How do I handle a vibecoding manager

[-]

blckshdw@reddit

Just close the PR. Don’t merge it, just close it. From the sounds of it he won’t know the difference and can think he did something

[-]

UUS3RRNA4ME3@reddit

Look this isn't strictly related to your point, but I can't stand self fulfilling tests.

They're usually more obfuscated than assert 3 == 3, but it'll be something like:

Mock mock = mock() mock.return=3

Something.method(mock())

Assert method.called_with(3)

Totally pointless

[-]

mpanase@reddit

40% of the last project I inherited were tests

Thousands of tests

EVERYTHING, including every module in the same codebase, mocked

I change code and functionalities again and again... no test breaks

And of course, tons of 2-4 year old bugs. Including crucial functionality that clients were cancelling their subscription for.

Impossible to reason with the testers... I had to simply end up telling the lead tester "fuck no, because I'm the project lead and this has proven to be useless; we are not doing it this way anymore"

[-]

Repulsive-Hurry8172@reddit

I change code and functionalities again and again... no test breaks

This is the problem. Inheriting a codebase with 40% tests is a good thing, but if the test code is buggy, then that's a problem

[-]

Kilazur@reddit

I haven't encountered those I think; this pseudo code does not match your description, right?

IMyLogic myLogic = new Mock<IMyLogic>();

var callToMock = CallTo(() => myLogic.MyMethod(3));

var testClass = MyClass(myLogic);

testClass.DoStuffWithLogic(3);

Assert callToLogic.Happened();

Despite a few years under my belt, my last job is the only one I've done unit testing at.

[-]

Yes, these kind of tests are typically written with devs who don’t know the critical logic and are filling in space hoping no one can see the slop. You can also find this madness with tests that test 3rd party logic while mocking the logic. Yes all kinds of bad tests exist. I find vibe coders do this a lot so it helps me to spot if they are critically thinking or just blindly accepting.

[-]

sintrastes@reddit

Damn, surprisingly enough I've actually never encountered this when using Claude Code personally. And I for sure know what kind of tests we're talking about here (because I've seen juniors make them before).

I typically say things like "Make sure this is well-tested", and the test cases are usually fairly reasonable.

I guess maybe the difference is detailed prompting v.s. vibe-coding? Or maybe I'm just lucky enough I haven't run into it yet. I've definitely run into other issues with agentic AI (my biggest pet peeve is it not doing what I tell it to do, and then it gaslighting me into saying it's "ready to ship").

[-]

MvKal@reddit (OP)

Yeah lol utterly useless and now i gotta explain why im not merging that

[-]

alee463@reddit

My manager actually did this to prove me wrong in regards to writing a test for something I checked in. This was the ammunition he used to fire me. He is now jobless as well. I made sure of that.

[-]

BadLuckProphet@reddit

I am on this bandwagon. This test is immediately followed by a production scenario where null is returned instead of 3 causing an issue. Talk to the dev and get met with "Well the issue can't be my code. The tests all pass."

Infuriating.

[-]

UUS3RRNA4ME3@reddit

Weaponised incompetence

[-]

systembreaker@reddit

Smile and nod and do the bare minimum to make them feel like they're managing and then do your job as you see fit as the person who was hired to do it.

[-]

MvKal@reddit (OP)

I get what ur saying but its like, maybe one out if ten tests has any semblance of usefulness in it.. the rest is just junk with llm comments that sound plausible enough to anyone who doesnt actually understand any of the code. So there is nothing to fix really, and he doesnt actually understand what the tests do, otherwise he wouldnt have submitted them for review.

[-]

Acceptable-Hyena3769@reddit

Its hard because as you said it sounds like it does something if you dont know the code well. Llms love to add 3==3 assertions and gaslight you into beliveing it will prevent service failure lol.

I would add a readme section and a template in your git ci/cd section that asserts that every pr is opened to list the issue, link the ticket, and explain briefly how it addresses the issue.

That would be enough to at least give you a solid ground for questioning.

But in the end if the manager is arrogant and ignorant you'll never get them to change. I may just be an introvert but i would either ignore their bullshit, change teams if it's really bad, or look for a new job lol

[-]

mpanase@reddit

I mean...

If later that week there's another PR that alongside other stuff undoes his changes... that'd be unlucky.

[-]

laueos@reddit

Turn it around and ask them to explain how this solution fits the problem (if there even is one). Ask more questions. Try to be open and friendly, as always if when you gently need to guide someone to get to their own conclusions.

