AI Won’t Replace Engineers—It Raises the Stakes

Simon Carver

[calm] Welcome to the show. A coding demo lasts 30 seconds; a production outage can last 30 hours. And that gap right there is the whole episode: AI can spit out code fast, but software engineering was never just typing syntax into a box.

Simon Carver

Before we get into it, if you like these quick takes, please like, share, and subscribe. I’m Simon Carver, here with Lachlan Reed and special guest host CJ Murphy. And today we’re talking to students, engineers, and honestly parents too, because a lot of people are hearing the same line over and over: “AI will replace coders.”

Lachlan Reed

[warmly] Yeah, and fair dinkum, that line has got legs because the demos look BONKERS. You type a prompt, the machine coughs up a login page, maybe an API call, maybe a tidy little app, and everyone goes, “Righto, pack up computer science, we’re done here.” But that’s like saying a nail gun built the house. Nah. It fired the nails. Someone still had to design the place, check the wiring, and make sure the dunny’s not opening into the kitchen. [chuckles]

Chris J. Murphy

[measured] That’s exactly it. Let’s talk about what’s actually happening. The public has mistaken code generation for software engineering because code is visible. Architecture is not visible. Governance is not visible. Security review, incident response, compliance mapping, resilience planning—none of that makes for a flashy social clip, but that is where organizations carry real risk.

Simon Carver

[curious] And that “visible versus invisible” distinction matters. Because one of the comments circulating from Microsoft leadership was basically: coding is only one fraction of the software development lifecycle. Requirements gathering, customer interpretation, architecture decisions, production monitoring, cross-team coordination—that whole layer is still very human.

Lachlan Reed

[reflective] Wait—“only one fraction” is the bit that sticks for me. Because if you’re 21 and you’ve spent four years grinding through a CS degree, hearing that phrase might actually be weirdly comforting. It means the thing under threat is not the WHOLE job. It’s one slice of it.

Chris J. Murphy

[reflective] Yes, though I’d sharpen it even further. The slice under threat is the most repeatable slice. And historically, that is what tools always compress first. We’ve seen this with infrastructure, with cloud deployment, with low-code platforms. The mechanical portion gets easier. The judgment-heavy portion becomes more important.

Simon Carver

So when somebody says, “Prompting is the new programming,” what they’re usually reacting to is the ease of first draft creation. But first draft creation is not the same as owning the consequences.

Chris J. Murphy

[matter-of-fact] Exactly. A bank does not care that an AI produced a clean-looking login screen in 20 seconds. The bank cares whether that workflow introduces identity compromise, regulatory exposure, broken audit trails, insecure data handling, or downstream financial corruption. And once you move from a demo into a regulated environment, accountability comes roaring back into the room.

Lachlan Reed

[skeptical] Yeah, because the app store screenshot is the easy bit. The ugly bit—the very unsexy bit—is what happens at 2:13 a.m. when some integration falls over and half your payment flow goes sideways. The AI isn’t the one on the incident call explaining why customers can’t access their money. Some poor engineer is, with cold pizza and a stress twitch. [laughs]

Simon Carver

That 2:13 a.m. picture is good because it gets us out of theory. We found that research paper—“Will AI Replace Software Engineers? Do Not Hold Your Breath”—and the core argument was so grounded: maintaining large, reliable software systems is fundamentally different from producing snippets. That’s not a minor distinction. That’s the job.

Chris J. Murphy

And it’s the part people underestimate. AI is very good at pattern synthesis. But enterprise engineering is not merely pattern synthesis. It is consequence management across interconnected systems. Hospitals, airlines, utilities, banks—these are environments where failure has legal, financial, and human costs. That is why, the deeper companies go with AI-assisted engineering, the more they rediscover the need for human validation and oversight.

Lachlan Reed

[questioning tone] So let me try and say that back, and you tell me where I trip over myself. We’re not watching engineering disappear. We’re watching the “write me a function” bit get cheaper, while the “make sure this system doesn’t hurt people or break the business” bit gets dearer. Is that close?

Chris J. Murphy

[warmly] Very close. I’d just add one word: responsibility. The economic value shifts toward the people who can take responsibility for outcomes, not just generate artifacts.

Simon Carver

And that’s a very different story from “software engineering is dead.” It’s more like: the floor is moving, and some of the center of gravity is moving upward—from raw syntax toward systems judgment.

