More Engineering Culture by InfoQ episodes

Doing The Right Thing: Accountability In The Age Of AI Coding Summary thumbnail

Doing The Right Thing: Accountability In The Age Of AI Coding Summary

Published 13 Nov 2026

Duration: 00:23:18

Human accountability remains paramount as AI blurs engineering roles, requiring leaders to foster transparency and better questioning to maintain software quality.

Episode Description

In this podcast, Shane Hastie, Lead Editor for Culture & Methods, spoke to Hugo Venturini about who is accountable when AI writes the code, how AI is...

Overview

Hugo Venturini, a software engineer and former CTO, discusses how AI is reshaping accountability and team dynamics in software engineering. He argues that while AI is a powerful tool, "humans are in charge and humans should be accountable." He notes that the growing size of pull requests is not a new problem introduced by AI but one that "was already existing before," and that teams should not review everything - automated verification can handle much of the checking, and senior engineers can be trusted to merge non-critical code without review.

He draws on his experience at Microsoft in 2014, when merging the software engineer and tester branches raised quality by making engineers accountable for the final product rather than one step in the chain. He sees AI as collapsing similar role boundaries - engineers, testers, product managers, architects - back into fewer people, but warns that quality suffers when people blame the tool rather than taking ownership. He emphasizes that the critical skill becomes asking the right questions, and that organizations must cultivate a culture of transparency and accountability without blame to empower people to step beyond their comfort zones.

What If

  • What if you stopped reviewing every line AI generates and trusted your automated gates instead?

    • Move: Configure your CI pipeline so that compilation and your full test suite act as the primary quality gate, and reserve human code review exclusively for the "heart of the product" logic rather than boilerplate.
    • Why Now? Venturini cites stories of "the 2000 lines, a senior engineer who's got a backlog of over 300 pull requests" waiting for review, and notes that "if it compiles, that's already a good thing" and passing tests may be sufficient for non-critical code.
    • Expected Upside: You could potentially cut your review backlog substantially by letting automated checks handle verification, freeing your attention for architectural decisions and the code that actually differentiates your product.
  • What if you treated every AI-generated output as a junior engineer's submission that you are personally accountable for?

    • Move: Before merging any AI-generated code, write down the specific edge cases, security implications, and performance concerns you would flag for a junior engineer, then work through that list yourself rather than accepting the output at face value.
    • Why Now? The source states that "if it fails, the one conducting this great tool is accountable, is in charge and should answer what goes wrong," and notes that quality is currently suffering because "people are still blaming the AI and don't feel accountable for what the tooling is doing."
    • Expected Upside: You shift from passive consumer of AI output to active quality owner, mirroring the Microsoft 2014 finding that making engineers "accountable for the final product" raised quality.
  • What if you started a recurring "I don't know" session with your team or collaborators, even without a management title?

    • Move: Hold a short weekly or biweekly session where anyone can ask you anything, and you answer honestly - including "I don't know" or "I made a mistake" - with no judgment, and follow up on unanswered questions.
    • Why Now? Venturini describes this as the key move that shifted an organization where "people would lie, hide, steal" into one of transparency, noting that "after a few sessions, people started to ask me difficult questions."
    • Expected Upside: Over a few sessions, your team may begin surfacing harder problems earlier and solving them collaboratively, as the source describes questions growing in difficulty and volume once trust is established.

Takeaway

  • Automate verification before human review: if your code compiles and passes your test suite, you may not need to review every line - reserve human review for the "heart of the product" logic rather than boilerplate.
  • Take explicit ownership of AI output: before merging AI-generated code, run through a checklist you'd give a junior engineer (edge cases, security, performance) rather than accepting output at face value, since "the one conducting this great tool is accountable."
  • Define the problem before prompting: spend time clarifying what "done" means and what the user actually needs before asking AI to generate, because "if you're asking the wrong questions... AI barely helps you."
  • Practice transparency at your level: share what you don't know, what you got wrong, and what you're working on with your team or collaborators, even without a management title, to build the trust that enables harder questions.
  • Rotate between deep technical work and coordination: the source describes this pendulum as building "all the understanding of the scope" needed to create products that answer user demands, and notes it need not be permanent.

Recent Episodes of Engineering Culture by InfoQ

18 Sept 2026 Signals and Levers: Building Thriving Engineering Organizations

"Engineering leadership requires respect, systems thinking, and adaptability to navigate non-linear software delivery, AI challenges, and culture - balancing technical debt, well-being, and real-world execution for effective team dynamics."

10 Jul 2026 Formal Methods for Every Engineer in an AI-Powered Future

"Formal methods, like the new **Quint** language, enhance software engineering by defining system behaviors and edge cases, with AI expected to automate specification generation by 2026, improving accessibility and validation."

More Engineering Culture by InfoQ episodes