The Weekly Dev's Brew

The Weekly Dev's Brew

By Jan-Niklas WortmannNewsTechnologyEducationTech News
Download on the App Store

All The Weekly Dev's Brew episodes

  • Why AI agents belong in the sandbox | Darren Shepherd
    Why buy a sandbox when you can just launch a container? Darren Shepherd spent years at Rancher building on Kubernetes, so sandboxes looked dumb to him. Then he built his own coding agent to run 15 things in parallel without work-tree chaos, and the most interesting part of it turned out to be a sandboxing system. He had proved himself wrong.
    The sharp part is where the sandbox boundary goes. The common pattern keeps the agent loop outside and sandboxes only its tool calls. Darren calls that completely flawed, because the loop itself directs code that holds secrets and talks to external systems. His answer: put the agent and its tools in one sandbox, with one policy for secrets and egress.
    Darren is on X and GitHub at @ibuildthecloud. https://obot.ai/
    Where does your boundary sit: around the agent itself, or only around the tools it calls?

    Why buy a sandbox when you can just launch a container? Darren Shepherd spent years at Rancher building on Kubernetes, so sandboxes looked dumb to him. Then he built his own coding agent to run 15 things in parallel without work-tree chaos, and the most interesting part of it turned out to be a sandboxing system. He had proved himself wrong.

    The sharp part is where the sandbox boundary goes. The common pattern keeps the agent loop outside and sandboxes only its tool calls. Darren calls that completely flawed, because the loop itself directs code that holds secrets and talks to external systems. His answer: put the agent and its tools in one sandbox, with one policy for secrets and egress.
    Darren is on X and GitHub at @ibuildthecloud. https://obot.ai/
    Where does your boundary sit: around the agent itself, or only around the tools it calls?

    Show notes, transcript, and links: https://www.wordman.dev/podcast/darren-shepherd-ai-agent-sandboxes/

    Key Takeaways
    • A container is not a sandbox. Sandboxes sit one layer up: create, clone, copy files in and out, on a base image that mostly just carries language runtimes.
    • Parallel agents need isolation. Running many tasks at once without Git work-tree chaos means giving each task its own sandbox.
    • Egress is the policy surface, not ingress. Agents sit on the consuming side of services, so outbound traffic is what you control.
    • Keep real secrets out of reach. Hand the agent a fake key with the right shape and swap in the real one at a transparent proxy, where policy and exfiltration checks live.
    • Sandbox the agent loop, not only the tool calls. The loop directs code that holds secrets and talks to external systems, so the agent and its tools belong in one sandbox with one policy.
    • Client-side agent loops beat centralized ones. Server-side provider tools such as web search open exfiltration paths that are hard to reason about.
    • MCP's lasting value is the interface, not the protocol: a contract designed for AI, which is why it fits the enterprise.
    • Output is not progress. Letting a model barf out thousands of lines feels productive until the regressions pile up, and one estimate raised in the conversation puts a skilled engineer's real gain at around 5 to 10 percent.
    • Chapters
      • 00:00:00 Why sandboxes felt dumb
      • 00:01:38 Building a coding agent
      • 00:02:36 Parallelism and work-tree chaos
      • 00:03:25 What a sandbox actually is
      • 00:04:27 Files, secrets, and egress
      • 00:08:20 Tool-call-only sandboxing is flawed
      • 00:12:28 Put the agent in the sandbox
      • 00:13:31 Client-side vs centralized loops
      • 00:14:51 GPTScript, Clio, and MCP
      • 00:20:00 MCP, OAuth, and staying in lane
      • 00:30:01 Without fundamentals, AI gets dangerous
      • 00:35:00 The 100x developer myth
      • 00:40:00 AI is not a compiler
      • 00:45:01 Distributed systems instincts
      • 00:51:09 ADRs and keeping quality
      • 00:55:04 Regression holes with agents
      • 00:57:19 Model personalities and greenfield work
      • 01:01:50 Skills as behavioral scripts
      • 01:05:16 Plan-first and spec-driven development
      • 01:12:30 Claude Code and tooling shifts
      • 01:21:16 Wrap and finding joy with AI
      • Mentioned
        • Obot AI
        • Obot on GitHub
        • Model Context Protocol
        • GPTScript (archived)
        • Clio (archived)
        • Rancher
        • Kubernetes
        • Claude Code
        • Codex CLI
        • Architecture Decision Records
        • EP. 22 with Dillon Mulroy
        • 1 hr 34 min
        • Why AI Coding Made Me More Productive - And Less Happy | Dillon Mulroy
          Why did AI coding make Dillon Mulroy more productive and less happy? He stopped writing most of his own code after Claude Opus 4.5. Six months later he ships more than he ever has, still reads every line, and enjoys the work less. The small implementation hits that used to create flow are gone. The day is now one hard design problem after another, and that is more exhausting.
          Dillon is a principal engineer at Cloudflare. This episode answers what Cloudflare's May layoff of about 1,100 people looked like from inside, why lab-style agent loops are a poor default for most teams, and how he keeps a model on a short leash with Ghostty, Herdr, Pi, and Plannotator. Specs are TypeScript types, call stacks, and tests. Pull requests stay in the 300 to 800 line range. He uses Pi's slash tree instead of sub-agents.
          Dillon is on X at @dillon_mulroy, GitHub at @dmmulroy, and Twitch at @dillon. https://dillonis.online