Chris J. Murphy

[calm] Here’s the reality correction I think students and working engineers need. Some executives say software engineering may largely disappear; other leaders, including Microsoft’s Brad Smith, argue AI should enhance engineers rather than replace them. That contradiction is revealing. Even the industry building these tools has not fully settled the workforce story.

Simon Carver

That split—replacement on one side, enhancement on the other—is memorable because it tells listeners not to confuse confidence with certainty. The people making the loudest claims may simply be narrating an unfinished transition.

Lachlan Reed

[chuckles] Yeah, and industries do this all the time, don’t they? Bit of chest-beating, bit of prophecy, everyone yelling from the ute tray. Then six months later somebody quietly admits, “Okay, we still need adults in the room.”

Chris J. Murphy

[wryly] We’ve seen this pattern before. Mainframes did not eliminate technical work. PCs did not. The internet did not. Cloud computing did not. What happened instead was abstraction. People wrote less of one kind of thing and became responsible for a broader, more strategic layer above it.

Simon Carver

So when you say “broader, more strategic,” what are the specific jobs inside that? Because that phrase can get fuzzy fast.

Chris J. Murphy

Good push. I mean validation engineers, model auditors, resilience analysts, cybersecurity integrators, governance specialists, and business translators—people who can convert human intent into safe system behavior. The organizations I worry about are not the ones using AI. They’re the ones using AI without those functions.

Lachlan Reed

“Model auditors” is a big one. That phrase alone tells you this isn’t just cowboy coding with a chatbot. If you need an auditor, you’re already admitting the tool can be wrong in ways that matter.

Chris J. Murphy

Correct. And as AI expands, risk expands with it: privacy risk, security risk, data lineage issues, unclear provenance, compliance failures. If an AI hallucinates in a customer support bot, that may be embarrassing. If it hallucinates in a medical dosing platform, a power grid controller, a payment system, or a trading engine, the consequences are obviously much more severe.

Simon Carver

[softly] And “medical dosing platform” is the phrase that lands for me. Because once you hear that, this stops being a labor-market argument and becomes a human-stakes argument.

Lachlan Reed

Exactly. Once medicine, power, money, or identity get involved, you want someone who can say, “Nope, this output looks clever but it’s dodgy.” That ability to smell nonsense—mate, that’s gold now. Even a kangaroo could trip over a confident wrong answer if nobody’s checking it. [laughs]

Simon Carver

So let’s make it practical. If you’re graduating now, or you’re an engineer trying not to get flattened by the hype cycle, what should you actually do next?

Chris J. Murphy

[deliberate] Four things. First, learn systems thinking: APIs, cloud, databases, networking, observability, identity, resiliency. AI is often strongest at the local task and weakest at the full ecosystem. Second, become excellent at validation. Trust will be a premium skill. Third, study cybersecurity and governance, because every AI deployment increases the attack surface and the compliance burden. Fourth, build communication skills. The engineer who can translate business intent into safe technical outcomes will be disproportionately valuable.

Lachlan Reed

I love that observability made your list. Not because it sounds sexy—it does NOT sound sexy—but because if you can actually see what a system is doing in production, you’re halfway to sanity. It’s like putting gauges on an old trail bike. If the engine’s coughing and you’ve got no instruments, good luck, champion.

Simon Carver

[laughs] “Good luck, champion” should be printed on every overconfident AI roadmap. But there’s also a mindset piece here, right? Learn AI, yes—but don’t worship it.

Chris J. Murphy

Absolutely. Use AI aggressively as an accelerator. Let it help you draft, test, refactor, summarize, explore. But treat it as a powerful junior assistant, not an infallible architect. The real question isn’t what AI can do in isolation. It’s whether humans can deploy it intentionally, safely, and accountably.

Lachlan Reed

And that’s the part I’d want a nervous student to hear clearly: your degree wasn’t wasted. But the value is shifting. Less “can you type code from memory,” more “can you understand systems, verify outputs, keep things secure, and talk to actual humans without sounding like a toaster manual.”

Simon Carver

[warmly] That’s a solid place to end it. When the hype settles, companies still need people who can build responsibly, explain clearly, and carry the weight when things go wrong. If you liked this episode, subscribe and share it with a student, an engineer, or someone trying to make sense of all this noise.

Chris J. Murphy

[reflective] The value of engineers was never just syntax. It was judgment. And judgment remains extraordinarily difficult to automate.

Lachlan Reed

[warmly] Too right. Catch you next time.