Signals: DevTalk with Michael Dowden

From automated bug fixes to everyday errands: how to delegate clearly, verify results, and set boundaries on cost and autonomy.

What does it take to manage AI agents rather than just ask them questions? Debbie O'Brien shares examples from her development work and personal automation experiments.

Topics covered:

- Delegating to agents like new team members: clear instructions, small tasks, and increasing responsibility as trust develops.

- Automating bug discovery, reproduction, fixes, and verification, with screenshots and video for human review.

- Starting locally before moving automated checks into CI.

- Keeping research in Markdown documents to provide context for planning and execution.

- Managing token costs by adjusting routine frequency, reviewing expensive tasks, and clearing unnecessary context.

- Setting boundaries around purchases, email, and travel tasks—and recognizing when agents act on incomplete information.

- Checking information across multiple sources rather than relying on a single model or document.

- Why Debbie believes junior developers still have an important place on engineering teams.

Find Debbie's videos, blog posts, and social profiles at debbie.co.

Beyond faster code: evaluations, human oversight and the ethical choices behind useful, inclusive products.

What does trustworthy AI look like when its behavior isn't deterministic? This conversation connects practical software delivery decisions with their wider impact on users and society.

Topics include:

- Bridging AI's trust gap through evaluations, guardrails, monitoring and clear routes to human intervention.

- Planning for mistakes with reversible actions and compensating transactions.

- Why faster code generation can shift bottlenecks to problem framing, testing, release and maintenance.

- Managing token costs and communication overhead in agentic workflows.

- Protecting user experience through research and testing rather than treating more features as more value.

- Digital sovereignty, AI regulation and the case for collaboration and federated infrastructure.

- Digital inclusion across connectivity, age and socioeconomic circumstances.

Jeff's closing takeaway: engineers should take more interest in the humanities to better understand the people and societies their systems affect.

In this episode, host Michael Dowden sits down with Venkat Subramaniam to discuss the real strengths and limitations of AI in software engineering. Rather than treating AI as true "artificial intelligence," Venkat describes it as a fast "accelerated inference engine" that amplifies human capability rather than replacing it. Developers can delegate code generation to AI, but not their responsibility or accountability for the final product. Navigating this shift requires moving from a mindset of "trust but verify" to "don't trust, verify heavily" through rigorous CI/CD pipelines, unit testing, and continuous human oversight. Instead of eliminating developers, AI will reshape their role, removing tedious boilerplate work and elevating the focus toward critical thinking, software architecture, verification, and long-term production stability.

In this episode, host Michael Dowden sits down with platform engineer and author Russell Miles to explore how software engineering is fundamentally changing in the era of artificial intelligence. Drawing from the concepts in his book, The Sovereign Engineer, Russell offers a practical vision for developers facing rapid technological shifts. Rather than viewing AI as a replacement for software practitioners, the conversation explores how engineers can adapt their practices, preserve their authority, and construct better environments to thrive.

In this episode, host Michael Dowden sits down with user experience expert Christian Kuhn to explore the evolving relationship between UX and AI design, drawing parallels to the early smartphone revolution. They examine the limitations of prompt-based interfaces and advocate for more intuitive, prompt-free experiences that reduce cognitive load. Kuhn also addresses ethical concerns around addictive design, synthetic empathy, and user trust, introducing the Human AI Experiences (HAX) framework as a guide for responsible AI design. Ultimately, both Dowden and Kuhn argue that as AI development matures, user research and human-centered UX will once again become essential to building meaningful products.

In this episode, host Michael Dowden speaks with Angular expert Manfred Steyer about integrating artificial intelligence into large-scale enterprise applications. Moving beyond text-heavy chats, the discussion covers core takeaways on building Agentic User Interfaces (UIs) that can stream dynamic components, dashboards, and co-created workflows on demand. To keep non-deterministic AI applications reliable and secure, developers must implement robust engineering frameworks—including standardized protocols like AG-UI and A2UI to avoid backend technology lock-in, deterministic safety harnesses, advanced human-in-the-loop patterns, and modern component-level blackbox testing using Vitest.

In this episode, host Michael Dowden sits down with Sebastian Bergmann, the creator of PHPUnit, to discuss the evolving landscape of open-source software development. Bergmann shares details on upcoming features in PHPUnit 13.3 designed to tackle flaky tests. The conversation delves deeply into modern supply chain security, highlighting recent changes in Composer's dependency handling and the necessity of automated dependency hygiene. Bergmann also addresses the dual-edged sword of AI in engineering—explaining how AI tools are driving an onslaught of security reports while warning against the dangers of developer "de-skilling." Finally, they examine the long-term financial sustainability of open-source projects and call on businesses to support the shared digital infrastructure they rely on.

In this episode, host Michael Dowen sits down with API strategy expert Dr. Matthias Biehl to discuss how the Model Context Protocol (MCP) serves as the missing link connecting generic Large Language Models (LLMs) to real-time, proprietary enterprise data and systems. The discussion explores the architectural shift from passive LLMs to active, real-time AI agents that are capable of interacting with their environments and collaborating via Agent-to-Agent (A2A) protocols. Dr. Biehl outlines how enterprises can seamlessly upgrade their existing REST API investments—specifically process and context APIs—into MCP interfaces without massive technical reengineering. Crucially, the episode highlights how organizations can overcome data safety concerns and build operational trust not by hardcoding agent limitations, but by utilizing proven API gateways and governance methodologies to enforce guardrails at scale.

In this live podcast episode, host Michael Dowden sits down with Pieter Buteneers, CTO of the legal tech workspace Emma, and Martin Stypinski, founder of the computer vision consulting firm VMG. The conversation bridges two distinct areas of artificial intelligence: highly deterministic, traditional computer vision applications and the rapidly evolving world of Large Language Models (LLMs). Together, they argue that AI is an accelerator rather than a solution in itself. The true value of any technology company lies in a deep understanding of the customer's problem and the user experience built around solving it.

In this episode, Michael Dowden sits down with Google Developer Advocate Abdellfetah Sghiouar to explore the current state and history of Kubernetes, tracing its 11-year evolution from a Google open-source project to a mature industry standard. The discussion delves into the rise of platform engineering and internal development platforms (IDPs), which aim to provide developers with higher-level abstractions that hide underlying infrastructure complexities. Looking ahead, they analyze emerging trends such as the swing between cloud and on-premise infrastructure, the growth of edge computing, and the integration of AI into development workflows.