The article <a href="https://mikelovesrobots.substack.com/p/wheres-the-shovelware-why-ai-coding" source="" author="Mike Judge">Where's the Shovelware? Why AI Coding Claims Don't Add Up</a> is an interesting read that makes the following argument, more or less, <bq>If so many developers are so extraordinarily productive using these tools, where is the flood of shovelware? <b>We should be seeing apps of all shapes and sizes, video games, new websites, mobile apps, software-as-a-service apps — we should be drowning in choice. We should be in the middle of an indie software revolution.</b> We should be seeing 10,000 Tetris clones on Steam.</bq> As bad as the world of existing non-LLM-generated slop already is, the author's point is that it's not <i>gotten measurably worse.</i> If AI allowed pretty much anyone to build an app---the proposal buoying the AI bubble---then we'd be <i>flooded</i> with a <i>tsunami</i> of crapware rather than just <i>drowning</i> in a <i>ocean</i> of it. <bq>[...] billions of dollars have been invested in these tools. <b>Billions of dollars will continue to be invested in these tools.</b> The problem is that <b>they’re being sold and decisions are being made about them</b> — which affect real people’s lives — <b>as if they work today.</b> Don’t parrot that nonsense to me that it’s a work in progress. It’s September 2025, and we’ve had these tools for years now, and they still suck. Someday, maybe they won’t suck, but <b>we'd better see objective proof of them having an impact on actually shipping things on the large.</b></bq> From the <a href="https://old.reddit.com/r/programming/comments/1n7vpvi/wheres_the_shovelware_why_ai_coding_claims_dont/">comments on Reddit</a>: <bq>Today (actually not joking) a manager told me<bq>AI should make you 10x more productive, what takes you 10 days should take you 1.</bq>Which I figured was bullshit because Tuesday he asked<bq>Can we compile OpenSSL v3.6 for RHEL-5? Docker makes this easy right?</bq><b>IDK how AI makes me 10x more productive when I spent 4 hours in meetings to realize we actually needed to update our LuaJIT (on RHEL-10) not compile a version of OpenSSL</b> (???)</bq> <img attachment="streetlight_effect.webp" align="right" caption="Streetlight effect">This is a much better point to consider. People are <a href="https://en.wikipedia.org/wiki/Streetlight_effect">searching for their keys on the sidewalk under the streetlamp</a> when they lost them in the bushes. Getting people to address inefficiencies in priority order would be a much bigger lever than letting them take the easy way out by bike-shedding with AI or by trying to force people to <i>USE AI DAMMIT</i> to run in the wrong direction. What's the point of doing something faster when it doesn't need to be done in the first place?