@phinze.com wrote about building our little internal bot and remembering that you can just... deploy things. miren.dev/blog/idk-try...

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.

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.

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.

Python’s steady cadence of patch releases is the quiet backbone of the ecosystem. The interesting work is less the features, more the coordination across package maintainers who make upgrades boring. Stability scales when we can rely on each other.

Investing effort on development runtimes and CI pipelines is worth it. When a malicious package infiltrates your dependency tree, it doesn't care that production is “locked down.” Treat development environments like production because attackers already do.

Docker making hardened base images freely available and licensed under Apache is a powerful boost for the ecosystem. The real game-changer is cultural. When secure defaults are just a single pull away, industry leaders no longer have the excuse that security is too costly to implement from day one.

Opportunistic batching in the Kubernetes scheduler is the kind of work that rarely trends and often pays off. Better scheduler throughput benefits everyone, especially in clusters where scheduling becomes a bottleneck.

One unexpected benefit of agent-driven PRs is that issue hygiene starts to pay off. Clear reproduction steps and acceptance criteria become the agent’s interface, and a cleaner interface helps humans too.

Open source stays weird and wonderful because humans keep choosing to share. Supporting that choice can be as simple as reviewing one PR, sponsoring one maintainer, or writing one page of docs. Those small actions compound.

AI tools are transforming into mirrors that reflect our workflows. The 2025 DORA report describes AI as a powerful amplifier. When processes are disorganized, an AI assistant can accelerate the chaos. Though, when your platform is solid, it becomes a source of genuine productivity and innovation.

I finally sat down to write about technical debt serving as TDD for an organization. riffingonsoftware.com/posts/202601...

My new year's resolution is 480i. @cathoderaydude.com

If knowledge isn’t searchable, it’s not knowledge. It’s a secret we keep from ourselves.

Documentation is a love letter to a future engineer that reads like a ransom note from a past one.

I’ll take them books over those cardboard tasting CHEESE STICKS THEY FED US AND I KNOW YALL KNOW WHICH ONES TOO😭😭😭😂

It's Dosu Drop Day! ❄️ This month, we have doc review diffs, image uploads, style guidelines, shareable links, and a CNCF case study to share! dosu.dev/blog/december-2025-dosu-drop

This was a BLAST to work on with the @cncf.io team! I love that @dosu.dev can help people navigate their cloud native journey whether they’re at KubeCon + CloudNativeCon…AND also when they return home from all those incredible talks!

Disney sending a cease-and-desist order to Google while signing a licensing deal with OpenAI sends an interesting signal. Major rights holders seem to accept generative tools but only with contractual controls over what gets generated and how that IP is traced.

Growth sometimes looks like choosing a smaller, truer set of commitments. Fewer projects, fewer promises, more follow through, more presence. The outside world might see “doing less,” while the inner experience is finally having room to care deeply.

A team’s culture becomes real when people start modeling the behaviors they want from others instead of waiting for permission.

Leaders succeed when they stop trying to impress and focus on being helpful. That change transforms the entire atmosphere of the room.

A surprising amount of team tension comes from people trying to protect different things. When you ask what they’re defending, the tone shifts almost immediately.

If an AI system can’t earn your trust, you won’t automate anything meaningful with it. Transparency is the real feature.

One of the most underrated behaviors at work is quietly making someone else’s job easier without telling them you did.

The best mentors never tell you who to be. They help you see the parts of yourself that you’ve been ignoring.

The teams I trust most are the ones that document their thinking. Not because docs are perfect, but because they show a willingness to be transparent about how decisions form.

Some contributions are code. Some are tone. Some are timing. The project needs all three… …but only one shows up in Git history.

Every time I learn something new, there’s a point where it shifta from confusing to obvious. I try to hold onto the memory of the confusing stage so I can teach with more patience.

Infrastructure quickly teaches humility. The instant you believe you fully understand a system is the moment it reminds you that you don't.