Agentic software engineering is a discipline shift rather than a tooling shift. The field can be situated historically (rule-based systems, statistical machine learning, transformers, the agent loop) and conceptually as the move from author-of-code to director-of-workflow. What feels like a sudden shift has a decade of preparation behind it.

The longer arc, running from the 1940s, shows the change is unprecedented in pace alone, not in kind. Programmers in the 1940s mediated intent by physically setting switches and connecting patch cables on machines like the ENIAC. Assembly languages in the 1950s introduced symbolic mnemonics. High-level languages abstracted from hardware. Structured and declarative languages shifted expression from how-to-compute to what-to-compute. Object-oriented programming reorganised intent around entities. Design patterns codified recurring solutions. Low-code and no-code platforms replaced authoring with configuration. Statistical code completion interpreted developer intent from local context. Generative AI extended interpretation to whole functions. Vibe coding from 2025 onward mediates intent through natural language and probabilistic inference rather than deterministic syntax. The current transition is the first to redistribute most of the implementation work to the machine.

The central claim of the course is that what separates professional from amateur use of agents is the presence of structure around the interaction. Over 68% of agent-generated pull requests face long delays or remain unreviewed in practice. True solve rates on real bugs collapse under rigorous audit from apparent success to a fraction of it. The speed-versus-trust gap is the defining tension of the current moment.

The autonomy-level framework for software engineering gives precise vocabulary for where current tools sit and what the next level would require. The levels run from manual coding through task-agentic and goal-agentic assistance to specialized and general domain autonomy. The foundational distinction at the heart of the course is ASE versus casual prompting. A casual prompter gives an agent an instruction and hopes. An agentic software engineer equips the agent and verifies the output.