Files
thereadme/slides/why_agentic_ai_wont_replace_engineers/script.md
T

4.7 KiB
Raw Blame History

Script: Why Agentic AI Won't Replace Engineers


The Hook

[VISUAL: Face Camera]

"Lately, the entire software industry feels like it's holding its breath. We see demos of 'Devin' or 'AutoGPT,' and the narrative is always the same: Software engineering is cooked. Writing your own code is a dead-end career."

"The story goes that programming is just a commodity now, and we should all prepare to be replaced by robots that are faster, cheaper, and don't need sleep."

"But... I don't buy it. I refuse to believe the future of my career is just becoming a 'shepherd of AI agents.' In fact, Im doubling down on coding."

[VISUAL: Switch to Screen Share - SLIDE 1: Title Slide]

"This is the 'Steel Man' argument for why the rumors of our demise are greatly exaggerated. Let's look at the math."


The Reliability Gap

[VISUAL: Advance to SLIDE 2: P(Reliability) < 1.0]

"It starts with this simple inequality: The Probability of Reliability is always less than 1.0."

"Here is the core conflict: Software engineering is deterministic. If you miss a semicolon, or pass the wrong type, or mishandle a null reference, the system crashes. It requires perfection."

"Agentic AI, by definition, is probabilistic. It predicts the next likely token based on training data. You can push that probability closer and closer to 1.0, but you can never actually reach it."

"In high-stakes environments—healthcare, finance, infrastructure—'mostly good' isn't good enough. As long as that gap exists, you cannot remove the human from the loop. You can't build a deterministic skyscraper on a probabilistic foundation."


The Economics

[VISUAL: Advance to SLIDE 3: The Review Tax]

"And because you can't remove the human, you run into an economic problem. I call this The Review Tax."

"We all know AI drives the Creation Cost way down. It feels like magic to generate code instantly."

"But as the volume of code goes up, the Verification Cost skyrockets."

"Imagine your reaction today if a coworker opened a PR with 500 lines of code across 60 files. You'd hate it, right? Now imagine that PR was written by a robot. No human feels ownership of it, but someone still has to review it."

"If it takes you longer to verify the code than it would have taken to write it yourself, the AI actually has negative economic value."


The Architecture Problem

[VISUAL: Advance to SLIDE 4: Speed != Coherence]

"That leads us to the difference between Speed and Coherence."

"LLMs are incredible at speed. They can generate functions instantly. But they struggle with long-term coherence."

"Projects eventually grow larger than the context window. An agent might write a feature that works locally, but breaks the architectural patterns of the legacy system because it lacks the 'Big Picture' intuition."

"If you rely solely on speed, you don't get a better product. You just get spaghetti code at 100 miles per hour."


[VISUAL: Advance to SLIDE 5: Liability]

"Then there is the question nobody likes to answer: Who do I sue?"

"If an autonomous agent deploys code that causes a massive data breach... whose fault is it? The vendor? The company? Or you, the engineer who prompted it?"

"Are you ready to carry malpractice insurance just to approve a PR from an AI? Companies need a 'throat to choke.' Until the legal liability is solved, full autonomy is a corporate fantasy."


The Human Cost

[VISUAL: Advance to SLIDE 6: Experience Required]

"But my biggest fear isn't legal. It's about how we learn. Build, Debug, Learn."

"This is the loop. You gain intuition by struggling through bugs. That is how you become a Senior Engineer."

"If we just hand off the 'Build' and 'Debug' phases to agents, we break the loop. We risk hollowing out the workforce. If you never struggle, you never learn. And if you never learn, how will you ever know if the AI is lying to you?"

"We are creating a generation of pilots who only know how to use autopilot."


Conclusion

[VISUAL: Advance to SLIDE 7: What to do?]

"So, what is the move? Do we give up?"

[VISUAL: Click to reveal point 1: "- Don't surrender your skills"]

"First: Don't surrender your skills. Do not let your intuition rot on the vine. Keep coding. Keep building difficult things from scratch."

[VISUAL: Click to reveal point 2: "- Avoid dependency"]

"Second: Avoid dependency. Use AI as a tool—like a really fast intern—but never let it become the architect."

[VISUAL: Face Camera]

"The future isn't about replacing us; it's about augmenting us. There is so much left to build, and the world still needs engineers who know what they are doing. So double down on your craft."

"Thanks for watching."