Is agentic AI making you procrastinate?

Eratticus · vor 2 Stunden

I definitely feel this. My company has mandated LLM tools to write code while at the same time prohibiting allowlists on commands for any reason so every decision needs my manual approval. So much of my time is now spent babysitting Claude. I have seen it make mistakes and delete things as we all know. But needing to approve an `ls` command is such a time sink in a completely different way than anything before LLMs.

GriffinMakesThings · vor 1 Stunde

The comments in here are incredibly depressing. So many people have clearly just given up. Are ya'll really happy to just sit and lose your ability to write software while you watch LLMs vomit out code? No way people using agents this way are actually reviewing everything, that's just impossible.

This shit's dark.

jbcaprell · vor 1 Stunde

There’s a line in the Ada 83 Language Reference Manual that goes:

The need for languages that promote reliability and simplify maintenance is well established. Hence emphasis was placed on program readability over ease of writing.

… and has been similarly expressed in things like Clean Code by Robert C. Martin as:

Indeed, the ratio of time spent reading versus writing is well over 10 to 1. We are constantly reading old code as part of the effort to write new code. ...[Therefore,] making it easy to read makes it easier to write.

… and this sort of thing has always been true. In an environment where LLMs have dramatically reduced the burden of writing code (of various quality), that ratio only grows: which means that when-and-where your work looks more like supervision, you really have to foreground directing agents to never give you big, sprawling commits that are utterly unreviewable. You need explicitly direct the agent to tell you a story the way you that a good senior would direct a junior developer to break up a 5K LOC pull request, because you’re on the hook for maintaining what been written, and the LLM will not at-all prioritize review-ability unless you direct it to do so.

dcabines · vor 1 Stunde

Stop having it write features for you. Ask it how it would write a feature then type it out yourself.

You’re robbing yourself of actual on the job learning and growth by babysitting the agent while mindlessly scrolling on your phone. Don’t let it steal your education or your critical thinking skills so easily.

C0c04l4 · vor 1 Stunde

I play chess.

destel116 · vor 1 Stunde

Same thing. And if I work on several projects at the same time this way, I also end up in the situation when most sessions wait for my input for too long.

digitalbananax · vor 1 Stunde

Crank some tunes on the good ol' transistor radio, maybe cook a steak, mow the lawn, yeeeah booi

sateliteconstelation · vor 1 Stunde

A few months back I started making poly-clay sculptures to fill this gap. I have over 200 now.

Also, and on a brighter productive note. I’m starting to get the hang of working on 3-4 tickets at a time, so there’s always one that needs my attention.

taotau · vor 1 Stunde

Which YouTuber put out a video about this topic? There's been this exact question quite a few times today.

lucasg115 · vor 1 Stunde

I've got ADHD, so when I'm in a flow state I'll just have two or three agents coding at the same time, often even for different projects. Then I just bounce between them for approvals and follow-up prompts.

abeuscher · vor 1 Stunde

You do not have to fill every moment of your workday with activity. If you have a process running you get to wait for it to end that is not procrastination nor is it lazy. If your employer is monitoring you so closely that you feel like you can't just relax while your machine is working then that's a culture problem.

We are thought workers. Brains are like muscles and need rest and breaks. Brains have a fixed capacity and when you go over it your brain does not instantly recover. Be kind to your brain. Give it space to breath. Look away from a screen. Bring a book. Better yet walk away and outside if you can set an alarm to pull you back in when it is time to review the work from the prompt.

And listen - if you get shit for it, then make sure whoever gives you that shit is working every single minute of every single fucking day and make sure they know you're watching them always. Force the anxiety back on them until they break. Because bosses can be made to burn out - often more easily than plebes. And forcing you to work when you're already busy and waiting is a great way to make people crazy.

Sapn1s · vor 1 Stunde

Not that this addresses you main question, but
"asks for permissions" you can use skip permissions guard, depends on what harness you use but like claude code has already built-in protection to not wipe your drive. It can still happen though, so ideally you setup the projects to run in docker containers or anything else , whatever is your pick of damage management, maybe u can just backup all your files, whatever you choose, but there are options to avoid perma permission yes accepting, if you plan to use it a lot, id invest time solving it first to save time/effort long term.

bunzelburner · vor 1 Stunde

maybe I'll get shit on for this but I have started to use it to do tasks that have a clear plan. for example, I had dashboard pages set up for 2 resource types. I needed a dashboard page for a third resource type. So essentially the template was there from the first 2 including the API endpoint shapes. I told Claude where to look, had it develop a plan, and had it go. Then I worked on something else. Came back later, tested and checked. Now I have the base dashboard page for that resource ready to go and I could just add the necessary customizations for that page on top of it. Saved me probably and hour or two in that instance. I still understood how it all worked because I basically wrote the templates it was using, and I checked over the plan before letting it go to make sure it wasn't getting cheeky on me.

darkn1 · vor 49 Minuten

I've taken to improving the "softer" side of the nature of our work. Design, layout, readability, accessibility. Those sorts of things, rather than just raw code or engineering effort.