[-]

MvKal@reddit (OP)

Yeah i tried that, he essentially pointed at a hallucinated claude comment being like "well it unblocks this" when it doesnt, which i have to then explain.

[-]

ribenakifragostafylo@reddit

Why not ask verification that proves the block is resolved? Not straightforward? If it's unblocked add an integration test that proves it

[-]

SolarNachoes@reddit

Use the 5 whys until they crumble in defeat.

[-]

SpritaniumRELOADED@reddit

So your complaint is that you, a professional with deep knowledge of the codebase, have to explain the codebase to somebody who has been newly tasked with managing it?

[-]

NeonQuixote@reddit

No, he’s complaining that the manager is submitting “fixes” that aren’t solely on the basis that “Claude said so.”

[-]

Wonderful-Habit-139@reddit

That are*

[-]

Material_Policy6327@reddit

No the issue seems to be a newbie is vibe coding their way to stupidity on a code base they barely know

[-]

hurricaneseason@reddit

You can't just have them talk about it, you need to have them follow through and actually produce and maintain something generated. I had to do this simply enough in the early GPT days by having a micromanager PO use his generated nonsense to create and maintain some basic spreadsheets. That was enough for him to understand how bad things get after even minor complexity, let alone at a full project level. Of course, he still pushes that devs are simpletons to be replaced, but you can't argue the stupid brackishness out of people's minds.

[-]

zzzthelastuser@reddit

From my experience this doesn't work in reality. You still have the work to ask the right questions and they just pass them over to Claude or whatever. I might as well talk directly to the AI and get the usual "good call! blablabla" response.

[-]

laueos@reddit

No, I mean talk to them in person. Not code review comments.

[-]

SquiffSquiff@reddit

FAANG EMs are a complete waste of time outside of FAANG in my limited experience. They re expecting to simply interface with a magcial system. First question here would be - 'why is manager coding at all?'

[-]

mpanase@reddit

Honestly, wouldn't be the first time that a "security review of the user role management system" has to happen... and it happens to remove the ability to write or even create PRs from a few managers...

And it can't be changed. It's a requirement by the high security council so we can keep our certifications.

[-]

ninetofivedev@reddit

A lot of companies want technical, hands-on managers who lead by example.

A lot of other companies have managers who sit a top their throne and delegate.

Both of these places exist, and both are valid ways to work.

[-]

g2gwgw3g23g23g@reddit

Manager with no understanding of code base is useless and incompetent imo

[-]

SquiffSquiff@reddit

Sure. Now explain why an incompetent coder and incompetent manager was put in a position where they think it is ok to drag the team down like this, as opposed to finding something where they can stay out of the critical path

[-]

positivelymonkey@reddit

Ego

[-]

Onedome@reddit

They typically suck outside of the framework they got all their experience in, they are very opinionated and cant seem to adjust to “cheaper” strategies. Some are good though, it usually just depends on the experience they had before FAANG.

[-]

SquiffSquiff@reddit

I could roll with that

[-]

MvKal@reddit (OP)

interface with a magical system

Thats what it feels like yeah. Imo it does make sense for that role to exist for some purposes but yes very much not coding.

[-]

mpanase@reddit

I usually play the fool

"I don't understand, could you explain this to me?"

"Could we capture some metrics to mesure the improvement introduced by this PR?"

[-]

cknipe@reddit

Have Claude review it.

[-]

CodelinesNL@reddit

Is there someone "in charge" you can go talk to? Even if they were contributing meaningful code, that's simply not what they're being hired for. If they wanted someone to write code, I'd assume they would hire a developer and not a manager.

[-]

MvKal@reddit (OP)

Yeah, my old boss/his direct boss, just wasnt sure if its worth that kinda escalation.

I also do not like the new manager on like, management style so might bring that up as well

[-]

riricide@reddit

It's probably best to nip this in the bud early.

[-]

the_useful_comment@reddit

Ask your old manager bait questions first until they get upset then show them the leet code grinder’s work once you have him ready to explode 😂

[-]

ObsessiveAboutCats@reddit

My boss is similar. At this point I'm covering my own ass by documenting my own work well (have already documented my concerns), and am sitting back to wait for the inevitable dumpster fire.

