Why Software Is Never “Done” Anymore with Iccha Sethi, SVP of Engineering at Vanta

Modern CTO54mMay 7, 2026

Get the full intelligence

Search transcripts, export clips, track mentions, and explore all topics from “Why Software Is Never “Done” Anymore with Iccha Sethi, SVP of Engineering at Vanta” inside PodZeus.

AI-Generated Summary

In this episode of Modern CTO, Joel Beasley sits down with Iccha Sethi, Senior Vice President of Engineering at Vanta, to explore why software is no longer 'done' in the modern era. Sethi explains that the rise of AI and agentic features has fundamentally changed software development, introducing constant drift and non-determinism from underlying models—vendors like OpenAI and Anthropic continuously evolve their models without clear announcements. This means engineering teams can never truly 'ship and forget' a feature. Instead, they must maintain ongoing quality through continuous evaluation, both offline using golden datasets and online via real customer data. Vanta has built a robust AI evaluation maturity model with five levels, including traceability, golden data, offline and online evaluation, experimentation, and self-improving feedback loops. Sethi emphasizes that engineering leadership now requires a new mindset: staying ahead of drift, managing AI adoption responsibly, and fostering a culture of continuous learning and improvement. The conversation also dives into practical tools and leadership philosophies. Sethi shares how Vanta uses SPAN to track onboarding success, investment mix, and AI tool adoption—especially to monitor 'AI slop' (overly large PRs). She reflects on her career journey from Atlassian to GitHub to Vanta, highlighting how each role offered unique learning opportunities and shaped her leadership style. Key themes include the importance of customer exposure, giving direct feedback with empathy, and adopting proven operational frameworks from mentors. Sethi’s advice to aspiring leaders centers on observing great behaviors and integrating them into your own toolkit—building a personal leadership 'tool belt' over time.

Key Takeaways
1

Software is never truly 'done' due to constant AI model drift and non-determinism from third-party vendors.

2

Implement a five-level AI evaluation maturity model: traces, golden datasets, offline/online evals, experimentation, and self-improving feedback loops.

3

Use tools like SPAN to monitor onboarding, AI adoption, and PR quality to prevent 'AI slop' and ensure healthy engineering practices.

4

Leadership success comes from observing and adopting proven behaviors—building a personal 'tool belt' of effective practices.

5

Give direct, kind feedback early and often; avoid the 'ruinous empathy trap' by naming patterns and co-creating solutions.

…and 3 more takeaways available in PodZeus

Chapters
0:00
10 min

Why Software Is Never 'Done' Anymore

Software is never really done anymore. Even if you're pinned to a specific version of a model, there's always drift happening underneath.

Highlight
10:00
10 min

The AI Evaluation Maturity Model at Vanta

We’re not just running tests. We’re running experiments to find the best model for each use case, and then feeding those insights back into the cycle.

Highlight
20:00
10 min

Managing AI Drift and Quality at Scale

It’s like having an alert in a channel when your P99 latency is higher. We’re doing the same thing, but for AI quality.

Highlight
30:00
10 min

Leadership Lessons from a Career in Engineering

Sethi reflects on her journey from Atlassian to GitHub to Vanta, emphasizing learning from each role. She shares how she prioritizes growth over brand, and how customer exposure and feedback shape her leadership.

40:00
10 min

The New Art of Prompting and Task Chunking

Sethi discusses how AI changes the skill set of engineers: breaking tasks into small, reviewable chunks and mastering prompt engineering. She compares pair programming with AI to mentoring a junior engineer with senior confidence.

High-Impact Quotes
Treat AI agents as junior engineers with senior confidence—pair programming with AI requires new skill sets like chunking tasks and prompt engineering.
Iccha Sethi25:00
Viral: 88.0
Software is never really done anymore. Even if you're pinned to a specific version of a model, there's always drift happening underneath.
Iccha Sethi1:45
Viral: 85.0
Timing is everything. Don’t over-invest in architecture too early. Accept trade-offs to find product-market fit first.
Iccha Sethi66:00
Viral: 82.0
Speakers

Host

Joel Beasley

Guest

Iccha Sethi
Topics Discussed
AI-Driven Software Maintenance95%AI Evaluation Maturity Model92%Continuous Evaluation and Quality Monitoring90%Engineering Leadership in the Age of AI88%Customer-Centric Engineering85%AI Tool Adoption and Risk Management80%Leadership Development and Mentorship78%Team Onboarding and Engineering Health75%
People & Brands

Vanta

organization

18xPositive

Joel Beasley

person

15xPositive

Iccha Sethi

person

12xPositive

SPAN

product

10xPositive

GitHub Actions

product

5xPositive

Andy

person

4xPositive

Atlassian

organization

4xPositive

Envision

organization

3xNeutral

NPM

product

3xPositive

Gong

product

3xPositive

Get the full intelligence

Search transcripts, export clips, track mentions, and explore all topics from “Why Software Is Never “Done” Anymore with Iccha Sethi, SVP of Engineering at Vanta” inside PodZeus.

Start discovering podcast insights today

Start with a 7-day trial and explore a growing catalog of popular podcasts. No credit card required.

No credit card required • 7-day trial • Cancel anytime