          Why did AI coding make Dillon Mulroy more productive and less happy? He stopped writing most of his own code after Claude Opus 4.5. Six months later he ships more than he ever has, still reads every line, and enjoys the work less. The small implementation hits that used to create flow are gone. The day is now one hard design problem after another, and that is more exhausting.

          Dillon is a principal engineer at Cloudflare. This episode answers what Cloudflare's May layoff of about 1,100 people looked like from inside, why lab-style agent loops are a poor default for most teams, and how he keeps a model on a short leash with Ghostty, Herdr, Pi, and Plannotator. Specs are TypeScript types, call stacks, and tests. Pull requests stay in the 300 to 800 line range. He uses Pi's slash tree instead of sub-agents.
          Dillon is on X at @dillon_mulroy, GitHub at @dmmulroy, and Twitch at @dillon. https://dillonis.online

          Show notes, transcript, and links: https://www.wordman.dev/podcast/dillon-mulroy-i-enjoy-coding-less-than-ever/

          Key Takeaways
          • Dillon Mulroy has not written most of his own code since Claude Opus 4.5, from about mid-December 2025 or mid-January 2026. He still reads every line.
          • He is more productive and less happy because the small implementation hits that created flow are gone. The day is one hard design problem after another.
          • A single human still owns the output. Greater reach in less time means more review, not less.
          • Training juniors is the open problem: the struggle that built architecture intuition is easy to skip with a prompt.
          • Cloudflare's May layoff cut about 1,100 people, roughly 20% of the workforce. In his view it was mostly an AI-driven change in which roles exist. They opened different roles the next day.
          • Lab-style agent loops are impractical for a median developer at today's prices. He wants receipts before treating them as the default.
          • Anthropic Fable is a non-starter for a company like Cloudflare, he says, because it does not ship with zero data retention and can drop a session onto Opus 4.8.
          • His stack is Ghostty, Herdr, Pi, and Plannotator. Specs are TypeScript types, call stacks, and tests. Pull requests stay in the 300 to 800 line range. He uses slash tree instead of sub-agents.
          • Chapters
            • 00:00:00 I have not written my own code
            • 00:01:18 From bearish to all-in
            • 00:06:55 Tools do not define the work
            • 00:10:02 I enjoy this work less
            • 00:14:16 Macro problems, no flow state
            • 00:17:38 Feeling junior again
            • 00:19:46 Training the next generation
            • 00:23:34 Cloudflare cut 1,100 people
            • 00:26:12 Roles are compressing
            • 00:34:55 Loops are not for the median dev
            • 00:41:50 Why Fable is a non-starter
            • 00:44:05 Getting the model to write your code
            • 00:46:20 Herdr, Pi, and Plannotator
            • 00:51:20 Specs that look like code
            • 00:52:44 Slash tree instead of sub-agents
            • 01:02:00 A queue he can see coming
            • Mentioned
              • Dillon Mulroy
              • The Next Token
              • Not Holding Back the Ocean (Ethan Niser)
              • The Coming Loop (Armin Ronacher)
              • Cloudflare Artifacts
              • Ghostty
              • Herdr
              • Pi
              • Plannotator
              • Graphite
              • 1 hr 5 min
              • Dex Horthy - What Actually Gets You 2-3x With AI Coding
                You can't get 10x with AI coding. Not today. Dex Horthy (Human Layer) makes the case for 2-3x if you refuse to ship slop, and why a perfect spec still produces garbage if you give the model permission to skip quality. That permission is how you burn a codebase into ash.
                We get into context engineering, where AI coding actually fails, why PR review is the bottleneck, how to use specs without worshipping them, cheap models vs Opus, compaction and sub-agents, and what to tell a 20-year engineer who is not living on Twitter.
                Dex is the founder of Human Layer. https://humanlayer.dev

