Powered by Neurons with Ewelina Kurtys

Today, we have Peter Morelli, Co-Founder of bitdrift and former VP of Engineering at Lyft, on the show! Pete shares how the challenges of monitoring millions of mobile devices at Lyft led to a new approach to mobile observability. In this episode, we explore why traditional observability tools struggle with mobile applications, from fragmented devices and unreliable network connections to the limitations of sampling and high data ingestion costs. Pete explains how bitdrift approaches these challenges by moving telemetry collection and processing onto the device itself, giving engineers a different way to investigate issues. We also discuss the engineering behind bitdrift’s SDK, its open-source blob-stream project, and its integration with ClickHouse. Finally, we dive into how AI agents are changing observability workflows, why access to the right telemetry data matters for automated debugging, and how engineering teams can use these tools to identify and fix more customer issues. Tune in for a deep dive into mobile observability, on-device telemetry, and the future of debugging mobile applications.

Today, we have Or Weis, Co-Founder and CEO at Permit.io, on the show! Or has spent over two decades in software engineering and is now building tools to help developers manage authorization and access control as AI agents become an increasingly important part of software systems. In this episode, we explore how AI agents are changing the way we think about identity, authentication, and authorization. Or explains why agents can’t simply be treated like human users or traditional machine identities, and how their unpredictable and autonomous behavior introduces new security challenges, from prompt injection attacks to the confused deputy problem. We also dive into the Model Context Protocol (MCP), why delegated access requires a different approach to security, and the shift from session-based permissions to action-level authorization. Or walks us through Permit.io’s MCP Gateway and introduces the concept of agentic intent—how organizations can track changes in an agent’s intended actions and intervene when its behavior starts drifting from the original task. Finally, we discuss how enterprises are navigating the transition to agentic workflows, the challenges of connecting AI agents to sensitive data and production systems, and why existing security and compliance practices need to evolve as AI adoption accelerates. Tune in for a deep dive into AI agent security, authorization, MCP, and what it takes to build secure and reliable agentic systems. Visit permit.io to learn more

Today, we have Dan Manges, Co-Founder and CEO at RWX, on the show! Dan previously served as CTO at Braintree and co-founded Root Insurance, and is now building a CI/CD platform designed for the agentic era. In this episode, we explore how AI coding agents are changing software development and why the bottleneck is shifting from writing code to validating it. Dan explains why traditional CI/CD workflows are struggling to keep up, and how RWX uses content-based caching, graph-based execution, and a CLI-first approach to make builds faster and more efficient. We also dive into how engineering teams can rethink their development workflows around AI agents, from validating code changes before opening pull requests to using sandboxes and distributed compute. Dan shares his thoughts on harness engineering, why developer tooling needs to evolve alongside AI models, and what it takes to build reliable software when code is becoming increasingly cheap to produce. Finally, we discuss how teams should measure engineering productivity beyond build times and costs, and why Dan believes the shift to agentic development is creating opportunities to rethink how software gets built and shipped. Tune in for a deep dive into CI/CD, AI coding agents, and the future of software engineering. Visit https://www.rwx.com/blog/series-a for more info.

Today, we have Robert Brennan, Co-Founder and CEO at OpenHands, on the show! Robert previously worked at Google and Fairwinds, and is now building an open-source platform for AI-powered software development. In this episode, we explore how AI agents are changing the way developers work, from automating code reviews and debugging to proactively monitoring production systems and fixing issues. Robert explains what makes an AI agent reliable, why agent harnesses are becoming commoditized, and how teams can continuously improve agent performance. We also dive into how software teams should prepare for agentic development, including the trade-offs between monorepos and microservices, trunk-based development, and traditional code review workflows. Robert also walks us through OpenHands’ SDK and Agent Canvas, and how they enable developers to build and manage autonomous workflows. Finally, we discuss the idea of an AI “dark factory,” where agents handle everything from writing code to testing and deployment. Robert shares his prediction that source code could become an antiquated abstraction layer within the next few years, and what that means for the future of software engineering. Tune in for a deep dive into AI coding agents, developer tools, and the future of autonomous software development. Visit https://www.openhands.dev/ to learn more.

