Agents and Engineers | Agentic AI, Software & Agentic Engineering

Mariatta Wijaya is a multi-award-winning open source maintainer and contributor with over 20 years of software engineering experience. A Python core developer focused on workflow and documentation, she advises and administers Global PyLadies, co-founded PyCascades, and formerly chaired PyCon US, PyCascades, and PyLadiesCon. Dan and Mariatta discuss why unreviewed AI-generated contributions burden CPython maintainers, even as Mariatta uses AI to write code. She shares how she reviews plans and results, records decisions, and builds Claude skills to prevent agents from repeating mistakes. They also discuss her self-hosted personal apps, the limits she places on AI advice, and how she uses AI to help promote community events while keeping time for speaker mentorship. Mariatta explains why agent instructions should be useful to humans and why she packages recurring work as reusable scripts. Full episode notes Transcript Chapters (00:00) - Introduction (01:35) - AI contributions and CPython maintainers (05:45) - Experience, review, and learning to code (09:16) - Migrating off Heroku with AI (12:17) - Documenting decisions and iterating on plans (19:26) - Claude skills and repeated Django mistakes (23:19) - Reviewing software as product owner and tester (28:01) - Where Mariatta prefers human judgment (34:24) - Personal apps and alternatives to subscriptions (43:43) - AI for community organizing (50:44) - Respecting project guidelines and reviewing output (53:27) - Documentation for humans and agents (55:36) - Git skills and reusable automation scripts (59:46) - Speaker management and spreadsheet parsing ⠀ Links from the show -------------------- CPython Claude skills Django Heroku Claude Calendly Doodle Sessionize architectural decision record ⠀ Guests ------- Mariatta Wijaya, Python Core Developer Website Mastodon Bluesky LinkedIn ⠀ Follow the podcast ------------------- LinkedIn Threads Instagram TikTok ⠀ Follow Dan Gerlanc ------------------- X LinkedIn Threads Bluesky

Trevin Chow is has more than 25 years of experience in product and design, held product leadership roles at Axon, Sketchy, Nike, and Microsoft, and co-founded Chewsy. He is a core maintainer of Compound Engineering, a plugin for Claude Code, Codex, and other agent harnesses. Dan and Trevin discuss how AI changes the work of building software, especially design and product judgment. Trevin describes starting with a value proposition before generating designs, using visual references to guide exploration, and keeping humans involved in critique. His “AI sandwich” puts human attention on defining the work and polishing the result while agents handle implementation. Concrete examples include illo's character packs for consistent illustrations and Compound Engineering's reviews across model families. Trevin explains why newer models call for clearer goals and fewer prescribed steps, while extensive evaluations, changing model behavior, and outdated plugins make maintenance costly. The conversation also covers the supervised Hermes Agent setup Trevin built for his daughter, including boundaries around companionship and homework help. It closes with Jev-powered prototypes and Energon, his open source service for sharing documents and files across agents without losing control of the underlying context. Full episode notes Transcript Chapters (00:00) - Product optimism and the changing craft of software (09:53) - Design exploration with AI agents (14:24) - Start with the value proposition, then explore designs (21:21) - Agent primitives and the problem of sharing context (24:54) - Cross-model review and the AI sandwich (32:19) - illo and consistent character illustrations (40:08) - Writing skills for changing models (43:03) - Evaluations, model regressions, and maintenance costs (47:36) - Future skills and the plugin update problem (52:13) - A supervised Hermes Agent for a child (01:02:41) - Jev experiments in document analysis and book discovery (01:06:54) - Energon and context that travels between agents ⠀ Links from the show -------------------- Compound Engineering illo Energon Hermes Agent Impeccable Jev Grok bot Muse Claude Code Codex Cursor Cloudflare Workers CleanShot X Figma Daniel Lemire simdjson design system risograph ⠀ Guests ------- Trevin Chow, Product Leader and Builder Website LinkedIn X ⠀ Follow the podcast ------------------- LinkedIn Threads Instagram TikTok ⠀ Follow Dan Gerlanc ------------------- X LinkedIn Threads Bluesky