                You can't get 10x with AI coding. Not today. Dex Horthy (Human Layer) makes the case for 2-3x if you refuse to ship slop, and why a perfect spec still produces garbage if you give the model permission to skip quality. That permission is how you burn a codebase into ash.

                We get into context engineering, where AI coding actually fails, why PR review is the bottleneck, how to use specs without worshipping them, cheap models vs Opus, compaction and sub-agents, and what to tell a 20-year engineer who is not living on Twitter.
                Dex is the founder of Human Layer. https://humanlayer.dev

                Show notes, transcript, and links: https://www.wordman.dev/podcast/dex-horthy-what-actually-gets-you-2-3x-with-ai-coding/

                Key Takeaways
                • Dex's ceiling for production work is 2-3x, not 10x: 99% of human-quality code, written as if you had typed every character, just faster.
                • A perfect spec still produces garbage if you give the model permission to skip quality.
                • PR review is the bottleneck. Spend the human time on design and breakdown so review is 20 minutes, not six hours.
                • A spec should give leverage so you can re-steer a short markdown file instead of thousands of lines of code.
                • Use the smartest model until it works, then make it cheaper. Kill a session when the model starts flailing on tests.
                • Vibe-coding a side project and keeping a 10-year enterprise system alive share almost no constraints.
                • Chapters
                  • 00:00:00 You can't get 10x (not today)
                  • 00:07:41 Is software engineering disappearing?
                  • 00:13:50 Where AI coding actually fails
                  • 00:19:21 Review is the bottleneck
                  • 00:24:34 The new software factory
                  • 00:29:12 Specs that give leverage
                  • 00:43:02 Stop chasing engineering metrics
                  • 00:48:13 Cheap models and shipping slop
                  • 01:01:15 Compaction, context windows, sub-agents
                  • 01:14:01 Vibe coding vs production loops
                  • 01:21:24 What juniors still need to learn
                  • 01:28:02 What Human Layer is building
                  • 01:36:50 Advice for working engineers
                  • Mentioned
                    • HumanLayer
                    • Dex Horthy on X
                    • 12-Factor Agents
                    • The Goal by Eli Goldratt
                    • Pi (Mario Zechner)
                    • Addy Osmani on Loop Engineering
                    • 1 hr 40 min
                    • How To Ship Real Code With AI (Not Junk) ft. David Cramer
                      Everyone is selling 100x developer productivity, but David Cramer runs production software at Sentry and says LLMs are not making his own work faster. We talk about what actually holds up when AI meets real codebases, security review, open source economics, and why verification beats generation.

                      Everyone is selling 100x developer productivity, but David Cramer runs production software at Sentry and says LLMs are not making his own work faster. We talk about what actually holds up when AI meets real codebases, security review, open source economics, and why verification beats generation.

