Summary
Every engineering leader is getting the same question right now: where are we applying AI? I think that's the wrong question. The better one is which problems actually deserve it. AI capability is no longer scarce. Judgment about where to deploy it is. This roundtable is built on a single premise: generation got cheap; trust didn't. The cost of producing code, configs, and decisions has collapsed. The cost of verifying them, governing them, and standing behind them has not. That asymmetry, not model capability, is what should drive problem selection.
To make this concrete, we'll pressure-test a simple triage framework you can take back to your teams. Every candidate problem gets scored on four questions:
The Trust-Cost Filter
1. Verification. When the AI is wrong, how expensive is it to find out? A failing test is cheap. Eroded customer trust discovered six months later is not.
2. Reversibility. When it ships wrong, what does undo cost? A rollback is cheap. Data corruption, regulatory exposure, or headlines are not.
3. Accountability. When no human authored the work, who signs for it? If nobody will, that's your answer.
4. Ground truth. Does the workflow give you an observable signal of success, or does quality only show up as slow decay?
Problems that score well on all four are AI-native. Automate aggressively, prompt to production. Problems that fail on verification or reversibility need boundaries: human signatures, staged autonomy, or a deliberate no. The leader's job isn't finding the places where AI can work. It's drawing the lines where it may, and defending those lines upward as rigorously as any yes.
I'll bring what I've learned scaling an autonomous SDLC across a platform organization, along with earlier chapters building privacy and trust and safety programs at companies where mistakes were measured in headlines. Come ready to trade real examples: your most expensive yes, your most defensible no, and the metrics that separate leverage from displaced work.
Target audience: Engineering directors, VPs, and platform leaders who want a decision framework for AI investment, not another capability tour.
Host
Andrew Swerdlow
Senior Director of Software Engineering @ Roblox
Andrew Swerdlow
Sr. Director of Software, Roblox
https://www.linkedin.com/in/andrewswerdlow/ (LinkedIn)
https://www.andrewswerdlow.net/ (Book and Coaching)
Andrew Swerdlow is the Sr. Director of Software at Roblox, where he leads Engineering Acceleration (EA) and Core Platforms. In this role, he spearheads the AI-native transformation for a global platform serving tens of millions of daily active users.
A veteran executive with a career defined by building at planetary scale, Andrew previously served in executive leadership roles on Google Assistant, Instagram, as well as work at YouTube and a variety of other core Google products. His extensive experience includes leading globally distributed teams of hundreds of engineers and navigating the technical complexities of the world’s most ubiquitous products.
Andrew is a prolific inventor holding 39 U.S. patents and is the author of the Amazon best-selling book, "Tech Leadership: The Blueprint for Evolving from Individual Contributor to Tech Leader." His work focuses on the intersection of high-scale technical infrastructure, team productivity, and the future of software development through AI. He holds graduate degrees from the University of Victoria and Drexel University, along with a Stanford University certification.
Sign up to receive news and updates on ELC Annual 2026
