What if the best way to get good work out of AI is to stop being nice to it? Adam and Tim have both landed on the same uncomfortable discovery: when you pit AI agents against each other, with fake points, opposing incentives, and competing models, the output gets dramatically better than anything a single polite prompt can produce.

Adam's bug-hunting pipeline hands fake rewards to sycophantic agents and then throws the scores in the trash. Tim made Claude and ChatGPT argue for twelve rounds straight until they both said "ship it".

Follow the show and be sure to join the discussion on Discord! Our website is workingcode.dev and we're @workingcode.dev on Bluesky. New episodes drop weekly on Thursday.

And, if you're feeling the love, support us on Patreon.

With audio editing and engineering by ZCross Media.

Full show notes and transcript here.

Tim spent a single Sunday afternoon with Claude and built Show Bot -- a sarcastic Discord bot trained on every Working Code transcript, complete with a Dungeon Crawler Carl personality, fallacy detection badges, and a talent for roasting anyone who tries to prompt-inject it. The conversation turns into a deep technical walkthrough of RAG pipelines, local models, cross-encoder reranking, and what happens when you just start building things that make you laugh.

Follow the show and be sure to join the discussion on Discord! Our website is workingcode.dev and we're @workingcode.dev on Bluesky. New episodes drop weekly on Thursday.

And, if you're feeling the love, support us on Patreon.

With audio editing and engineering by ZCross Media.

Full show notes and transcript here.

Testing sounds simple until you actually try it. Private methods that can't be reached without hacks. Dependency injection that doubles your architecture's complexity before a single assertion runs. Production code that slowly warps around your test suite instead of the other way around. Ben has spent his entire career shipping code without tests, and this week he decided to change that. The crew walks him through every trap he steps on, and a few they've been stuck in themselves.

• Ben Nadel's Blog

Follow the show and be sure to join the discussion on Discord! Our website is workingcode.dev and we're @workingcode.dev on Bluesky. New episodes drop weekly on Thursday.

And, if you're feeling the love, support us on Patreon.

With audio editing and engineering by ZCross Media.

Full show notes and transcript here.

Do commit messages even matter anymore, or did pull requests kill them? Ben works one commit per PR and thinks the commit message is the PR description. Carol and Tim put all the context in the PR and treat commits as disposable breadcrumbs. Adam's somewhere in between — when he's not pushing thirty knife emojis and "nope, still not working" to QA. Meanwhile, Tim's back from emergency eye surgery with a gas bubble floating around his eyeball.

• Ben Nadel's Blog
• Conventional Commits

Follow the show and be sure to join the discussion on Discord! Our website is workingcode.dev and we're @workingcode.dev on Bluesky. New episodes drop weekly on Thursday.

And, if you're feeling the love, support us on Patreon.

With audio editing and engineering by ZCross Media.

Full show notes and transcript here.

For ten years, Adam's codebase has carried an ORM layer that everybody knew was wrong and nobody was touching. Nine hundred functions. Fifteen hundred files. The kind of job that gets solemnly nodded at in architecture meetings and quietly dies on the roadmap — every single year. So he stopped waiting for a volunteer and handed it to an AI agent instead. Claude's problem now.

Follow the show and be sure to join the discussion on Discord! Our website is workingcode.dev and we're @workingcode.dev on Bluesky. New episodes drop weekly on Thursday.

And, if you're feeling the love, support us on Patreon.

With audio editing and engineering by ZCross Media.

Full show notes and transcript here.

Ben had been riding high on vibe coding—throwaway scripts, zero attachment, pure productivity magic. Then he tried the same approach on a project he actually cares about and watched that 10x feeling crater to something closer to 10%. The bottleneck, it turns out, was never the typing.

The hosts dig into what it feels like to let go of code you used to care about, whether "write-only code" is actually the future, and the growing gap between building software and keeping it alive.

• Vibe Coding by Gene Kim & Steve Yegge - The audiobook on AI-assisted development
• 1Password: From Magic to Malware - How OpenClaw's agent skills became a supply chain attack surface
• TLDR Newsletter - Source of the "write-only code" concept

Follow the show and be sure to join the discussion on Discord! Our website is workingcode.dev and we're @workingcode.dev on Bluesky. New episodes drop weekly on Thursday.

And, if you're feeling the love, support us on Patreon.

With audio editing and engineering by ZCross Media.

Full show notes and transcript here.

Adam built a Claude Code skill for his Taffy REST framework and wanted to share it with the CFML community. Simple enough—create a GitHub repo, add some markdown files, done. But somewhere between "this is cool" and "anyone can install this," a familiar chill crept in. These skills are just text files. No checksums. No digital signatures. No verification that the thing you're installing won't quietly exfiltrate your code to some server in Eastern Europe. Sound familiar? It should. We've been here before—back when passwords lived in plain text and "security" meant hoping nobody looked too hard.

The hosts dig into the unsettling parallels between today's LLM plugin ecosystem and the wild west of early internet security.

• Adam's Dotfiles Blog Post - Getting his shit together with dotfiles, Brewfile, and 1Password SSH agent
• CF Community LLM Marketplace - Adam's community marketplace for CFML-related Claude skills
• Steve Yegge's Google Platforms Rant - The infamous accidentally-public Google+ post
• Vibe Coding by Gene Kim & Steve Yegge - The audiobook Ben's been enjoying
• Socket.dev - Supply chain security for npm dependencies

Follow the show and be sure to join the discussion on Discord! Our website is workingcode.dev and we're @workingcode.dev on Bluesky. New episodes drop weekly on Thursday.

And, if you're feeling the love, support us on Patreon.

With audio editing and engineering by ZCross Media.

Full show notes and transcript here.

Ben's been circling vibe coding for months, kept at bay by a simple fear: what if he spends more time fighting the AI over formatting than actually building anything? What if he has to bolt on linters and test runners just to babysit the output? Then his work handed him a Claude plan, and he decided it was finally time to take the plunge. And then something unsettling happened—the code looked like his code. Same line lengths. Same method ordering. Same obsessive formatting. Nobody told it to do that. It just... knew.

Meanwhile, Adam has gone full mad scientist. His "Ralph" workflow runs Claude in a loop, feeding it tasks from a JSON file while he walks away to eat dinner. When he comes back, features are done. Tests pass. The machine just keeps building. It's the kind of setup that makes you wonder why you're still manually typing commands into a terminal.

• Adam's Ralph Workflow for Claude Code - Adam's blog post with his implementation
• Matt Pocock's Ralph Primer Video - The workflow Adam adapted for automated iterative development
• Algorithm Maze Race - Tim's vibe-coded game on itch.io
• Pro tip: Use /resume in Claude Code to return to prior sessions

Follow the show and be sure to join the discussion on Discord! Our website is workingcode.dev and we're @workingcode.dev on Bluesky. New episodes drop weekly on Thursday.

And, if you're feeling the love, support us on Patreon.

With audio editing and engineering by ZCross Media.

Full show notes and transcript here.