Most important take away

Stop reinventing the wheel: organizations should use existing, proven, and maintained solutions for common problems (auth, databases, queues, UI frameworks) rather than building, maintaining, and operating their own. The bigger industry challenge is open source sustainability — projects need real, predictable funding (paid licenses, support contracts) rather than fickle donations, and engineers and leaders need to budget for “free” software as a real line item.

Summary

Dylan Beattie joins Carl and Richard to introduce usewhatworks.org, a manifesto and community pushing back on the developer instinct to build from scratch when robust, off-the-shelf solutions already exist. The discussion centers on actionable career and architecture advice for developers, tech leads, and maintainers navigating the modern open source landscape.

Actionable insights for engineers and architects:

• Default to established, maintained software for commodity concerns (authentication, password reset, MFA, social login, key/value stores, databases, queuing, container orchestration). A homegrown login system rarely accounts for timing attacks, GDPR retention, 2FA, or federated identity; a good off-the-shelf product already does.
• Distinguish between “open source” and “maintained.” Existence is not a quality signal — community continuity and active maintenance are. Garnet (Microsoft Research’s Redis-compatible drop-in) is cited as an example where longevity is still an open question.
• Learn complex patterns (distributed systems, custom databases, your own auth) on your own time as a craft, but do not ship them at work unless absolutely necessary. Dylan teaches distributed systems with an explicit warning: “don’t go back to work next week and do this” — instead learn the signs that indicate when the complexity is finally justified.
• Career value comes from knowing how to evaluate, integrate, tune, back up, and operate off-the-shelf systems (e.g., understanding SQLite WAL mode means a file copy is not a real backup). These transferable skills are more marketable than bespoke internal frameworks you can’t take with you.
• The MVP of any infrastructure component is deceptively easy; the real cost is the long tail of maintenance, compliance, and 3 AM pages. Frame the buy-vs-build conversation in terms of total cost of ownership, not initial development hours.
• “If you can’t sell it, buy it.” Engineering time should be spent on the differentiating “secret sauce” of the business, not on commodity infrastructure.

Architecture and licensing patterns discussed:

• The Redis case study: permissive licensing enabled hyperscalers (AWS) to resell the product without contributing back, leading Redis to move first to a restrictive source-available license and eventually to GPL. GPL acts as a strong deterrent to cloud vendors because it would force them to open source their billing, metrics, and orchestration code.
• Avalonia as a sustainability success story: an open source WPF replacement that survives largely on a single major corporate sponsorship plus paid commercial licensing. It fills the cross-platform desktop gap Microsoft never cleanly solved (WinForms aging, WPF Windows-only, MAUI/WinUI churn, Blazor desktop = Electron).
• The ServiceStack precedent (2014): when a project shifts from BSD/MIT to a commercial license, the inevitable community fork attracts the entitled users while serious customers follow the upstream. Forks then collapse under unfunded support demand.
• Sponsorship is fickle and comes with strings — sponsors expect priority and influence. A paid license model on a line-item budget is more sustainable than donations, which are the first thing cut in a downturn.
• For maintainers: think about commercial viability before you publish, not after. Picking a license is a long-term sustainability decision, not just a vibe.

Career advice for new maintainers and developers:

• If you release something useful, plan early for how it could pay its own bills and how it could eventually be handed off — bus-factor-of-one is the silent killer of “small but critical” projects.
• AI coding assistants do not fix the fundamental problems that structured programming, OO, functional, Scrum, Agile, Kanban, or cloud did not fix — confidence in what reaches production. Reliance on AI assistants instead of junior developers risks drying up the pipeline of future seniors and architects, who historically learn the craft by building “workpieces” (Tetris, toy databases, toy auth) on the side.
• Code-as-craft still matters: build your own X to learn, then use what works at work.

Get involved at usewhatworks.org — sign the manifesto, open GitHub issues/discussions, contribute case studies, or push back if you disagree with the approach.

Chapter Summaries

• Intro and 2002 retrospective: Episode 2002 milestone. Recap of 2002 — Euro launch, SARS, Salt Lake Olympics, Halle Berry’s Oscar, Lord of the Rings: Two Towers, Pioneer 10’s last signal, SpaceX founding, Mars Odyssey detects water ice, Phoenix/Firefox browser, LinkedIn founded, Trillian, Handspring Treo, Roomba, Battlefield 1942, Audioslave, and the RTM of .NET 1.0.
• Better Know a Framework: GitHub Copilot CLI now supports bring-your-own LLM via OpenAI-compatible APIs, including local Ollama. Carl runs Qwen Coder 30B locally — fast, good for tactical work, weaker on architecture; pulls enough power to trip a small UPS. Copilot price hike in June makes local inference attractive.
• Listener comment: Rob Howard recalls a Y2K rollover support call from Australia.
• Dylan intro and current work: Porting Next.js to Astro, livestreaming “bad Claude code” with Rendle, building MCP servers, learning modern CSS. Riff on conference panel-show format ideas (Last One Laughing / QI style).
• Use What Works origin: ~35 signatories on the manifesto at usewhatworks.org. Emerged from a year of conference talks on open source sustainability and the Reddit-fueled “open source rug pull” narrative.
• Redis case study: License changes driven by hyperscalers reselling free software; community backlash; Microsoft Garnet as a drop-in; eventual move to GPL as a stronger anti-hyperscaler stance.
• Avalonia and desktop UI: Mike James’s London talk on sustainability; Avalonia as a cross-platform WPF replacement filling the gap left by Microsoft’s desktop churn (WinForms, WPF, MAUI, WinUI, Blazor + Electron).
• The build-vs-buy trap: How “we’ll just build our own auth” snowballs into 2FA, social login, timing attacks, GDPR — the MVP is easy, the maintenance is the cost.
• AI does not fix this: AI hasn’t solved the production-confidence problems prior methodologies didn’t solve; uptime of major AI providers is dropping below meaningful nines; putting AI in things is not interesting — new capabilities are.
• Career and craft angle: Learn complex systems on your own time; ship the boring proven thing at work; transferable operational skills (backups, tuning) beat bespoke internal frameworks; concern about junior pipeline drying up if AI replaces entry-level work.
• Sustainability models: Sponsorship is fickle and political; paid licenses on a budget line are durable; ServiceStack’s 2014 commercial pivot as precedent — serious users follow upstream, freeloaders go to the fork and the fork dies under unfunded demand.
• “If you can’t sell it, buy it”: Spend engineering time on the differentiating secret sauce; everything else, use what works.
• Call to action: Visit usewhatworks.org, sign the manifesto, file issues/discussions, share case studies, contribute to the conversation on long-term project continuity.