Allen Hutchison is the co-founder and CTO of Vycari, building AI agents for everyday tasks with Chris Perry. He spent 21 years at Google, most recently as a Distinguished Engineer at Google DeepMind, helped found Gemini CLI, and built the Antigravity SDK. His earlier leadership work included Google's mapping products, Core ML, and the AI Developer organization. He writes Letters from Silicon Valley.Dan and Allen discuss why consumer agents need familiar interfaces. Vycari embeds agents in mini apps, including a calendar booking tool that works with meeting preferences and buffer time. Allen wants users to think about the assistant's usefulness rather than the technology behind it.Vycari's two founders have landed well over a thousand PRs in roughly a month. Planning documents guide implementation, nightly jobs keep those documents aligned with code, and review follow-ups become issues that agents can implement. Independent reviews and reflection on user trajectories help identify problems, while Allen retains the final merge decision.Their engineering choices favor quick feedback and manageable costs. Vycari is separating microservices into their own repositories and runs Python and Postgres on VPSs. Allen explains how inexpensive models, stronger models writing their procedures, and deterministic integrations support the goal of a $5 monthly subscription.Full episode notesTranscriptChapters(00:00) - Allen's path to Gemini CLI (06:46) - Open-source contributions and approval fatigue (11:08) - XLA and the infrastructure behind machine learning (13:52) - Vycari and agents inside familiar apps (21:02) - A calendar booking tool that knows your preferences (22:34) - Two founders, a thousand PRs, and startup economics (29:29) - Planning documents and the automated development loop (37:55) - Reflection agents and nightly codebase audits (39:57) - Product context and tiers of coding agents (42:24) - CI, test coverage, and smaller repositories (48:06) - Python, Postgres, and conventional infrastructure (51:53) - Refactoring at a scale that used to be impractical (54:18) - Unexpected agent behavior and database migrations (59:19) - Inexpensive models and a $5 subscription goal (01:06:02) - The invisible work in a reliable agent harness ⠀Links from the show--------------------VycariGemini CLIClaude CodeGreptileGitHubObsidianAntigravity SDKRoborevHostingerPythonPostgresXLATheanoJAXCUDAGemini Flash LiteClaude HaikuGPT LunaGemmaGLMDeepSeekSupabaseNeonModalmicroservicescontinuous integrationfine-tuningyou ain't gonna need it⠀Guests-------Allen Hutchison, Co-Founder & CTO, VycariWebsiteVycariLinkedInGitHubX⠀Follow the podcast-------------------LinkedInThreadsInstagramTikTok⠀Follow Dan Gerlanc-------------------XLinkedInThreadsBluesky

Ben Morss is a Developer Advocate at Mistral. He previously spent 9.5 years at Google as a Product Manager for advanced APIs in Chrome and a Developer Advocate for the web. He earned a bachelor's degree in Computer Science and Music from Harvard University and a doctorate in Music Composition from the University of California at Davis. He continues to make music and write about music theory. Dan and Ben discuss how LLMs help people learn while still missing context and intent. Their discussion of information theory and training data leads into a debate about software craft when agents make code much cheaper to produce. Ben describes Chat Invaders, a cat-themed demo game, and a reunion sing-along app that broke with forty people present. He had not load-tested the app and did not determine what caused the failure. They explore why working output does not guarantee value for a business or an audience. Ben wants musical assistance that can develop ideas and handle arranging work, but finds music theory explanations unreliable. The conversation ends with the human experience behind fiction and songs, and open questions about whether models could be conscious. Full episode notes Transcript Chapters (00:00) - Learning with LLMs and the limits of explanation (05:24) - Model progress and conversational regressions (09:02) - Keeping up with AI and understanding the mechanics (11:10) - Information theory, entropy, and model compression (19:19) - Computing, music, and understanding structure (23:35) - Software craft versus speed of iteration (30:55) - Vibe coding and educational sample code (34:22) - The sing-along app that broke live (37:38) - Software demand and the power of verification (40:58) - Human judgment and useful business adoption (48:47) - AI composition and the missing musical assistant (52:38) - Music theory and gaps in training data (59:31) - Fiction, music, and the human behind the work (01:05:14) - Thinking, consciousness, and continuity ⠀ Links from the show -------------------- Mistral Genius information theory entropy lossy compression Claude Shannon Magenta Suno Markov chains WebSockets music theory augmented sixth chord stems ⠀ Guests ------- Ben Morss, Developer Advocate, Mistral Website LinkedIn ⠀ Follow the podcast ------------------- LinkedIn Threads Instagram TikTok ⠀ Follow Dan Gerlanc ------------------- X LinkedIn Threads Bluesky

