I think most people react most strongly to the lack of quality in purely AI generated things. Especially art projects. No one likes buggy games and AI doesn’t change that. If they can manage to be one of the few to pull off a polished and successful game entirely generated by AI the I’d wager the actual backlash will be far more limited. But most people have now come across at least a few generated slop projects and they tend to leave a bad taste. So resistance to supporting the decision to go all in on something that has generally provided distasteful outcomes is to be expected, no?
I have serious and significant objections to how LLM codegen models have and are being created/trained, marketed, deployed, and used, for a whole litany of ethical, personal, and professional reasons.
To my deep chagrin, I and my team have effectively been forced to start using those tools to expedite some of our work lately, largely due to the very fucky and understaffed state my team has been in for the better part of a year. So, yeah, I do use claude, and I am fully versed in how to get higher-quality output from it by including adversarial and specialist agents in workflows, and how to present goals in a way that you minimize nonsense and maximize targeted, incisive agentic changesets. But Ialwaystreat that agentic output as a draft, and wherever possible, give my workflows a manually, tightly coded baseline example to serve as a change template. I categorically refuse to yeet LLM output into a PR - in fact, there was a multi-month period earlier this year where a bunch of contractors were doing just that, and I consistently took the piss out of them - and after enough restatements to make it clear they weren’t listening and didn’t care, we fired three of them.
There is a reasonable and incisive way to employ codegen tools, but you absolutely have to know what you’re doing, and you have to be conscientious about it.
Fully cuck coding is generally a recipe for disaster, including horrifying security vulnerabilities, awful performance, and various exciting categories of bugs that can yield interesting prod issues and outages. It will tank your code quality if you let it - and worst of all, it will make you a worse engineer if you intellectually lean on it too much. Delegation of your critical reasoning and technical aptitude to the clankers is a great way to allow those skills to atrophy to an alarming degree, and on top of that, you will have a much shakier understanding of the systems you are building and supporting.
I think most people react most strongly to the lack of quality in purely AI generated things. Especially art projects. No one likes buggy games and AI doesn’t change that. If they can manage to be one of the few to pull off a polished and successful game entirely generated by AI the I’d wager the actual backlash will be far more limited. But most people have now come across at least a few generated slop projects and they tend to leave a bad taste. So resistance to supporting the decision to go all in on something that has generally provided distasteful outcomes is to be expected, no?
Precisely.
I have serious and significant objections to how LLM codegen models have and are being created/trained, marketed, deployed, and used, for a whole litany of ethical, personal, and professional reasons.
To my deep chagrin, I and my team have effectively been forced to start using those tools to expedite some of our work lately, largely due to the very fucky and understaffed state my team has been in for the better part of a year. So, yeah, I do use claude, and I am fully versed in how to get higher-quality output from it by including adversarial and specialist agents in workflows, and how to present goals in a way that you minimize nonsense and maximize targeted, incisive agentic changesets. But I always treat that agentic output as a draft, and wherever possible, give my workflows a manually, tightly coded baseline example to serve as a change template. I categorically refuse to yeet LLM output into a PR - in fact, there was a multi-month period earlier this year where a bunch of contractors were doing just that, and I consistently took the piss out of them - and after enough restatements to make it clear they weren’t listening and didn’t care, we fired three of them.
There is a reasonable and incisive way to employ codegen tools, but you absolutely have to know what you’re doing, and you have to be conscientious about it.
Fully cuck coding is generally a recipe for disaster, including horrifying security vulnerabilities, awful performance, and various exciting categories of bugs that can yield interesting prod issues and outages. It will tank your code quality if you let it - and worst of all, it will make you a worse engineer if you intellectually lean on it too much. Delegation of your critical reasoning and technical aptitude to the clankers is a great way to allow those skills to atrophy to an alarming degree, and on top of that, you will have a much shakier understanding of the systems you are building and supporting.