Dosu documenting things to help out with your codebase.
One unspoken agreement among every engineering team is that someone will update the docs... later. So, we built a GitHub Actions workflow that catches doc drift on merge, and wrote down every sharp edge we hit. dosu.dev/blog/how-to-...
It's Dosu Drop Day! Import docs from Notion, Confluence, GitHub, or GitLab, and Dosu keeps them updated as your code changes. No more docs that go stale too soon after you write them.
dosu.dev/blog/februar...
I asked Claude to fix my printer's "Replace Toner" error. It found the web interface, changed a dropdown from "Stop" to "Continue," and the printer worked. Four minutes.
Then it decided the fix wasn't thorough enough.
bit.ly/4aiRlTS
@phinze.com wrote about building our little internal bot and remembering that you can just... deploy things.
miren.dev/blog/idk-try...
Power user transformation! One of my favorite things to see in the wild :)
It has been a wild month! Nolan Lawson went from calling LLMs "amusing toys" to admitting 90% of his code now comes from Claude. Linus Torvalds started vibe coding his Python visualizers. The tools are changing fast, folks. bit.ly/4tiO6mU
It's Dosu Drop day! 🎉 We have an MCP Server preview (your coding agents can access Dosu's knowledge), an improved diff UI, and we're searching for design partners to test Notion/Confluence/GitHub sync. Let us know if you're interested in taking a look! dosu.dev/blog/january-2026-dosu-drop
AGENTS.md is becoming the README for AI coding agents. 60K+ projects use it now, and I took some time to document what's working well across the ecosystem. bit.ly/4pPbEgf
I love what Kelsey said about the true goal of Kubernetes. That goal is to eventually disappear. The best infrastructure becomes invisible, stable enough that you build on top of it, so you can tackle what's next.
Sandi Metz wrote about the wrong abstraction years ago, and I still think about it. Duplication is cheaper than the wrong abstraction. If future maintainers can't understand why you built it that way, you've created cognitive debt rather than a shortcut.
AI can tell you what code does. It can't tell you why the maintainer wrote it that way three years ago, what broke when they tried it differently, or which edge case that weird conditional handles. That context lives in issues and PR discussions. At @dosu.dev we make that searchable. bit.ly/4qVInkP
Kubernetes has a deprecation policy that gives you years of warning before breaking changes land. Nobody talks about this because it's boring, but it's why enterprises trust the platform. Deprecation policies are promises to your users.
Someone asked me how to navigate the CNCF landscape without getting overwhelmed. My honest answer is that you're not supposed to adopt everything in the landscape. Most teams need maybe 10% of what's there. The value is knowing options exist when you need them.
Platform teams that treat internal developers like customers tend to stick around. The ones that act like they have a captive audience eventually discover that shadow IT is real and developers are creative about routing around tools they don't like.
A decade in open source taught me this: the people who last aren't the ones who shipped the most code. They're the ones who made others feel welcome enough to stay.
Docker launched Kanvas this week. It converts Docker Compose files into Kubernetes manifests automatically and provides visualizations for cloud deployments. Built with Layer5. Direct challenge to Helm and Kustomize. Worth trying if you've been frustrated with the Compose-to-K8s gap.
The tools that last are rarely the ones that promised to change everything. They're the ones who solved a specific problem well and then got out of the way.
Microsoft previewed Azure HorizonDB, a PostgreSQL service for cloud-native workloads with disaggregated storage, zone-redundant architecture, and a Rust-based storage engine. Neat to see that Postgres is receiving significant investment for high-scale AI and analytics.
CIDR without the apples, please
OpenCode gained 3,600 stars in five days. Memory infrastructure projects like claude-mem and memvid are both trending. Chrome DevTools MCP server hit 19,000 stars. The tooling layer around agents is where the energy is right now.
bit.ly/3NkU1ar
The Rust Foundation Maintainers Fund is a concrete example of putting money behind “maintenance.” Code review, refactoring, and release work are the parts that keep ecosystems reliable. Paying for that work is aligned with reality.
A good contributor experience is mostly about time-to-first-yes. Fast feedback, clear labels, and friendly review turn strangers into teammates. People keep showing up when they feel seen.
Foundations are often misunderstood as branding clubs. In practice, they are dispute resolution mechanisms. When two vendors disagree, a neutral process is cheaper than a fork war.
It's one of the best questions 😄
Engineering leadership is often deciding where ambiguity is allowed. Early exploration can be messy, but production work needs crisp constraints. Mixing the two creates chaos.
Burnout sneaks in when work has no edges. Healthy teams protect focus with small rituals, scope boundaries, and “not this quarter” lists. Saying no is a maintenance move.
Deprecations are where projects show their values. Clear timelines and good docs say, “we respect your time.” Surprise removals say, “Good luck.”
Training data is turning into a first-class asset, with price tags and provenance questions. That pressure might push better attribution habits. Datasets need governance too.
Release notes are an expression of trust. When projects are honest about tradeoffs, users can make sense of breaking changes. When they read like marketing, users assume the worst.
Docs are the closest thing open source has to a product manager. They decide who can succeed without a private Slack channel. Good docs are inclusive infrastructure.
