Searchable transcript of Agentic Engineering vs Software Engineering: Beyond Vibe Coding — IBM Technology (10:46). Search for a phrase, then click its timestamp to jump straight to that moment in the video.
Captions sourced from the original video on YouTube, published by IBM Technology. The video, its captions and all related intellectual property remain the property of their respective owners; AINotes claims no ownership. Provided for research, accessibility and search — see the Transcript Notice and Copyright Policy.
00:00 For years, computer science programs have prepared students for a clear next step, a career in software engineering, and that pipeline used to work. I've experienced it firsthand, but with AI assisted coding and agentic coding, reshaping how software gets built, that path is becoming far less conventional. We're seeing something new, something called agentic engineering.
00:31 In this video, we'll break down the difference between software engineering and agentic engineering and how this shift is changing the role of developers. For decades, software engineering meant writing explicit instructions for computers. Now we're entering a world where engineers increasingly define goals and AI agents figure out the execution. But let's be clear, this does not replace software engineering.
00:57 It just changes where engineering effort goes. Traditional software engineering is fundamentally deterministic. You give explicit instructions and control flow. The same input, if it happens twice, you expect the same output twice. And this meant manually defining edge cases, permissions, retries, validation, and execution paths. Every step is predefined.
01:23 But, large language models introduced something fundamentally different. Instead of only executing rigid logic, these systems can interpret goals. And once you combine this with memory, tool use, planning and feedback loops, you get something that's closer to an AI agent. So from a high level perspective in traditional software engineering, developers write behavior.
01:57 Whereas in agentic engineering, developers shape behavior. In short, agentic engineering allows us to shift away from deterministic logic toward the nuanced territory of probabilistic judgment. This becomes even more obvious if we break down the term. So agentic refers to an organization of agents that write the code and the human developer that oversees and validates the output, as the agent or multi-agent system iterates through the subtasks we maintain a human in the loop.
02:39 Engineering, on the other hand, describes a level of expertise required to use agentic workflows for a meaningful code production that doesn't jeopardize the code quality. The surge of terms like agentic coding, vibe coding, and agentic engineering reflects a broader shift in how developers interact with AI. However, it's easy to confuse the several terms for AI assisted coding, especially as new terms are coined each year.
03:09 So to help distinguish some of the most frequently used terms, I actually like to think of this as somewhat of a spectrum and not from best to worst, but from how much agency the human retains versus how much the agency is delegated to the system. So, on one end you have traditional software engineering, where developers explicitly define nearly every workflow, rule and implementation detail.
03:41 AI may assist, but the human remains fully in control of the architecture and decision-making process. Moving along the spectrum, you get AI-assisted coding. And this is where tools help... Auto-complete, refactor, or generate code snippets while the developer still drives the overall implementation. Then comes vibe coding. This is where developers increasingly define or describe intent in natural language and iteratively shape the outcomes rather than manually writing every line themselves.
04:26 The focus shifts from precise implementation to creative direction and rapid experimentation. Further, still, is agentic coding. And so this one is where systems can autonomously plan tasks, use tools, write and execute code, debug failures, and iterate toward a goal with limited supervision. And finally, we have agentic engineering, which expands beyond code generation entirely.
05:07 Here, engineers are not just building software, but describing environments and designing them in which autonomous systems can reason, coordinate, evaluate outcomes, and adapt dynamically under constraints. The important thing is that these are not rigid categories. They represent a gradual shift in abstraction from directly programming behavior to increasingly orchestrating intelligence.
05:38 These terms actually bleed into one another. Teams often start with vibe coding to prototype ideas quickly and then evolve toward agentic coding and ultimately agentic engineering as the stakes and complexity increase. Instead of treating AI as just a coding assistant, agentic engineering treats AI systems as collaborators, capable of handling multi-step workflows under human supervision.
06:08 The engineer increasingly becomes someone who orchestrates the system rather than manually executing every task themselves. And importantly, this is not about removing humans from the loop. In fact, one of the biggest misconceptions around agentic systems is the idea that they eliminate the need for engineering expertise. In reality, the opposite may be true.
06:36 As these systems become more autonomous, the value of human oversight actually increases. Because now engineers are responsible not only for writing code, but also for ensuring reliability across probabilistic workflows. Perhaps this means fostering a culture that encourages experimentation while maintaining accountability. Meaning, agentic systems should accelerate development, not replace software engineering expertise.
07:10 The developer community understandably has some mixed reactions to this shift. Some folks are embracing this as an opportunity to pursue more thought-provoking and fulfilling work, but others are skeptical. And honestly, that skepticism makes sense. Traditional software systems fail in relatively predictable ways. You think about it, if there's a bug, you can often trace it directly to a faulty condition, state transition, or implementation detail.
07:43 Agentic systems are different. Because these systems operate probabilistically, they can appear correct while still making flawed decisions underneath the surface. They can hallucinate. They can misunderstand intent, misuse tools. Or generate technically valid, but architecturally poor solutions. That's why the central challenges in agentic engineering are to orchestrate and verify.
08:17 The engineering bottleneck starts moving away from writing every line manually and towards supervising, constraining, and evaluating increasingly autonomous systems. This also explains why vibe coding becomes problematic at scale. Vibe coding works extremely well for rapid experimentation and prototyping. But without strong engineering discipline, these workflows can also create technical debt very quickly.
08:48 AI-generated code that lacks structure, testing, or architectural consistency can become difficult to maintain over time. And this is where I think the phrase agentic engineering becomes useful because it emphasizes there's still engineering happening. This isn't about replacing software engineering as a whole, but rather expanding it into a new abstraction layer.
09:14 Agentic engineering changes what engineers spend most of their time engineering. And you may be wondering, what's next on the spectrum we looked at earlier? As the landscape of AI assisted software development continues to evolve, its terminology will likely also continue to shift. While agentic engineering provides a more mature and accurate framing for how developers collaborate with autonomous systems today, the practices behind it will continue to advance as AI agents become more capable.
09:48 At the same time, the role of developers will evolve from simply writing code to designing it. To shaping the behavior of these AI systems and also supervising them. If you're feeling unsure about where to begin, start small. Start by using these tools in your current workflow. Pay attention to where they help and where they break. Then build from there.
10:18 But let's not forget to resist substituting automation for understanding and quality. If you enjoyed this video, make sure to like and subscribe, and let me know in the comments below what AI topic you want to see next. Thanks for watching!