AI producing excessive amounts of code is a self-reinforcing phenomenon wherein you have to write huge diffs to get anything done due to the way the existing code is implemented
1 Boost
0 Likes
@datarama @jcoglan path to a utopian society straightforward from here:
- all software is given markdown documentation and MCP backends, so that it can work with AI
- AI disappears due to collapse of economic logic and/or changes in IP law
- someone invents a programming language that can connect to MCP
This post is a joke, I think
@mcc @datarama @jcoglan MCP is just JSON-RPC with built-in documentation. Writing an MCP client library wouldn't be hard; no doubt many already exist.
The issue, of course, is that MCPs don't "need" any stability guarantees because "the LLM will figure it out!," so programs that depend on an MCP will probably break frequently.
@mcc @datarama @jcoglan In general I have been trying to survive through this era by, just improving our systems normally (lower latency, better APIs, better error messages, invalid states unrepresentable, DRY...). And claiming that the improvements make it work better with AI. Which for all I know could be true
Collapse the last two into "the AI corporate ecosystem no longer exists in its mid-2026 form and we, as a society, develop better mental models for what LLM-based instruction following systems are and (more importantly) aren't" and your joke becomes my actual prediction for the future.
@mcc I know this sounds like a fake answer but when people talk about it eliminating boilerplate my response it "I just don't write boilerplate and I don't understand why this is a problem" like genuinely this is not a pressing issue that I have when writing programs
@mcc like, the information density of code you write is directly under your own control. literally "you control the buttons you press"
@jcoglan @mcc I think that is generally true for a surprisingly small subset of programmers who are likely to be over-represented on Fedi. If you're writing business logic, which a whoooole lot of programmers are, usually you're stuck in languages and templating systems beyond your control, and there is a ton of boilerplate.
@vathpela @jcoglan so "AI" is a broken corporate system that refuses to fix itself giving itself permission to fix itself but ONLY if it does it through the jingly-keys "AI" technology. seems silly. what if instead we introduced macro systems and learning to program, and called it "supertechnology" or "hyperprogramming" and said it was worth $2 trillion dollars. would the corporation mandating the boilerplatey template system be willing to embrace a less-boilerplatey template system then
@mcc @vathpela @jcoglan the only flaw in this plan is that this technology would involve a well-compensated labor force whose psychological well-being was respected and cultivated, and the real purpose of “AI” is to destroy that. if they could destroy the power of labor without all the “boilerplate” stuff they’d be doing that instead
@glyph @vathpela @jcoglan perhaps at some point in the future, corporations will conclude that their purpose is to perform activities that result in a revenue higher than their expenses, rather than treating psychological terrorism against working classes as their primary objective.
unlikely, i know. but perhaps it is possible.
@mmcknett @glyph @vathpela @jcoglan problem is investors don't want profit they want *growth*
and investors have learned growth is exponential so they're ok with killing four businesses if the fifth is "growing". They think the one growing business will grow faster than the other businesses (the ones tearing themselves apart chasing "growth") fail
@mcc @mmcknett @vathpela @jcoglan for what it's worth this is a modern brain disease of "investors". I am investor. I don't mind growth but it is explicitly not what I am looking for. I would rather invest in dividends than buybacks. There *is* a tax advantage to buybacks which needs to be eliminated, so bits of this are structural, but much bigger chunks are in fact cultural or even spiritual
@mcc @mmcknett @vathpela @jcoglan I think we are going to see this culture shift, in the wake of the AI bubble burst. Venture capital was a culture that made sense in the late 90s, the fiber bubble notwithstanding; there were a bunch of genuinely transformative opportunities to build things. Investing in 5 risky startups when only 1 would succeed was a rational thing to do, and not particularly evil. That metastasized into private equity, where you would buy 5 going concerns and destroy 4
@jcoglan boilerplate is a failure of abstraction. some languages give you enough abstraction to prevent boilerplate; some make this hard. if you find you don't encounter boilerplate, this suggests to me you're good at abstraction and your tools are allowing you to be good at it.
if people keep encountering boilerplate, my rec would be to adopt tooling that allows them to do abstraction. but i guess some people think adopting and learning tooling is too hard and letting an RNG write code is easy
also the anti-DRY-ness it tends to produce makes it incredibly difficult to know if you've covered everything that needs changing b/c the same concept has dozens of representations. this directly leads to more bugs