@ReaperGorgo@yashmp2004 For real, people love to compare poorly prompted agents with imaginary perfect devs who create perfect documentation, thorough test cases, and never create tech debt.
@SuiBuilds We've replaced MVP launch tech debt with invisible tech debt. With the rate at which code gen tools are improving, do you think future tech debt with AI might be a good tradeoff?
@lucas_builds I use them to keep the context windows separate and not blow up my main session. That way, the subagents can do tasks and then only return what is relevant to the main agent loop
How I choose between traditional automation and AI:
If you need it done the EXACT same way, skip AI. If x then y, otherwise z.
If you need something that doesn't have such clear definitions, that's where AI shines. "Should this be marked as a marketing or product expense?"
@im_pranavkakde I agree it's not software engineering. It's project management. Of those 3, I would expect vibe coders and PMs alike to only own number 2.
@kapilansh_twt When I use those 3, it's not just the LLM but the entire harness that makes them so effective. I don't want to build out my own local functionality for each use case if I can use an existing tool that performs better and faster.
@gitghxst Twitter/X: I built an AI that built another AI, that made a harness so powerful that even small AI models can outperform flagship models
Real life: Day 271 - still trying to get IT to approve databricks
@Cristina_AC_X If you take a regular data engineer or analyst and put them in that same position, they will either have to check the codebase to see where and how a field comes from, or check with stakeholders to agree on a definition.
@Cristina_AC_X I half agree...maybe? It's spectacular at SQL, and it sounds like what you're describing is the lack of data governance and a data dictionary. And to be fair to our future robot overlords, a human couldn't solve that either without looking elsewhere for additional context.