Skip to content
Artwork for Engineering at Scale

Engineering at Scale

Unleash

Conversations with engineering leaders at the world's largest and most innovative enterprises on shipping software fast, safely, and at scale in the age of AI. Brought to you GetUnleash.io

Play
  • 5 episodes
  • fortnightly
  • Avg 40 min
  • English
  • Thursday · 40 min

    What breaks first when your team ships 10x more AI code

    At Uber, Anant Gupta (Chief Technology Officer @ Mixpanel) watched the business grow from a few hundred million trips to billions. At that scale, a one-in-a-million edge case stops being an exception and becomes the norm. Now he argues AI coding is a solved problem. The hard part is deciding what to build and knowing whether it worked once it shipped. On this episode of Engineering at Scale, Anant shares how he keeps the blast radius small when teams ship faster than ever. We cover: - AI coding is solved, deciding what to build isn't - What breaks first when teams ship 10x more code - Separation of concerns, redundancy and human fallbacks in healthcare - Why one-in-a-million exceptions become the norm at scale - Reducing blast radius with gradual rollouts by default - Why code freezes can cause more production issues

  • October 29 · 28 min

    AI-generated code in banking: why AI never touches the money

    Satya Singeetham (EVP & Chief Technology Officer @ New Horizon Bank) tells his engineers to use AI as much as they can, with one hard line. Anything that touches people's money is off limits for the AI, because he is the one signing with the OCC when a product goes live. On this episode of Engineering at Scale, Satya shares the guardrails he uses to get AI speed inside a regulated bank, from cross-model code review to feature flags on money movement. We cover: - Why anything touching customer money stays off limits to AI - Having a non-Claude model review Claude's code - Why engineers must be able to fix AI code without AI - Which banking features go behind a feature flag - Pausing feature work to fix incidents at the root - Going from building core banking to running a bank

  • October 15 · 42 min

    Why faster AI code generation doesn't make engineering faster

    Martin Georgiev (Former VP of Engineering @ Lendable) started an AI-first team, and one engineer told him they were embarrassed to put their name on a PR because the code wasn't generated by them. That's when Martin realized AI was changing his team's professional identity. On this episode of Engineering at Scale, Martin shares how he helped grow Lendable's engineering team from 3 to 45+, rebuilt a UK lending platform for the US, and kept a human on the final sign-off. We cover: - Why faster code generation creates queues everywhere else - Launching an AI-first team instead of retrofitting existing ones - Keeping human accountability at the final sign-off - Why US expansion meant building a multi-tenant lending platform - Hiring a trusted friend network to scale remote teams - Building a career framework at six engineers

  • August 27 · 47 min

    Ship more or cut headcount: the question engineering leaders are avoiding

    Kevin Hill has spent his career inside systems that don't forgive mistakes. From running Windows Error Reporting at Microsoft, where a single conflicting data point triggered an email from the president of Windows to two engineers at once, to leading engineering at Salesforce, where global enterprise infrastructure runs 24/7. He has built a specific point of view on what scale actually demands from leaders, and it's not what most engineering orgs are currently optimizing for. In this episode with Egil, Kevin gets into why the real bottleneck in AI-assisted development isn't code generation, it's review and production safety. He also shares how he's structuring teams, building AI adoption from the peer level up, and why the identity crisis hitting engineers and middle managers right now is the conversation most leaders are avoiding. Topics discussed: Spec-driven development as a forcing function for consistent AI output quality Why middle management identity is the most underdiscussed pressure point in AI transitions The three-tier adoption model and why the middle 60-70% is where leaders should focus AI champions at team level as a structured peer-to-peer learning mechanism Why reviewing and safely shipping AI-generated code is now the core engineering bottleneck Shifting team composition from skills-based to product and value-based structures Ship more roadmap or cut headcount: the question every enterprise engineering leader is actually wrestling with How the value of "right skills" is shifting away from language expertise toward architecture and engineering judgment

  • August 13 · 42 min

    Why regulated enterprises can't skip AI governance

    Kevin Trilli has spent multiple decades in product, founded a zero-to-one company where product and technology were one job by necessity, and now holds the combined Chief Product and Technology Officer role at FIS Originations. He comes to this conversation with a specific argument: the bottleneck in AI-accelerated product development is no longer writing code. It's everything downstream of it. His framing is direct. Slapping an agent on top of a legacy platform doesn't eliminate tech debt, it just exposes it faster. Customers in B2B, especially regulated industries, can't consume features at the pace you can now generate them. And if your organization hasn't built a proprietary knowledge pipeline, your AI tooling is running on generic context and producing generic output. Kevin walks through how he's restructuring the product and engineering process at each stage, where he draws a hard line on not outsourcing strategy to AI, and what governance for agentic products actually looks like when your customers get audited. Topics discussed: Why platform limitations don't disappear when you add an agentic layer on top Using AI as a signal aggregator for strategy without letting it become a "coin-operated strategy machine" Building a proprietary knowledge pipeline as the prerequisite for any intelligent tooling Customer consumption velocity as the real constraint on B2B product development speed Governance and compliance requirements for agentic products in regulated enterprise markets Why the strongest argument for top-down AI mandates is framing it as a career investment, not a company directive Why AI fluency for senior leaders means being personally hands-on, not directing from the outside Making customer and market context queryable by anyone in the org, so engineers don't need to be in the room to know what to build

Showing 1–5 of 5 episodes