                      Show notes, transcript, and links: https://www.wordman.dev/podcast/david-cramer-ship-real-production-code-with-ai/

                      Key Takeaways
                      • The 100x developer narrative breaks down once you optimize for quality, security, and maintainability instead of raw output.
                      • David built Warden inside Sentry and used it to find more than 100 previously unknown vulnerabilities in production code, including auth bypasses.
                      • Search and internal knowledge tools may be the highest-leverage LLM use case inside a company today, not autonomous code generation.
                      • MCP is useful when it gives agents structured access to real workflows, not when it is treated as a thin API wrapper.
                      • Open source sustainability and AI training economics are colliding, and the financial models behind many AI companies do not add up.
                      • Chapters
                        • 00:00:00 Cold open
                        • 00:01:51 Budgeting the AI hype
                        • 00:07:00 No 100X developers
                        • 00:14:25 Where LLMs actually help
                        • 00:35:29 Vibe coding limits
                        • 00:48:26 Open core is not open source
                        • 01:12:29 Bubble and economics
                        • Mentioned
                          • Sentry
                          • Warden
                          • Finding Vulnerabilities with Warden
                          • Skill Synthesis
                          • Agent Skills
                          • Open Source Pledge
                          • David Cramer on X
                          • 1 hr 17 min
                          • Faster Isn't Better — Mario Zechner on Everything AI Made Us Forget (creator of Pi)

                            Mario Zechner has watched people generate 500,000 lines of code in a week with a swarm of agents. He'll tell you exactly how that ends.Mario is the creator of Pi, the minimal, self-modifying coding agent that took off after Claude Code stopped fitting his workflow. He now builds it alongside Armin Ronacher at Earendil. This one is a grumpy, honest conversation about where AI coding is actually heading, and the places the hype quietly falls apart.We get into:- Why "code is never free," and how a lot of today's productivity just delays the bill- The case that spec-driven development is hyper-waterfall, repeating a mistake the industry thought it solved 30 years ago- Why an army of parallel agents wrecks his brain, and the "absolute caveman" workflow he actually uses to ship- Running AI fully local on a normal MacBook, and why that future is closer than most people think- The clanker problem- What we lose when AI strips out the friction we used to learn fromThe tension we kept circling: agents make exploring solutions faster than ever, but the thinking was always the hard part, not the typing. 0:00 Intro2:20 Why Pi Fits Workflows8:53 What Success Looks Like for Pi10:58 Building Beyond the Coding Agent13:13 Local AI Is Getting Real20:13 How Mario Actually Works26:38 Discipline, Atrophy, and Juniors29:08 Spec-Driven Dev Is Just Hyper-Waterfall35:03 Code Isn't Actually Free37:56 Async Agents and Thinking Time44:42 Learning Without the Pain49:13 AI's Sloppy Software Wave52:11 GitHub Under Clanker Load53:31 Family, Work, and Balance57:05 The Pi Team and Leadership58:52 Refactoring Pi's Core1:03:32 Security, YOLO, and Trust1:06:27 Taming the PR Flood1:12:11 Token Prices and Budgets

                            1 hr 18 min
                          • A Google Engineer's Honest Take on AI, Angular, and the Future of Your Career

                            In this episode, I sit down with Minko Gechev — Google Engineer, former Angular team lead, and one of the most thoughtful voices on AI and developer productivity.We get into why AI won't replace engineers anytime soon (and what that actually means for your career), how Google's Angular team is affected by AI, the security risks of the current AI coding wave, and the single most valuable skill developers should be building right now.If you're a developer trying to figure out where you fit in the age of AI, I hope this one helps you.Connect with Minko:X/Twitter: https://twitter.com/mgechevGitHub: https://github.com/mgechevLinkedIn: https://linkedin.com/in/mgechev


                            Topics covered:

                            AI-powered developer productivity • LLM skill evaluation • Angular + AI • Software engineering career • Imposter syndrome • Open source security • Future of coding

                            55 min
                          • Microfrontends: Cutting Through the Hype and Misconceptions (w/ Luca)