Cory O'Daniel is CEO and co-founder of Massdriver, an internal developer platform and platform orchestrator for governed self-service infrastructure using Terraform, OpenTofu, and Helm. He has spent more than 20 years building teams and startups and working with cloud infrastructure. He created Bonny, an Elixir-based Kubernetes operator framework, and is a co-founder of OpenTofu.Dan and Cory discuss what infrastructure agents need to act safely. Massdriver models environments and resource relationships so agents can query context directly, while separate agent identities and attribute-based permissions constrain their access. Cory describes a staging workflow that tests infrastructure changes through real deployments before he promotes a release to production.Cory explains how a tested, consistent codebase supports his team's Claude Code workflow, why enforceable tooling matters more than long instruction files, and why human review still matters. He argues that engineers should build quality controls and deployment paths for domain experts who can now create software themselves, with collaboration and operational ownership still needing attention.Full episode notesTranscriptChapters(00:00) - Self-service infrastructure and the arrival of agents (03:11) - An API built around infrastructure context (09:03) - Agent identity and attribute-based permissions (15:54) - A four-person team, agents, and technical debt (23:12) - Future engineering roles and enforceable quality (30:20) - Backlog zero and autonomy for citizen developers (34:29) - Claude Code workflows, examples, and memory (40:54) - Testing infrastructure changes before production (48:00) - Reusable presets and shared provisioners (55:06) - Terraform is easy; organizations are complex (57:09) - Citizen developers and the future of cloud platforms (01:00:53) - When production apps run on someone's laptop (01:03:23) - Domain expertise, autonomy, and engineering's value (01:10:30) - Collaboration beyond Git and code in OCI registries ⠀Links from the show--------------------MassdriverMassdriver ArchitectThe Citizen DeveloperCory O'Daniel's personal websiteTerraformOpenTofuPostgresGraphQLKubernetesattribute-based access controltest-driven developmentdomain-driven designGhosttyCheckovWizSnykGitOpsOCIElixirGolang⠀Guests-------Cory O'Daniel, CEO & Co-Founder, MassdriverWebsiteLinkedIn⠀Follow the podcast-------------------LinkedInThreadsInstagramTikTok⠀Follow Dan Gerlanc-------------------XLinkedInThreadsBluesky

Stephanie Jarmak is an AI Engineer at Omni working on multi-agent orchestration and code intelligence. She is a research affiliate for NASA's Science Explorer (SciX) and a maintainer of Gas City, an open-source orchestration builder for multi-agent coding workflows. Dan and Stephanie discuss how she became a Gas City maintainer by learning from review agents, encoding contribution standards in skills, and building an issue-triage workflow. Automating those steps exposed new reliability problems, from failures across dependencies to excessive alerts and agents that overreact to hypothetical risks. Their conversation connects these experiments to Stephanie's move from planetary science into search and AI engineering. They examine the appeal of building with agents, the responsibility for maintaining what those agents produce, and the mixed emotions of seeing models solve once-difficult research tasks. Personal game projects and Omni's approach to shaping her role show how her curiosity carries across both work and home. Full episode notes Transcript Chapters (00:00) - Contributing to open source with coding agents (08:51) - Learning the review standards and becoming a maintainer (12:58) - Automating a Gas City workflow (21:02) - Astronomy, computers, and a different path into development (27:40) - Working styles and responsibility for agent output (34:46) - Production responsibility and AI skepticism (40:12) - Human judgment and agents solving science tasks (45:55) - Pre-mortems, silent failures, and too many alerts (48:34) - Slack as an agent dashboard and shareable Gas City packs (53:45) - Building games and creative projects with the kids (59:06) - From planetary science to search and Sourcegraph (01:09:48) - Omni and a role shaped around the person (01:15:03) - Decision models and making automation easier ⠀ Links from the show -------------------- Gas City Beads Gas Town Dolt Steve Yegge Claude Code Codex The Six Types of Working Genius Terminal-Bench Science Harbor Slack Suno Forge Scryfall 17Lands SciX Sourcegraph Amp Omni information retrieval pre-mortem reinforcement learning from human feedback Jev ⠀ Guests ------- Stephanie Jarmak, AI Engineer, Omni Website LinkedIn ⠀ Follow the podcast ------------------- LinkedIn Threads Instagram TikTok ⠀ Follow Dan Gerlanc ------------------- X LinkedIn Threads Bluesky

