And I would like to suggest that slowing the fuck down is the way to go. Give yourself time to think about what you’re actually building and why. Give yourself an opportunity to say, fuck no, we don’t need this. Set yourself limits on how much code you let the clanker generate per day, in line with your ability to actually review the code.
This might be the most consequential thing I have read all year.
I cannot recommend it enough.
I think it is so important for all the rational people to say, and keep saying, this. The sirens’ lull for me comes not from the agents themselves but the unpaid spokespeople on substack or the Readwise newsletter, etc. saying “Here are my free premium prompts, my 5 part formula and github repository, get Cowork running your life by 9am tomorrow!” Sure, I realize they’re probably bullshit, but….what if??! 
And yet, I see firsthand the absolute tangled mess that occurs even at the document writing level when someone has mindlessly initiated a project with an agent and then changed the scope midstream before tossing it over the fence to me with a “TA-DA!” And that’s just 20 indecipherable pages of a Word doc that no human has built a mental model of, let alone thousands of lines of code 
You can sleep well knowing that you still have an idea what the fuck is going on, and that you have agency.
What worries me is that I think many people are now more concerned with sleeping well knowing they’re not being left behind in the AI race. To where or for what? They’re not sure.
Or you just need a rubber duck to bounce ideas against, which basically means bouncing your idea against the compressed wisdom of the internet and synthetic training data.
This is my current use case, and in the scheme of my work’s scale, already a pretty miraculous one at that. I try to remind myself not to be greedy beyond this huge gift that agents already provide and acknowledge the ways it has transformed my work for the better.
I can’t wait for opinions such as these to become mainstream, rather than merely the province of Old People Yelling At Clouds. I don’t want to yell at clouds, but I also don’t think I’m wrong.
“Everything is broken” is not a new problem. I was screaming this in 1998 and others were screaming it in 1974. The difference lies in how easily we now make our entire lives depend on all this brokenness. We seem to have seen Terry Gilliam’s “Brazil” as a blueprint, rather than a cautionary tale. (Well, that’s seems only reasonable, since we are doing the same with “The Handmaid’s Tale”. The novel, obv.)
We have basically given up all discipline and agency for a sort of addiction, where your highest goal is to produce the largest amount of code in the shortest amount of time. Consequences be damned.
When we play with Parent’s credit card, it’s fine to generate and ship bullshit, but suddenly, when you actually have to pay for it, you might actually decide that you care about what you’re shipping and you might, you know, make it work. Once again, this is not a new problem, but evidently The Tech Sector still has waaaaaaay too much cash to burn. I could burn it for you, and I would do it twice as hard for half as much.
You realize you can no longer trust the codebase. Worse, you realize that the gazillions of unit, snapshot, and e2e tests you had your clankers write are equally untrustworthy. The only thing that’s still a reliable measure of “does this work” is manually testing the product. Congrats, you fucked yourself (and your company).
I laughed, because I literally wrote about this myself today: it’s funny how They don’t trust your tests because They weren’t around when you wrote them, but They trust Their tests, even when Their tests are just as complicated as yours. Writing the code is a shortcut to understanding it, in part because you are forced to slow down long enough actually to read it. Writing it creates an intentional bottleneck that cultivates the context in which understanding can happen. Clankers eliminate that. It’s the Evil Wizard Problem (from The Pragmatic Programmer) on crystal meth.
The point is: let the agent do the boring stuff, the stuff that won’t teach you anything new, or try out different things you’d otherwise not have time for.
I’m so thrilled to see someone else write this. There are problems I want solved and problems I want to learn to solve. I want the clankers to deal with the first kind so that I can focus on the second. It’s really that simple.