Beyond Coding

Carlos Kelkboom, a lead architect and software engineer focused on compilers and deterministic design, lays out his rules for coding agents. They explore why AI excels at one-shot builds but falters during iteration, how scaffolding and Git worktrees shape better architecture, and why his main agent never edits files. The conversation also tackles heap-free languages, fading collaboration, progressive context, and throwing away code you cannot understand.

Mat Jones, a Staff Engineer at Adyen focused on platform engineering and distributed systems, explores coding agents with a view far beyond the repo. They discuss coordinating multiple agents, Architect’s nine connected graphs, processing seven million trace spans per second, tracing changes through payment flows, and Sentinel agents that turn anti-patterns into merge requests. They also tackle security risks, tool bloat, and smarter ways to start.

Dexter Horthy, CEO of HumanLayer and the engineer who coined “context engineering,” examines why AI coding agents can quietly worsen codebases. He introduces SlopCodeBench, challenges SWE-bench hype, and explores dark factories, token maxing, and realistic 2–4x productivity. Horthy also discusses design docs, engineering taste, pull requests, janitor bots, overnight agents, and focused parallelism.

Dario Hamidi, an AI engineer at Amp focused on agentic development, explores a radical alternative to pull requests and feature branches. He discusses permission design, trustworthy environments, and cloud-agent failure modes. Dario also introduces Orbs, persistent remote workspaces for parallel and multiplayer development, plus strategies for triaging 60+ bug reports daily without overwhelming engineers.

Tech layoffs collide with thousands of unfilled engineering roles, while AI reshapes interviews and hiring filters. Recruiters and engineers explore why referrals, personalized outreach, GitHub, and open source can beat polished resumes. They also discuss the future of junior developers, the soft skills that sustain careers, practical AI expectations, algorithm interviews, and growing through scope rather than titles.

Jordan Tigani, a database expert and entrepreneur who helped create Google BigQuery, shares how getting fired led to MotherDuck. He discusses partnering with DuckDB’s creators, building a paid product from scratch, and using AI for dashboards, pipelines, and data exploration. They also debate whether dashboards are obsolete, why engineers should look beyond query optimization, and how writing and product sense can shape a career.

Brian McClendon, the technology executive who helped create Google Earth and led Google Maps, discusses Niantic Spatial’s ambitious 4D model of the world. He explores visual positioning, Gaussian splats, robot training, and turning research into production in months. He also shares what he looks for in engineers and how to use AI with meaningful self-checks.

Mallika Rao, an engineering leader from big tech who levels up engineers to staff roles, discusses how judgment now matters more than raw coding chops. She explains rehearsing multiple futures, learning from incident case studies, cognitive coordination and meeting load, team design for AI reliability, and when to go deep versus broad with agents.

Michael Giannangeli, Head of Product for Agentic AI and Amazon Nova at Amazon, leads model training, evals, and RL gyms. He talks about building evals from real failures. He explains RL gyms that train models in real environments. He covers model routing, cost-per-token trade-offs, and when benchmarks get retired.

Wes Bos, front-end developer and educator known for JavaScript and React tutorials, talks about why code alone no longer proves skill. He discusses generative UI limits, iterative agent workflows, the tradeoffs of cheaper models versus local hardware, and why systems thinking and personal branding matter in an AI-assisted world.