Eric Ma is a Senior Principal Data Scientist at Moderna, where he leads the Data Science and Artificial Intelligence (Research) team. Previously, he conducted biomedical data science research at the Novartis Institutes for Biomedical Research and earned his doctorate in Biological Engineering from MIT. Eric is also an open-source software developer known for leading pyjanitor and nxviz and contributing to NetworkX and PyMC. Dan and Eric discuss Eric's rapidly changing practice of agentic data science, centered on Marimo Pair and reproducible Python notebooks. Eric demonstrates an analysis of protein activity and stereoselectivity data, explains why canonical sources and inline environment metadata matter, and shows how agents make ambitious custom visualizations more approachable even when a live demo fails. The conversation then turns to design documents, testable specifications, repository conventions, and the attention-management practices Eric uses to keep agent work from becoming a context-switching trap. Full episode notes Transcript Chapters (00:00) - Why Eric's data science workflow changed so quickly (05:49) - Marimo Pair puts agents inside the notebook (07:19) - Reproducing a protein-engineering paper (10:45) - Why enzyme stereoselectivity matters (13:20) - Canonical data sources and portable environments (18:08) - Exploring activity and selectivity with interactive plots (27:47) - Finding promising regions for protein engineering (30:24) - From heat maps to a 3D protein viewer (35:54) - Learning from agent traces and preserving intent (47:13) - Opinionated structure for agent-written code (52:25) - Managing attention across agentic tasks ⠀ Links from the show -------------------- Marimo Marimo Pair OpenCode cmux Machine-Directed Evolution of an Imine Reductase for Activity and Stereoselectivity cloudscraper Polars Plotly PEP 723 uv 3Dmol.js anywidget Protein Data Bank PyMOL The Arrow of Intent Hermes Agent Obsidian ⠀ Guests ------- Eric Ma, Senior Principal Data Scientist, Moderna Website LinkedIn GitHub X ⠀ Follow the podcast ------------------- LinkedIn Threads Instagram TikTok ⠀ Follow Dan Gerlanc ------------------- X LinkedIn Threads Bluesky

Andrew Zigler is a GTM Engineer at LinearB and the host of Dev Interrupted, a twice-weekly podcast and newsletter about AI-native development and agentic orchestration. A classicist by training, he previously taught in Japan, built e-learning platforms, and worked in developer relations at Mattermost. Dan and Andrew discuss how shared workflows and memory can help turn individual AI gains into team improvements. Andrew explains his Mise en Place planning methodology and how he uses Beads to translate ideas into connected tasks, while keeping human collaboration in tools such as Asana and Confluence. They examine why clear specs reduce rework without removing the need for iteration, how CI/CD and code review may adapt to agentic development, and what Andrew’s personal agent setup has changed about his work and learning. Full episode notes Transcript Chapters (00:00) - Introduction and the gap between individual and team gains (06:15) - Open source when agents become the users (10:09) - A personal agent stack built on simple primitives (17:37) - Mise en Place and planning before coding (21:58) - Connecting Beads to team workflows (26:10) - Why task graphs help agents maintain context (35:57) - Shared specs and human alignment (43:26) - Agile iteration with faster prototypes (44:52) - CI/CD and independent code review (52:12) - Humanities, learning, and managing agents (58:00) - Changing skills and software interfaces (01:02:25) - Fixing feedback loops and sharing the stack ⠀ Links from the show -------------------- Beads Beads Rust Mise en Place Andrew’s dotfiles Dev Interrupted LinearB Tailscale Agent Gateway systemd WireGuard DuckDB Wispr Flow Dolt Gas Town Steve Yegge Jeffrey Emanuel Robots Ate My Homework NumPy pandas The Diamond Age Snow Crash agile continuous integration progressive web app ⠀ Guests ------- Andrew Zigler, GTM Engineer, LinearB; Host of Dev Interrupted Website LinkedIn X GitHub ⠀ Follow the podcast ------------------- LinkedIn Threads Instagram TikTok ⠀ Follow Dan Gerlanc ------------------- X LinkedIn Threads Bluesky