But at the same time yeah I feel what everyone is saying

Electro_Nick_s · vor 42 Minuten

I work in security so I'm often juggling multiple projects but I have in all my agents.mds for my tools to split off into it's own worktree whenever editing files. So in the middle of the work day I might have two to four sessions going depending on how busy I am. At that point there is no down time. I would recommend 'avoid-ai-writing' as a skill as well. I also use openspec and build code to spec which prevents some of that drift

subnu · vor 41 Minuten

Procrastination culture is a massive thing in software development, even prior to AI. It was just easier to fall victim to when things took 15x as long to create.

Turbulent_Spare6385 · vor 35 Minuten

The problem I have with agentic AI is that the further you diverge from writing the code yourself, the harder it becomes to stay in sync with the codebase. We end up tailoring our workflows with loops of writing, verification, and rewriting just to build enough trust in the agent’s output. I can see the productivity benefits, but I’m not sure I like this. I actually enjoy writing code, or at least used to, since I’m not writing anymore

pierec · vor 26 Minuten

Try to setup your workflow in a way that doesn't require you as a semi-passive copilot. You scope out the work and come back for results. This may require some legwork first to get your project in an agent-friendly state.

I work out when agents are doing longer stretches of work.

I've setup a small corner with kettlebells and a yoga mat in my office some months back. My posture, energy and focus improved quite a bit. Now I have the drive to learn new stuff again.

Use the extra time to improve yourself, starting with your health. I'm considering the current agentic coding revolution a temporary boon to one's free time. The project management / planning side of things hasn't caught up yet, so engineers leveraging agents end up having less to do in my experience. Take advantage of that while it lasts. Work on yourself, use the time to poke around what other teams are doing, maybe look for opportunities to contribute to some org-wide initiatives.

Mediocre-Writer6635 · vor 14 Minuten

I read code it previously wrote while it's doing the next thing, and write down prompts to fix or explain whatever I find that i don't agree with.

I also do manual testing.

Yes the AI is waiting for me often, but that's just reality of how long it takes to sign off on code that is being committed under my name.

cmaxim · vor 1 Stunde

Yeah.. in the beginning I was feeling like I would be reviewing all the code that AI spits out, and then direct each granular choice where needed to keep things moving forward.

I'm finding the reality is that AI is just much faster and more efficient at reviewing and producing code than I am. While I still think it's important to regularly review what it's outputting I often just can't keep up on larger projects, and I end up using AI to help review the AI, otherwise I start to lag and slow down which ends up defeating the partial purpose of AI (increasing productivity).

Asura24 · vor 1 Stunde

I haven’t written 1 line of code in the past months, no realistic reason for you to do it either if you are using the best models out there.
Also I would recommend you to always run things in auto most of the time it will do things right, you shouldn’t be wasting time for day to day stuff anymore just checkout the final PR

Chemical-Ad-7982 · vor 1 Stunde

"the AI already prompts you for an approval or asks you what's next." -> This is your mistake, you need to set it up to work autonomously. That means enabling auto permissions, forming a plan with it through discussion, and then letting it go on that plan. When its running you start a new one to run it parallel and end up with 4-5 sessions running concurrently (if your agents work in the same codebase you have them work in separate worktrees). In this way you spend all your time working and 0% waiting. Basically you are correct that your waiting window is too small, thats why you need to make it bigger.

"Yes datacenters suck. Yes AI is totally hyped." -> ??? Complete reddit delusion AI works great for programming.

Du schreibst:	Du siehst:
Kursiv	Kursiv
Fett	Fett
[reddit!](https://reddit.com)	reddit!
* Aufzählung 1 * Aufzählung 2 * Aufzählung 3	Aufzählung 1 Aufzählung 2 Aufzählung 3
> Zitierter Text	Zitierter Text
Lines starting with four spaces are treated like code: if 1 * 2 < 3: print "hello, world!"	Lines starting with four spaces are treated like code: if 1 * 2 < 3: print "hello, world!"
~~strikethrough~~	~~strikethrough~~
super^script	super^script

webdev

Posting Guidelines

Related Subreddits

Discords

MODERATOREN