[-]

YoongiMarryMe95@reddit

Just commenting to say my non technical manager/owner has been vibe coding as well. Other devs have commented saying they can’t read his work (3k+ line slop). Not using our libraries, breaking the build, his stuff just doesn’t work. I just ignore his PRS and let everyone else tell him his stuff is crap.

[-]

briznady@reddit

Had my non technical skip open a pr up a few months ago. Treated it like I would anyone else’s pr and had a bunch of comments on it. He couldn’t fix it. I got on a call with him and walked through the fixes. He hasn’t attempted contributing at all since.

[-]

Qwertycrackers@reddit

Tell Claude to review it. Merge his stuff with many kudos. Number 1 rule is always kiss ass management.

[-]

ninetofivedev@reddit

Listen, I hate to have to defend AI, but you handle it all the same. You point out that the tests don't do anything but needlessly add additional code to the code base, and same for the change.

If they're your manager, your options are deal with it or go over their head.

Swear to god, AI came to the world and people can't rationale how to deal with it. In that regard, nothing has changed. It just happened less in the past because it's easier now.

[-]

Kilazur@reddit

Fight AI with AI. Have Claude setup in your pipeline to do a first review of every PRs.

[-]

maxip89@reddit

Learn by pain.

Diamonds are made by pressure.

Pessimism leads to doom. Optimism to power.

Three advices I can recommend.

[-]

Only-Fisherman5788@reddit

had one like this. came from a name-brand company, opened a PR that "looked like it worked," reviewers waved it through because arguing with someone's manager costs political capital. three weeks later a silent failure surfaced in prod and the autopsy pointed back to a condition the PR author hadn't thought to check.

the only thing that saved us long-term was making the review process hostile to plausible-looking code, not to the author. concretely: every infra PR needed a failure-mode section ("what breaks if X is null, if Y is racy, if Z arrives out of order") before merge. unblocked us from personal-politics pushback because the form was the gate, not the reviewer.

what's the leadership above him like, are they technical enough to push back if you escalate?

[-]

HansProleman@reddit

I'd probably just give him the rope to hang himself. It's easier and safer than anything else, though more deleterious to the codebase.

As you mention you have a good relationship with his manager, that can be worth a try. But it's politically risky.

[-]

pseudo_babbler@reddit

Tell him that his generated code isn't good and that you can clearly see that he doesn't understand why. It's not his job to waste your time generating code that he doesn't understand and you won't be putting up with it.

You're the expert, he's just a try hard who doesn't want to do the actual work of being a team admin and instead wants to do the fun stuff. Stand up to him.

[-]

zerocoldx911@reddit

FAANG managers are the worst

[-]

engineered_academic@reddit

Assert 3==3 is actually a valid test to ensure your test harnesses are working correctly. You generally want to approach this by asking him to explain this change and guide him by asking clarifying questions. Otherwise people will feel like you are attacking them for no reason.

[-]

Beneficial_Map6129@reddit

He probably is going to try for politics

I would just accept his PR and then quietly overwrite them citing added complexity

[-]

metaphorm@reddit

PR review comments are the first lever. A one-on-one meeting involving a guided tour of the codebase architecture and infrastructure is the second lever. Detailed technical documentation about the system is the third lever.

There's nothing inherently wrong with using Claude Code to assist on implementing stuff. But it still has to meet the quality standards, which means it needs to pass review, and it needs to be aligned with the architecture and design patterns of the codebase. Some of that comes from building up the human understanding of it. Some of it comes from building up the agent context. You'll have to do both.

[-]

shadytradesman@reddit

Ask him if he read the code before asking you to read it. Point out the basic mistakes. Tell him that pushing low quality code wastes everyone’s time. Tell him that it’s the developer’s responsibility to ensure their code meets a minimum quality bar before asking for review.

[-]

silhouettelie_@reddit

Don't they have to get approval for PRs? Raise comments asking them to explain/justify the bad code. Don't approve until they do?

[-]

dead-first@reddit

Tell him you wrote a dev manager agent and his services are no longer needed. Good luck on your future endeavors

[-]

MvKal@reddit (OP)

Lol i wish, hes one of the biggest users of the company llm chatbot lol

[-]

codemix@reddit

Forget that they're a manager, treat their contributions kindly in the same way you would anyone else new to the team - they're learning and it's your job to help them. You're not teaching them how to code, you're teaching them how the team works and what the quality expectations are. Rather than criticising their contributions, ask questions that help you figure out what their intention was. Then show them how you'd do it.