Key Takeaways

  • Dr. Gautham Pallapa banned the term "vibe coding" across engineering teams and interns at Scotiabank, requiring staff to act as systems engineers rather than syntax writers.
  • Scotiabank classifies writing raw code, generating unit tests, and drafting technical documentation as repeatable, low-value work best offloaded to automated systems.
  • Pallapa rejects AI proposals that cannot justify their specific business value, calling uncritical AI adoption "resume-driven development."
  • Engineering leaders must move from sequential software development life cycles toward rapid feedback loops that force humans to focus on architecture and system boundaries.

The Rule Behind Banning Vibe Coding

Dr. Gautham Pallapa, Principal Director of AI and Innovation at Scotiabank and author of Invaluable, gave his engineering teams a blunt directive: stop saying vibe coding. The ban applies to senior staff and interns alike.

Pallapa is clear about why. “I've told them they cannot ever use vibe coding, uh, as a word, um, or a phrase within, um, our premises, because what I really want is I don't want to have a developer, a coder anymore. What I want to have is a software engineer.” (play from 31:09)

Vibe coding treats software production like an improvisational chat session. A developer prompts an AI model, scans the output, shrugs, and commits the code because it runs without errors. In an enterprise handling billions of dollars in financial transactions, that habit creates hidden architectural rot. As Pallapa explained, “there's a lot of difference, and I actually walk them through the difference between vibe coding and, uh, enterprise software development. Um, very, very different beasts.” (play from 31:24)

Why Syntax Writing Is Now Low-Value Work

When language models can generate Python scripts or SQL schemas in seconds, writing syntax loses its commercial value. Pallapa forces his teams to separate human judgment from mechanical generation.

“It has helped us start looking at everything that we do and then identify things that are high-value activity for us as humans, and then our entire goal is to make these low-value activities, like coding, writing documentation, writing tests, all those various things that are systemic and repeatable, those are now low-value activities, and so they can be handled over by systems,” (play from 22:22) Pallapa said.

Writing boilerplate syntax, drafting basic documentation, and authoring standard unit tests used to take up eighty percent of a junior engineer's week. Pallapa treats those tasks as commodities. The actual work of an engineer is structural: mapping failure states, enforcing security boundaries, auditing data flows, and designing systems that survive production traffic. If your engineers spend their days writing routine loops, you are paying human salaries for automated tasks.

The Trap of Resume-Driven Development

Engineers love shiny tech. When large models and agentic workflows hit the market, teams rushed to plug autonomous agents into internal backends whether the workload needed them or not. Pallapa challenges this impulse directly during architecture reviews.

“And every time they say they want to do all these various things, I ask the question, like, 'Why do you need an agent for this? Why do you need AI?'” (play from 33:21) Pallapa noted. “Like, tell me tell me what is the value agentic AI is providing or generative AI is providing. And if they're like, 'Well, because everyone else is doing it,' then I'm like, 'Okay, you're doing resume-driven development at this point.'” (play from 33:29)

He is also pushing his teams to re-evaluate how code moves from concept to production. “One of the things I am experimenting with is, should the software development life cycle change with AI? Why are we going through the same steps? Why is it not looping multiple times with the feedback loops and continuous improvement rather than being like this sequential approach?” (play from 34:34)

When AI handles code production, linear waterfalled handoffs make no sense. The engineering team must operate in tight loops, testing architectures against continuous system feedback.

What to Do With This

Open your company's pull request tracker tomorrow morning and pull the last five merged code reviews. Check whether the comments debate syntax choices and formatting or system architecture, data dependencies, and failure handling. If your team spends review cycles nitpicking variable names and boilerplate, mandate that linters and AI handle syntax generation, and require that every PR description explain how the change impacts upstream services and system reliability.