"A stale AGENTS.md can be worse than no file at all." - @cloudflare.social So, how can we write a better AGENTS.md file? -> bit.ly/438ziv8

Multi-agent failure mode! Not only a great band name, but worth thinking about how orchestration is mostly about explicit handoff contracts.

AI Code review @cloudflare.social runs over 5K+ repos. There's even an agent they run that checks if your AGENTS.md is accurate! Great to see context-rot being used as a review gate. Keeping that context current is the problem we're working on at @dosu.dev blog.cloudflare.com/ai-code-revi...

“What happens when every company has access to the same model? The best riders win.” www.tomtunguz.com/harnessing-ai/

Letting coding agents handle their own MCP configuration hands-free via the CLI is the hero this city deserves. Thrilled to see bulk doc generation land in this drop, too. Give it a spin and let us know what you think! 🚀

"Signing is for the bad days" nesbitt.io/2026/05/24/s...

Every couple months I check in on the latest Adrian Chiles articles and I’m always absolutely blown away. These are all from the last couple months. The man is operating at an incredibly high level.

For anyone shipping against deadlines…Lucas Costa has a profound take on why those dates work as pressure, but you can do better by your team. Using preemption points might be a great perspective to bring to your next quarterly planning sesh. www.lucasfcosta.com/blog/deadlines

The NHS made its open source projects private after vulnerabilities got reported. While hiding potential bugs, that doesn't fix the problem. Security by obscurity is NOT security. I’m glad the GDS highlighted that fact. simonwillison.net/2026/May/17/...

I love spring in Los Angeles

If you run an open-source project and aren't using @dosu.dev, you're missing out! Dosu's responses to new issues are really on point, and it's great to have them reference other issues and PRs, and brainstorm within an issue to make better decisions.

A new tutorial has emerged (just in time for Friday)! Learn how you can turn "how fresh are our docs?" into a 0-100 score that runs in CI on every PR.

Thank you Bloomberg engineering for stepping up here, would love to see other companies do this with open source projects "Sustaining OpenTelemetry: Moving from dependency management to stewardship" www.cncf.io/blog/2026/03...

The April @dosu.dev Drop is out and it's an OSS-heavy one!

Nothing like a trail, some sun, and lots of quiet time in nature

New post on building /doc-it! Most interesting finding is around context files getting stale (silently) and your coding agent performance getting worse without telling you.

Today, I came up with a wild plan.

This drop is a real one. Doc Export helps close the loop with your workflows. Generate in Dosu, push to wherever your team works, and keep that knowledge up to date. Plus...doc reviews got a real quality bump. Would love to know what you think!

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...

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.