“…our cognitive abilities and creative capacities appear poised to take a nosedive into oblivion.”
Don’t sugarcoat it. Tell us how you really feel.
Don’t sugarcoat it. Tell us how you really feel.
Probably both are true: you should try them out and then use them where they are useful, not for everything.
None of my professional life reflects that whatsoever. When used well, LLMs are exceptional and putting out large amounts of code of sufficient quality. My peers have switched entire engineering departments to LLM-first development and are reporting that the whole org is moving 2x as fast even after they fired the 50% of devs who couldn't make the switch and didn't hire replacements.
If you think LLM coding is a fad, your head is in the sand.
I have no doubt that volumes of code are being generated and LGTM'd.
It used to take me days or even multiple sprints to complete large-scale infrastructure projects, largely because of having to repeatedly reference Terraform cloud provider docs for every step along the way.
Now I use Claude Code daily. I use an .md to describe what I want in as much detail as possible and with whatever idiosyncrasies or caveats I know are important from a career of doing this stuff, and then I go make coffee and come back to 99% working code (sometimes there are syntax errors due to provider / API updates).
I love learning, and I love coding. But I am hired to get things done, and to succeed (both personally and in my role, which is directly tied to our organization's security, compliance, and scalability) I can't spend two weeks on my pet projects for self-edification. I also have to worry about the million things that Claude CAN'T do for me yet, so whatever it can take off of my plate is priceless.
I say the same things to my non-tech friends: don't worry about it 'coming for your job' yet - just consider that your output and perceived worth as an employee could benefit greatly from it. If it comes down to two awesome people but one can produce even 2x the amount of work using AI, the choice is obvious.
For this kind of low stakes, easily verifiable task it’s hard to argue against using LLMs for me.
https://edition.cnn.com/2025/08/27/us/alaska-f-35-crash-acci...
But for it to be useful, you have to already know what you're doing. You need to tell it where to look. Review what it does carefully. Also, sometimes I find particular hairy bits of code need to be written completely by hand, so I can fully internalise the problem. Only once I've internalised hard parts of codebase can I effectively guide CC. Plus there's so many other things in my day-to-day where next token predictors are just not useful.
In short, its useful but no one's losing a job because it exists. Also, the idea of having non-experts manage software systems at any moderate and above level of complexity is still laughable.
Like any new tool that automates a human process, humans must still learn the manual process to understand the skill.
Students should still learn to write all their code manually and build things from the ground up before learning to use AI as an assistant.
personally I think everyone should shut up