emergency on my part...unless you're my manager...or a director and above...or a really loud-spoken end-user..All right - what was my emergency again?

  • Tom Uellner wrote:

    Until product owners are willing to put enough detail into requirements documents, we're not going to see AI build complete programs. I know that I routinely round trip back to product owners 4 - 5 ( 15 - 20 🙂 ) times to get enough clarity to build what they really want instead of what they said they wanted the first time. It's a very interactive process.

    That said, when AI gets to the point of being close to human intelligence, it could start interactively working with a product owner the same way we do. It needs to be an iterative process. That's not going to happen in my lifetime but there's no reason it could not eventually happen.

    And yes, I use code to build code quite often.

    Why would this be far off? You could do this now for UI testing, probably, assuming that you had the feedback mechanism designed in a way that made it make sense to the AI, but it would essentially be the A/B testing that your optometrist does when you get a prescription for glasses. This one better? or this? rinse and repeat. gradually the AI would learn the user preferences and tend toward those first.

    -------------------------------------------------------------------------------------------------------------------------------------
    Please follow Best Practices For Posting On Forums to receive quicker and higher quality responses

  • jonathan.crawford wrote:

    Tom Uellner wrote:

    Until product owners are willing to put enough detail into requirements documents, we're not going to see AI build complete programs. I know that I routinely round trip back to product owners 4 - 5 ( 15 - 20 🙂 ) times to get enough clarity to build what they really want instead of what they said they wanted the first time. It's a very interactive process. That said, when AI gets to the point of being close to human intelligence, it could start interactively working with a product owner the same way we do. It needs to be an iterative process. That's not going to happen in my lifetime but there's no reason it could not eventually happen. And yes, I use code to build code quite often.

    Why would this be far off? You could do this now for UI testing, probably, assuming that you had the feedback mechanism designed in a way that made it make sense to the AI, but it would essentially be the A/B testing that your optometrist does when you get a prescription for glasses. This one better? or this? rinse and repeat. gradually the AI would learn the user preferences and tend toward those first.

    I think AI has to come a lot further than it's current state to be able to prod a product owner in the same way we do when we're looking for logical pieces that they are overlooking. It's not anywhere near that state yet. It will be at some point but not now or in the next several years.

    I just need to make it to retirement age before that happens... 🙂

  • Good point. User demand is a bigger issue than what is often thought. Remember English Query?

     

    Sue

     

  • I'm quite happy about using code to generate code,  and first started doing that in about 1970.  But I'm not convinced that an AI will help very much in doing that - at least until our AI capablities are a lot better than they currently are.  If AI was viable, Google could have combined it with their statistical interpretation model and vastly improved the quality of text produced by Google Translate, perhaps  by understanding enough grammar to recognise word classes and using it to distinguish between, for example,  "pants (verb)" and "pants (noun)" (but either they haven't tried that or they have and they couldn't get it to work) or by some other mechanism.  But they haven't yet got  decently accurate translator - any serious human translator will tell you that Google translate makes a mess a lot of the time.   I's better than in was a decade or so ago, but it's still not good.   Just try it on, say,  translating between a Celtic language (eg Scottish Gaelic) and a Germanic language (eg English) and you'll see what a mess it makes.  If Google can't make a decent translation when it has a corpus of hudreds of thousands of text translations between the two languages to work from, and AI is not a solution, I wouldn't imagine that AI-generated code could actually be useful with the current state of the AI art.

    • This reply was modified 4 years, 10 months ago by  TomThomson.

    Tom

  • Back in the day (1981) I developed a code generator for the place where I worked, for a mainframe online app.

    Simple in some ways but very helpful, it would modify a template to produce a runnable program to present a screen of data to a user. A programmer still had to code moving data to and from the DB, but inter-screen navigation, some standard calculations, and error handling were all in the template.

    Depending on complexity of data handling, the template provided 80% to 90% of the final program, with a corresponding beneficial impact on overall development and testing time and quality. I used this technique at a number of jobs over the years, until I moved into DB management.

    When I bother to think about it, I am still mystified why similar techniques are not standard today - I never met a Windows developer who thought providing a standard framework tailored to the site's requirements might be useful to them.

    BTW, it was in 1981 I first used a Wiki. The code for it had been built a few years previously by the lead Systems Programmer at that site. Obviously all mainframe-based, but you could select Add Page or Update Page, enter your desired text, set hyperlinks to other pages, and select Publish, so definitely it was a Wiki.

    Original author: https://github.com/SQL-FineBuild/Common/wiki/ 1-click install and best practice configuration of SQL Server 2019, 2017 2016, 2014, 2012, 2008 R2, 2008 and 2005.

    When I give food to the poor they call me a saint. When I ask why they are poor they call me a communist - Archbishop Hélder Câmara

  • Removed, duplicate post.

    • This reply was modified 7 months, 2 weeks ago by  EdVassie.

    Original author: https://github.com/SQL-FineBuild/Common/wiki/ 1-click install and best practice configuration of SQL Server 2019, 2017 2016, 2014, 2012, 2008 R2, 2008 and 2005.

    When I give food to the poor they call me a saint. When I ask why they are poor they call me a communist - Archbishop Hélder Câmara

  • So how do you know when AI is 'smart enough' to be safe?  Does it cover all the possibilities?  And most importantly, how safe are you from liability if it should fail?  Are you going to be happy with the decision?

    Years ago I was disgnosed with prostate cancer.  The science at the time said watch it and see what develops.  I said do the surgery now.  I'm still alive.

    I have a friend recently diagnosed with bladder cancer.  He can have the surgery now with an external bag installed, or he can wait to see how fast it progresses.

    Ever wonder how many research studies aren't published because the researchers don't like the results?

    AI is just another roll of the dice.

    • This reply was modified 7 months, 2 weeks ago by  skeleton567.

    Rick
    Disaster Recovery = Backup ( Backup ( Your Backup ) )

  • I think where AI helps is as a 2nd pair of eyes when you are coding.  It will question and suggest things that a person would be too embarrassed to ask.  Sometimes the embarrassing questions are precisely the questions that need asking.

    It is not subject to intimidation by developers with an aggressive personality.

    I think I have mentioned before that I use Sourcery.ai when coding Python.  I have found that it is a useful teacher.  In some cases I don't understand what it has recommended so I go and research its recommendations.  I don't accept them unless I understand them.

    Over time its recommendations get fewer because I've learned from past interactions.  Then the next generation of Python comes along and a whole raft of new suggestions appear.  Its handy because otherwise I'd be coding Python 3.12 in the same way as Python 3.6.

    As a declarative language SQL doesn't have quite the same needs as a imperative language.  The problem I see is in a non-SQL person summoning up the enthusiasm to build a useful SQL AI tool or a SQL person developing the skills to be able to do so.

    I honestly believe that AI can be a great benefit to us as DBAs, data engineers and developers.

Viewing 10 posts - 16 through 24 (of 24 total)

You must be logged in to reply to this topic. Login to reply