Julien Verlaguet is Founder and CEO of SkipLabs, the company building Skipper, a closed-loop coding agent. He created Skip, a reactive programming language, and led the design of Hack, the language Meta developed to run its code base at scale. In this episode, we discuss what the job of programming language designers becomes when AI agents write most of the code. This includes the design decisions in the language and tooling when agents are the principal user. We also discuss: - How Julien refocused SkipLabs towards AI - What it would take for AI to be able to fully replace engineers - The use of formal methods with LLMs Full episode notes Transcript Chapters (00:00) - A closed-loop coding agent (04:30) - Reactive programming and technical debt (08:40) - Why SkipLabs moved toward AI tooling (11:07) - Incremental tools and low-latency feedback (14:20) - Reactive collections versus Buck and Make (21:06) - Language design in the age of LLMs (31:01) - When human coding still matters (34:16) - Working with four to six agents (41:04) - Why code quality still needs judgment (49:48) - The hidden cost of AI-generated tests (56:24) - The junior engineer dilemma (01:04:27) - Formal methods and LLM verification (01:13:35) - A bumpy road toward a brighter future ⠀ Links from the show -------------------- Skipper SkipLabs Skip Hack reactive programming incremental computation Buck SQLite Coq Rocq Lean Curry-Howard correspondence formal methods model checking Emacs Bun ⠀ Guests ------- Julien Verlaguet, Founder and CEO, SkipLabs Website LinkedIn ⠀ Follow the podcast ------------------- LinkedIn Threads Instagram TikTok ⠀ Follow Dan Gerlanc ------------------- X LinkedIn Threads Bluesky

Thomas Wiecki, Founder, PyMC Labs, is the author of PyMC, one of the most popular frameworks for Bayesian modeling, and the founder of PyMC Labs, a Bayesian AI consultancy. He holds a PhD in Computational Cognitive Neuroscience from Brown University and previously served as VP of Data Science and Head of Research at Quantopian Inc., where he built and led a team of data scientists to develop a hedge fund leveraging insights from a crowd-sourced pool of 300,000 researchers.In this episode, we explore how AI agents have made Bayesian methods more accessible to non-experts. Bayesian models are useful for both forecasting and decision-making under uncertainty. These models can act as scenario simulators, making it possible to ask what-if questions about marketing, seasonality, or health interventions while exposing the mechanisms behind an outcome.We also discuss how agentic engineering has finally helped deliver on some of the original promises of data science, how to effectively embedding agents in tools like Slack or Discord to democratize access to company information, and what parts of data science still require talking with humans to understand the problem.Full episode notesTranscriptChapters(00:00) - Agentic engineering meets Bayesian decision science (05:27) - Bayesian models versus black-box prediction (10:22) - Confounders and causal mechanisms (12:41) - From notebooks to model-aware agents (15:41) - The rise of agentic interfaces (19:18) - Alchemize and verified code translation (29:49) - Agentic data science beyond the silo (34:25) - Daimon and collaborative multiplayer AI (37:43) - Verification and unsettled methods (42:53) - Parallel agents and PyMC Forecast (51:08) - The Bayesian method of software engineering (53:59) - Agents as software's primary users (57:08) - Institutional knowledge and bus-factor risk (01:03:56) - Faster research and open source (01:12:05) - Causal models for AI's future ⠀Links from the show--------------------Master Agentic Data ScienceShow Us Your Agent SkillsPyMCPyMC LabsBayesian modelingcausal questionsmedia mix modelingDaimonAlchemizePyMC Forecastprior data fitted networksNumPyro ForecastStanJAXPyTorch⠀Guests-------Thomas Wiecki, Founder, PyMC LabsWebsiteLinkedInGitHubTwitter⠀Follow the podcast-------------------LinkedInThreadsInstagramTikTok⠀Follow Dan Gerlanc-------------------XLinkedInThreadsBluesky