I get that LLMs can be modestly useful coding assistants. But this whole ecosystem of "agents" and harnesses and subsidized million dollar spends at pennies on the dollar, I just don't get. What are you building? What useful work is being done? Because all I see is worse software being built at higher prices, an astronomical technical debt, and a thriving ecosystem of charlatans cashing in before the inevitable bust. Now I'm just an inebriated spud shaking my fist at clouds here, but I am also honestly curious what nuggets of reality are to be gleaned from the hype. What tools and workflows are people finding actually valuable? (Obligatory acknowledgment of lampshade: deliberately provocative title.)
Comments
com•23m ago
I see well-described bugs that have been backlogged for 4 years out into the hopper of a simple harness-based bug fixer - and a tested, ready for review fix pop out USD2.40 later. This is game-changing for us and our customers.
I’m seeing quite decent data analysis work done by otherwise busy people that they’d never otherwise have the time to do, supported by readable python code showing how the stats were generated, and allowing us to run actual experiments and rerun the stats to see if the analysis and suggested actions has a meaningful impact.
I don’t know about whether the trillions and trillions of dollar people are barking up the wrong tree, but this stuff I’m talking about is cheap and easy and we see the results.
com•23m ago
I’m seeing quite decent data analysis work done by otherwise busy people that they’d never otherwise have the time to do, supported by readable python code showing how the stats were generated, and allowing us to run actual experiments and rerun the stats to see if the analysis and suggested actions has a meaningful impact.
I don’t know about whether the trillions and trillions of dollar people are barking up the wrong tree, but this stuff I’m talking about is cheap and easy and we see the results.