Today we are talking with Vignesh Baskaran, the CTO and co-founder of Hexo Labs, about teaching AI agents to improve themselves. Vignesh has been training neural networks since 2012, back when he was still called a data scientist. Then he became an ML engineer and now an AI engineer, though he says the underlying work has never really changed. It's to figure out how to make a system behave the way you intend it to. He built the litigation search engine that Google itself became a customer of, and now he's chasing something new, agents that rewrite and retrain other agents without a human in the loop. We dig into Sia, the meta-agent at the center of Hexo's research, and why improving an agent means touching both its harness and its actual model weights, not just one or the other. We talk about proxy evals for when you don't have much to ground truth. The Darwin-Gödel machine and why formal verification is too strict a bar for anything commercial. How Hexo's work echoes DeepMind's Alpha lineage from AlphaGo to AlphaEvolve, and the spectrum from clearly verifiable to totally subjective tasks? Why VAE evals are quietly wrecking agent quality across the industry, and a great story about an agent that discovered a customer's own eval file was silently corrupted, something buried in hundreds of thousands of traces that no human would have caught.

Today we're talking with Kolton Andrus, the Founder and CEO of Gremlin, about what happens to reliability when AI is writing most of the code. Kolton helped build the Chaos Engineering practice of both Amazon and Netflix before starting Gremlin. In our conversation we talk about scar tissue, the intuition engineers develop from being woken up at 3:00 AM to fix production outages and how AI doesn't have any of it. It generates code in an afternoon that maybe took a team previously weeks to build, but none of those painful lessons come along for the ride. We dig into why 10x more code might mean 10x more failures. The concept of reliability guardrails, think ethical guardrails, but for keeping your systems up. Why you still have to test in production no matter how good your staging environment is? How Gremlin is rethinking their product for the world where agents, not engineers, are essentially the primary users.And why we're entering a painful, narrow part of the hourglass before AI gets good enough to handle all of this on its own.

Today we are talking with Andre Elizondo, the Director of Innovation at Mezmo about their open source agentic harness for SREs called AURA. Mezmo got their start handling observability data at scale. Logs, traces, metrics, the usual stuff. AURA is their answer to a growing problem, as system complexity outpaces humans' ability to make sense of all that data, how do you actually make it actionable for AI agents? We get into their approach to context engineering, essentially making data agent ready before it hits the model. Why they built their own orchestrator in Rust? How they handle memory and self-correction in agent loops? Their take on MCP and where it fits versus Skills and code sandboxing and how the SRE role is evolving as agents become trusted teammates. Visit mezmo.com/aura

Today on the show, we have a special guest — Ashmeet Sidana, the founder of Engineering Capital. Ashmeet started his career as an engineer at some great companies like Hewlett-Packard and Silicon Graphics before founding his own company, getting it acquired, and eventually starting his venture capital firm, Engineering Capital. With his strong engineering background, Ashmeet looks for startups that have a technical insight — something unique that gives them an edge over their competitors. This focus on technical insight sets Engineering Capital apart from other VC firms that often emphasize market insight or distribution insight or some other kind of advantage. We talked about AI, Exponential Engineers, Entrepreneurship, and had a lot of fun.

Today we have Dr. Ewelina Kurtys on the show. Ewelina has a background in Neuroscience and is currently working at FinalSpark. FinalSpark is using live Neurons for computations instead of traditional electric CPUs. The advantage is that live Neurons are significantly more energy efficient than traditional computing, and given all the energy concerns right now with regards to running AI workloads and data centers, this seems quite relevant, even though bioprocessors are still very much in the research phase.

Today we're talking with one of our favorite engineers, Rafal Wilinski. Rafal has been on the cutting edge of AI development in the last few years as he has led AI teams at Zapier and Vendr. Rafal walks us through the hard-won lessons about actually integrating AI tools into the applications you're building. One of the hardest things in integrating these AI tools is how to ensure you're getting better and not regressing as you improve your prompts and upgrade your models. He shows how using evals is one part of the story along with deeply investigating customer signals to see how they are or aren't succeeding with AI. Along the way, we also talk about RAG, his favorite models, his AI development toolset, and why Poland has been killing it lately. Check it out and be sure to follow Rafal if you want to learn more on building with AI.