#TimTalk - Stop the Vibe Coding: How "Boring" AI Engineering Actually Wins | Krishna Kumaar Sharma

Ex Head of AI – Amazon, Building Omokai: Physical AI OS to turn speech into autonomous missions for drone/robot swarms, Enabling security and inspection teams to scale their operations.   / krisberlin   https://forms.fillout.com/t/iUWCb97oBbus https://omokai.com/ Here is his research paper: https://arxiv.org/abs/2512.01155 Here is a TLDR blog he wrote about it: https://leoniscounsel.substack.com/p/... Are sub-agent swarms and autonomous "loop-until-green" coding pipelines actually transforming software development, or are they just fueling a massive agentic hangover? In this episode, we sit down with Krishna Kumaar Sharma, former Head of AI at Amazon and founder of Omokai. While the tech world is obsessed with autonomous AI agents generating code end-to-end, Krishna exposes the reality behind the hype: spiraling compute costs, multi-agent coordination tax, and codebases broken by unchecked, "vibe-coded" updates. We dive deep into Krishna’s battle-tested alternative: Agentic Coding, The Boring Way. Learn how to transition from chaotic hype to disciplined engineering in complex legacy systems using his D3 Framework (Discover, Define, Deliver). Key Topics Covered: • The Hidden Costs of Unchecked Agents: Why throwing more agents at a problem amplifies errors by 17.2x and drives up coordination costs without delivering ROI. • The Death of Vibe-Coding: How unconstrained AI releases create production surprises and break downstream environments. • The D3 Framework: How structuring workflows into explicit Discover, Define, and Deliver phases constrains blast radiuses and preserves developer intent. • Multi-Model Orchestration: Leveraging model disagreement as a signal and routing tasks across diverse LLMs for cost-aware, audited execution. • Bounded Autonomy: Why the most effective agentic system isn't the one that writes the most code, but the one that creates zero surprises in production. Whether you are an engineering leader, software developer, or AI architect struggling with AI code quality, this conversation provides a realistic roadmap for building sustainable, production-grade AI engineering workflows. For more information on myself Timothy Hughes (Tim Hughes), social selling, digital selling, Revenue Operations RevOps, AI Agents or my company, DLA Ignite and Azpertilo.AI then follow these links Company Website: DLAIgnite.com | Azpertilo.AI Linkedin:   / timothyhughessocialselling   X (Twitter): https://x.com/Timothy_Hughes YouTube:    / timothyhughes1   Meta (Facebook):   / timhughessocialselling   Instagram:   / tim_hughes1   Clubhouse: @timothy_hughes Tik Tok:   / timothy_hughes