                            Micro-frontends promise to help teams scale development, but AWS Principal Architect Luca Mezzalira has seen some succeed and some just fall apart. In this conversation, he shares what he's learned from years of implementation work at companies like DAZN and through consulting with Fortune 500 organizations.
                            Luca explains why micro-frontends are fundamentally different from components—something many teams get wrong from the start. He discusses the common pitfalls he encounters: teams treating them like distributed components, sharing state across boundaries, and copying architectures from companies with completely different constraints and resources.
                            The discussion covers practical implementation details—from communication patterns and tooling strategies to when approaches like iframes actually make sense. Luca also touches on newer developments in server-side rendering and how they're changing the micro-frontend landscape.
                            This isn't about micro-frontends being good or bad, but about understanding when and how to use them appropriately. Luca emphasizes the importance of context and organizational design in making these architectural decisions work.

                            Our Fantastic Guest - Luca Mezzalira
                            I’m a software architect with over 20 years of experience helping organizations—from startups to Fortune 500s—design scalable, resilient, and modern architectures. My focus: evolving frontend systems, driving cloud-native modernizations, and enabling autonomy at scale through architectural clarity.
                            As Principal Serverless Specialist Solutions Architect at AWS, I advise global enterprises on designing and implementing efficient, event-driven systems. I specialize in serverless, micro-frontends, and distributed architectures that align technology strategy with long-term business outcomes.

                            LinkedIn
                            X/Twitter

                            Links and Resources
                            Luca's Newsletter 
                            Single-spa 
                            Module Federation 
                            Native Federation 
                            Web Fragments (Microsoft) 
                            Astro Server Islands 

                            Thank you very much for listening!

                            We are also pretty much on all social media platforms, so make sure to like and subscribe!

                            Homepage - https://www.weeklybrew.dev/

                            BlueSky - https://bsky.app/profile/weeklybrew.dev
                            Instagram - https://www.instagram.com/weeklydevsbrew/
                            TikTok - https://www.tiktok.com/@weeklybrew.dev
                            YouTube - https://www.youtube.com/@theweeklydevsbrew
                            LinkedIn - https://linkedin.com/company/the-weekly-dev-s-brew

                            1 hr 12 min
                          • Why CSS is Now the Fastest-Moving Space in Web Development (with Una Kravets)

                            Why CSS is Now the Fastest-Moving Space in Web Development (with Una Kravets)


                            While everyone's talking about AI disrupting development, Una Kravets from Google Chrome reveals a surprising twist: CSS has quietly become the fastest-evolving part of web development. In this conversation, Una breaks down how features that required months of JavaScript engineering are now landing as native platform capabilities. Think customizable dropdowns, anchor positioning, and scroll-driven animations. She shares insights from working directly with Chromium engineers and explains why senior developer expertise is becoming more valuable than ever, even as AI lowers the barrier to building applications.
                            Una walks through her process for identifying platform gaps and working with standards bodies like Open UI. She also tackles the unique challenge of scaling web platform knowledge in an AI-driven development landscape, sharing her mixed feelings about AI's current applications. From her success building Chrome extensions with Gemini to her frustration with chat-based customer service, Una argues for using AI where it makes sense while maintaining the human elements that make the web engaging. Whether you're skeptical about AI's role in development or curious about the cutting-edge of CSS, this conversation offers a grounded yet forward-thinking perspective on the future of web development.


                            Our Fantastic Guest
                            Una Kravets
                            Una Kravets leads the Web UI Developer Relations Team at Google Chrome with a mission to make the web platform easier to build on and more powerful. She hosts the CSS Podcast and has spoken at over 100 conferences around the world helping folks build better web interfaces. When Una isn't online, she loves to craft and recently became a mom.

                            X
                            Bluesky
                            Website


                            Links and Resources
                            Emil Kowalski's React motion course animations.dev
                            Google's web development guidance and feature announcements
                            Open UI Community Group
                            CSS Working Group GitHub
                            Web Platform Tests

                            Thank you very much for listening!

