Upgrade to Pro — share decks privately, control downloads, hide ads and more …

From vibe coding to reliable (agentic) development

Sponsored · SiteGround - Reliable hosting with speed, security, and support you can count on. →

From vibe coding to reliable (agentic) development

Good #software #engineering practices become even more useful when we could use #AI to #generate bad software faster

Jordi Cabot

November 15, 2026

Transcript

18 slides
  1. 2

  2. 3

  3. Especially compared with the “classic” / “formal” approach Database Schema

    Frontend Graphical User Interface Code Generation State machine 4 Backend
  4. AI has also improved a lot Sceptics changed their minds

    with Opus 4.5 and the arrival of agents (with memory, skills, context, etc.) 5
  5. … but it is still far from perfect Buggy features

    Security, scalability issues, etc. . Tests that test nothing 6
  6. Spec-driven Development Before generating anything, the agent creates an application

    specification and checks it with the user: is it correct, which features belong in v1, and how will they be tested? 13
  7. Step 3– Generate the app: maximise reliability, minimise cost AI

    generates only the parts of the application not covered by an existing generator (available as “skills”) 17
  8. Comparing approaches Approach Specification Code generation Drawbacks Best suited to

    Classic / “formal” A person analyses and designs Deterministic generators Limited flexibility. Outside the scope: manual coding. Critical systems Purely agentic The agent interprets the requirements Generative AI Possible errors. Non-deterministic. High token usage. Prototypes and small, low-risk apps. Exploring ideas. Hybrid The agent creates the model, a human validates it Deterministic generators. AI for the rest. ???? Everything else 18

More Decks by Jordi Cabot

Other Decks in Technology