So LLMs are the 3d printers of software. Great for niche use-cases without enough market demand for a proper solution, and generally scale very poorly vs proper industrial processes.
The big difference from 3D printers is that there is zero upfront investment required. The number of non-technical people I've seen making simple tools/scripts to automate bits of their workflow is astonishing. Costs them nothing to at least try it out.
> The number of non-technical people I've seen making simple tools/scripts to automate bits of their workflow is astonishing.
The overwhelming majority of the population doesn’t even know what a tool or script is. Of the remaining who do, I would not be shocked that they’re capable of asking an LLM to produce one for them.
>> The overwhelming majority of the population doesn’t even know what a tool or script is.
They don't need to know what it is. "Hey ChatGPT I have to check this report daily and update a spreadsheet with the latest changes, can you do it for me?"
Can you come up with a better example? I ask ChatGPT your question verbatim:
> Hey ChatGPT I have to check this report daily and update a spreadsheet with the latest changes, can you do it for me?
And it recommends:
- process automation using a script (I don’t know what that is)
- parse the report (what’s parsing?)
- update the spreadsheet automatically (cool, how?)
- highlight differences (no I want it done automatically)
Obviously I want it done automatically, so I tell it it’s excel and its next response tells me to use “pip install pandas”. I close the browser and go back to work.
So yeah, trivial for you and me, but you can’t act like we’re in magic wizard territory.
Try not taking things so literally. Here's a better example:
- Company has a Cursor project you can download and open in Cursor.
- It's connected to all of the companies internal data sources.
- Now business people (who would have previously had to make requests to the BI team) can query the databases in natural language.
- They can tell Cursor to export the data as a CSV and they can open in their Spreadsheet tool.
- They need that report every day. They can tell Cursor and it can suggest the best automated way to do that. Once they ok it, Cursor can write a script and run it daily. The csv file shows up in the persons Downloads folder every morning.
Aside from this, I've seen people with no coding background creating Chrome extensions to help them with things. I've seen them making little React Native apps without knowing what React Native is.
The preponderance of evidence to date would point to systems engineered by meatbags are of a significantly higher quality than vibe coded ones.
My use of the word “system” is very intentional vs. something less qualifying like “program”.
Edit to add: my take is proper system engineering and design basically requires general intelligence, so the bar for LLMs to reliably produce high quality systems is AGI.
This seems to be a non-sequitur, did you respond to the wrong comment? And investors are famously stupid on aggregate, so I wouldn’t trust their judgment in terms of predicting the future of systems engineering.