                            We are also pretty much on all social media platforms, so make sure to like and subscribe!

                            Homepage - https://www.weeklybrew.dev/

                            BlueSky - https://bsky.app/profile/weeklybrew.dev
                            Instagram - https://www.instagram.com/weeklydevsbrew/
                            TikTok - https://www.tiktok.com/@weeklybrew.dev
                            YouTube - https://www.youtube.com/@theweeklydevsbrew
                            LinkedIn - https://linkedin.com/company/the-weekly-dev-s-brew

                            53 min
                          • Writing a Web Browser in 2025 (with Andreas Kling)

                            What does it take to build a web browser when everyone says it's impossible? In this episode, we sit down with Andreas Kling, the engineer behind Ladybird—the only major browser project that doesn't take money from Google.
                            Andreas breaks down a uncomfortable truth: Google funds every major browser through search deals. Chrome, Firefox, Safari—they're all dependent on Google's advertising revenue. Ladybird is building the first truly independent alternative since the early Firefox days.
                            We dive deep into the technical challenges of implementing web standards from scratch, why their 700,000 lines of code can compete with Chrome's 100+ million, and how they're making browser code that actually mirrors the specifications. Andreas reveals why they switched from UTF-8 to UTF-16, why they didn't choose Rust, and how they handle the constant evolution of living web standards.
                            From the "draw the owl" problem of CSS specifications to building a sustainable nonprofit model with sponsors like Shopify, Andreas shares the engineering and business decisions behind their ambitious timeline: alpha in 2026, beta in 2027, and v1.0 by 2028.

                            Our Fantastic Guest
                            Andreas Kling
                            President of the Ladybird Browser Initiative.

                            X

                            Links and Resources
                            Ladybird Browser Website
                            Web Platform Tests
                            Fil-C (memory-safe C++ compiler)

                            Thank you very much for listening!

                            We are also pretty much on all social media platforms, so make sure to like and subscribe!

                            Homepage - https://www.weeklybrew.dev/

                            BlueSky - https://bsky.app/profile/weeklybrew.dev
                            Instagram - https://www.instagram.com/weeklydevsbrew/
                            TikTok - https://www.tiktok.com/@weeklybrew.dev
                            YouTube - https://www.youtube.com/@theweeklydevsbrew
                            LinkedIn - https://linkedin.com/company/the-weekly-dev-s-brew

                            1 hr 8 min
                          • From Discord Helper to OSS Maintainer (w/ Dominik Dorfmeister)

                            What does it take to accidentally become the maintainer of one of React's most popular libraries? In this episode, we sit down with Dominik Dorfmeister (aka. TkDodo), maintainer of React Query (now TanStack Query), to unpack his unexpected journey from answering Discord questions during COVID to maintaining a library with 10 million weekly downloads.
                            Dominik reveals how spending months just helping people on Discord led to becoming a core maintainer of the TanStack ecosystem. We dive deep into React Query's stability-first approach, the challenges of managing breaking changes across multiple frameworks, and why the React adapter is surprisingly the most complex to maintain.
                            From deleting 20,000 lines of dead code at Sentry to building platform teams that enable developer productivity, Dominik shares insights into maintaining software at scale.

                            Our Fantastic Guest
                            Dominik Dorfmeister
                            Software Engineer from 🇦🇹, working at Sentry, maintaining TanStack Query

                            BlueSky


                            Links and Resources
                            Dominik's Blog
                            Query.gg
                            Knip Docs 

                            Thank you very much for listening!

                            We are also pretty much on all social media platforms, so make sure to like and subscribe!

                            Homepage - https://www.weeklybrew.dev/

                            BlueSky - https://bsky.app/profile/weeklybrew.dev
                            Instagram - https://www.instagram.com/weeklydevsbrew/
                            TikTok - https://www.tiktok.com/@weeklybrew.dev
                            YouTube - https://www.youtube.com/@theweeklydevsbrew
                            LinkedIn - https://linkedin.com/company/the-weekly-dev-s-brew

                            54 min