Ivy: You're writing code ten times faster, just to wait a hundred times longer for the test pipeline to clear.
Marcus: This is Marcus.
Ivy: And this is Ivy.
Marcus: And you're listening to Broke It for Tuesday, September 22nd. This is the show about what's new and what's breaking in AI.
Ivy: We have a packed show for you today.
Marcus: We'll dig into why generating code at lightning speed has developers just... sitting around waiting.
Ivy: Plus, a new IBM study shows a huge gap between execs who love AI and the employees who are stuck using it.
Marcus: And we'll look at New York's new AI safety law—and why it could set the standard for the entire country.
Marcus: Alright, so let's talk about speed. AI coding assistants like GitHub Copilot are everywhere, and they're letting developers crank out code at an unbelievable rate.
Ivy: Unbelievable is right. But the problem is, writing the code was never the real bottleneck, was it? It's making sure the code actually works.
Marcus: Well, yeah, testing. But we're generating so much more, so much faster! That's a huge win.
Ivy: Is it, though? From what developers are saying, we've just moved the traffic jam. Code that takes minutes to write now sits for hours waiting to get through the CI/CD pipeline.
Marcus: So the build queue is just a parking lot now. Developers are pushing dozens of changes a day, and the testing infrastructure just can't keep up with the firehose of new code.
Ivy: Exactly. The AI is a Ferrari engine, but we've strapped it to the chassis of a 1998 sedan. The whole system can't handle the speed.
Marcus: [laughs] Okay, that's a fair point. But isn't this a good problem to have? It means the next frontier for AI is optimizing the testing and validation part of the process.
Ivy: It's a problem managers need to wake up to. Buying Copilot licenses doesn't mean a 10x productivity boost if your team spends most of the day waiting for a green checkmark.
Marcus: So for developers, the real skill isn't just prompting an AI anymore. It's building a pipeline that can actually keep up. That's where the new value is.
Ivy: And if you're a manager, don't celebrate your team's code output. Measure how fast that code actually gets validated and deployed. Otherwise, you're just measuring how fast you can fill a waiting room.
Ivy: Alright, let's shift gears and talk about the human cost of all this. A new IBM study on AI in the workplace just dropped, and the findings are... pretty stark.
Marcus: Stark how? I thought the general consensus was that AI is taking over tedious tasks and freeing up humans for more creative work.
Ivy: [dry] That's the executive summary. The study found a deep divide. Executives see AI as a simple validation task—the AI does the work, a human just checks it. But employees tell a different story.
Marcus: Which is...?
Ivy: They call it the 'weight of unacknowledged work.' They're not just 'validating' things; they're constantly debugging, correcting, and cleaning up after the AI. They've become auditors for the algorithm, but without the training or the pay for that new, high-stakes job.
Marcus: Whoa. 'The Algorithm's Auditor.' That's a powerful phrase.
Ivy: And it gets worse. Get this: 80% of execs admit they're pushing AI faster than employees are comfortable with. And those same employees say their own skills are eroding because they're spending all their time just babysitting the machine.
Marcus: That's a recipe for burnout. You're giving people a tool that's supposed to help, but it just adds invisible work and makes them feel de-skilled.
Ivy: It's the classic disconnect. Management buys the shiny object, sees a dashboard that says 'efficiency up!', and doesn't see the human cost of tidying up all the AI's little mistakes.
Marcus: So if you're in that spot, you have to start documenting that 'auditor' work. It's a real, valuable skill. Make it visible.
Ivy: And if you're a manager, go talk to your team. Don't just look at the output metrics. Ask them how the work feels. The study is a huge red flag that you might be burning out your best people with 'help'.
Marcus: Now for our last story, let's talk regulation. New York's Governor Hochul just announced a huge new AI law, and this could be a very big deal.
Ivy: Let me guess, another commission to write a report about the ethics of AI?
Marcus: [laughs] No, this one has teeth. It's a new law requiring 'major developers' of large-scale AI models to register with the state.
Ivy: Okay, registration. What else?
Marcus: They'll also have to report any significant incidents. We're talking major security breaches, new dangerous capabilities, or just repeated, harmful bias. It essentially creates a public database of AI failures.
Ivy: Hmm. The devil is in the details, as always. Who defines a 'major developer'? What qualifies as a 'significant incident'? Is it self-reported? Because if it is...
Marcus: The details are still being hammered out, but the intent is clear. This is the 'California emissions standard' play. New York is such a huge market that companies will likely adopt these rules nationwide rather than build a separate, compliant model just for one state.
Ivy: So it's de-facto national regulation, driven by a single state. We've seen that movie before. The question is whether it just creates a compliance cottage industry for lawyers or actually improves safety.
Marcus: I'm optimistic! It moves the conversation from 'AI is magic' to 'AI is a powerful industrial product that needs safety standards and public accountability.' That's a massive shift.
Ivy: So, if you're building foundational models, your legal and compliance team is about to get a lot busier. Get ready for reporting requirements.
Marcus: And for the rest of us, it means that for the first time, there's a government agency you can point to when a major AI model goes off the rails. It's the beginning of a real accountability framework.
Marcus: So to wrap things up: AI code generation is hitting a wall, and that wall is the testing pipeline.
Ivy: A new study confirms it: executives and their employees are living in two different realities when it comes to AI at work.
Marcus: And New York state may have just set a new, national standard for AI accountability.
Ivy: Before we go, Marcus, speaking of AI getting things wrong... did you see that new image model that still, in late 2026, cannot correctly count the fingers on a human hand?
Marcus: [laughs] No! Still? I saw a picture of a beautiful, photorealistic family picnic, and the dad had seven fingers on one hand. It's the one thing they can't seem to fix.
Ivy: [dry] Some problems, it seems, are eternal. It's good to know there are still some things only humans can get right.
Marcus: That's our show. We'll be back tomorrow with more Broke It.
Ivy: Until then, try not to get stuck in any endless testing pipelines.
This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.