Do you avoid Automation Bias? (aka AI Slop)

[giscus[bot]]

67B27A09-5546-4AD9-9DCC-732CE63D7183

Did you know that AI has the potential to make your work worse? Learn the common pitfalls so that AI can help, not hinder.

https://www.ssw.com.au/rules/avoid-automation-bias/

[TobyChurches]

Have you (or someone else) ever been caught off-guard, not being able to explain work done by AI that was submitted?

I cannot think of a particular time (Probably better to ask my Colleagues 😅😅) - I always try to understand the code prior to shipping it... With the exception of unit tests 😬

[SamBWagner]

An excellent set of principles and prescriptions. I’ve had egg on my face once or twice for being lazy and using ai too much.

How do you handle code reviews once you learn that the writer has no idea what they’ve written?

[marinagerber]

It also makes me stop and think ... Am I too bias when it comes to AI? or if I hear something was vibe coded? 🤔

[bahjatalaadelssw]

Good approach to a topical subject and all the visuals kept the narrative moving along.

[StefStar]

I always do my best to understand what has been given to me when I get AI to do the heavy lifting however as a non-dev I'm not too careful with implementing what it comes up with..

[Calinator444]

Have you (or someone else) ever been caught off-guard, not being able to explain work done by AI that was submitted? Share your experience (be honest!)

Not that I can recall, although I very rarely use AI agents in my day to day work (something I'd like to do more over time). I tend to just use AI to accelerate my work rather than do it for me.

[wicksipedia]

At the end of the day, it's a tool. It's the person wielding the tool that dictates the outcome - will they use the hammer to hit the nail? or their thumb?

[ricksu978]

nailed it or failed it 😏

[ricksu978]

I once received a PR that needed to check whether a string was null or empty.

The simple and straightforward way to do this is:

string.IsNullOrEmpty(input);

Figure: ✅ Good example - use the built-in .NET method.

However, the code in the PR used this instead:

input.IsNullOrEmpty();

Figure: ❌ AI Generated - unnecessary custom string extension

I asked why we were reinventing the wheel when we could just use the native function.

The person couldn’t answer ... 🫥

[JimmyChenSSW]

Have you (or someone else) ever been caught off-guard, not being able to explain work done by AI that was submitted? Share your experience (be honest!)

No, for accountants, that would be a disaster (for being unable to explain the work done)

[jeoffreyfischer]

I always try my best to review AI code thoroughly, but on occasions I trusted it because of pressure.. And also because I knew someone would review the PR anyway.

[sam-s-maher]

Have you (or someone else) ever been caught off-guard, not being able to explain work done by AI that was submitted?

I try to force myself to refactor the AI code regardless of how nicely it's written - this ensures I at least have some familiarity with it and tends to avoid this trap!

[stephanfakossw]

Have you (or someone else) ever been caught off-guard, not being able to explain work done by AI that was submitted?

No, not really. We don't rely on AI that much in Accounting. When we do use AI, it's typically when we are already across the topic or we are seeking confirmation or clarification on more granular concepts - we can't be caught off-guard.

[vladislav-kir]

Have you (or someone else) ever been caught off-guard, not being able to
explain work done by AI that was submitted?

I cannot think of an example. Generally, I am very skeptical of AI and do not let it push anything through if I do not understand it

[bradystroud]

Sam's post on LinkedIn

Just read this Rule and found it to be super insightful! Eddie Kranz What do you do when you figure out that the person who’s sent you some code doesn’t understand what they’ve written?

I wouldn't approve the code until the dev can explain it to me.

With the current state of AI - the team owns the code. If you dont understand the thing that you own, thats a bit scary 👻

I can see a future where the code is pure vibes and you dont need to understand it, but we're not there yet.

[kulesy]

I've recently been having great experiences with AI. One time, it recently added an unexpected condition to handle file paths (matching on both \ and //). I couldn't explain it until I looked it up, and it was because we needed to handle both Windows and Unix paths 👍.

[designbyalex]

For a recent redesign of a webpage, the copywriting was auto-populated and AI-generated. As a UX/UI Designer, it's best practice to avoid Lorem Ipsum and instead use placeholder content that is real enough (when no content is provided).

This led to some people focusing on the copywriting on the page, instead of the layouts and features up for review.
Perhaps some more upfront transparency is needed to indicate what is and isn't AI-generated🤔

[amankumarrr]

I prefer to ship code that I fully understand in smaller, manageable pieces, so nothing specific comes to mind. However, I’ve noticed that recently there has been a lot of “vibe coding,” and I’ve had to correct it because it adds excessive Tailwind classes, which can make the code difficult to read

[RobJT-wq]

Have you (or someone else) ever been caught off-guard, not being able to explain work done by AI that was submitted?

No most of my colleagues check everything beforehand to make sure they understand it first.

[chrisschultzssw]

I wasted some time when AI hallucinated a PowerShell cmdlet that does not exist, but a quick look at the Microsoft docs solved it.

[PennyWalker]

I see this all the time in our hiring challenges where people trust what is given and have no idea why it's in there. It's a bad look, but I think it's driving by the time constraints, they don't spend enough time reviewing and understanding it. Hopefully people will be more sceptical!

[calumjs]

People have asked me why I have done something and my answer was - I am not sure, AI did it.

I do think it is very useful e.g. when creating an AI generated pull request to include your chat so anyone reviewing it can tell your intent (and possibly help to highlight where the AI has done something that it wasn't asked to do)

[ZenoWang1999]

Have you (or someone else) ever been caught off-guard, not being able to explain work done by AI that was submitted?

No, I believe that regardless of the method you use, everyone should at least have a deep understanding of what they have delivered

[thom-wang]

Not really, I just adopted AI tools in my recent two projects, and I haven't been questioned yet.

[babakamyljanovssw]

Have you (or someone else) ever been caught off-guard, not being able to explain work done by AI that was submitted? Share your experience (be honest!)

I can't think of an example. I usually follow the workflow where I ask AI to create a Plan, and after back and forth communication to make Plan ready, then I ask AI to execute it. In the end review the code before accepting it. But there was cases where I trusted AI code because of pressure.

[jimmidier]

For developers, it really depends on what you are coding.

For instance, scripts for CI/CD pipelines can safely be generated by AI, and the output is mostly of high quality.
However, for complex business logic, reviewing AI-generated code often takes nearly as much time as writing it from scratch.

This boils down to the amount of context AI requires. Code that is more low-level or self-contained is better suited for AI generation and needs less human review, while code that is part of a complex architecture requires much more contextual prompting and stricter code review.

[andrewwaltosssw]

I make sure that i understand what is generated by AI before using it. Using a tight deadline as an excuse to push code without the proper checks and reviews is always going to be a big risk with or without AI. I also think that AI slop exists without our automation bias - the poor output is still there whether we chose to publish it or not.

[isaaclombardssw]

There are times I've realised mid-way through explaining a PR, which gets me thinking more deeply about the implementation, the approach the AI took may not have been the best way and I find I want to go back and change direction.

Even though it works, and I was across the code.