Cosmic
August 22, 2026
This article is part of our ongoing series exploring the latest developments in technology, designed to educate and inform developers, content teams, and technical leaders about trends shaping our industry.
The Model Context Protocol got a roadmap, a new Rust LSP promises to slash memory usage by 100x, and Dan Luu makes the case that slow software is now a choice. Here is what developers are talking about.
The MCP Roadmap Arrives
Anthropics's Model Context Protocol team published The New MCP Roadmap, outlining where the protocol is headed. MCP has become the standard way to connect AI assistants to external tools and data sources, and this roadmap signals the project is moving beyond its initial launch phase.
The Hacker News discussion digs into implementation details and what developers want to see next. For teams building AI-powered applications, MCP compatibility is increasingly table stakes.
Cosmic's own MCP Server lets you connect Claude, Cursor, or any MCP client directly to your content model. The roadmap suggests more capabilities are coming to the protocol, which means more ways to wire AI into your content workflows.
Rust Glancer: 100x Less RAM for Your LSP
Rust Glancer is a new Rust language server that claims to use 100x less RAM than rust-analyzer. The project takes a different approach: instead of loading the entire crate graph into memory, it analyzes only what you need when you need it.
The discussion covers the tradeoffs. Rust-analyzer provides more features, but Glancer may be enough for many workflows, especially on memory-constrained machines. The broader pattern here matters: developer tools are getting serious about resource efficiency.
Dan Luu on Software Performance
Dan Luu published There's no reason for software to be slow anymore, arguing that the hardware excuses for poor performance have expired. Modern machines are fast enough that slow software is a choice, not a constraint.
The Hacker News thread predictably turned into a debate about tradeoffs, developer time versus user time, and whether performance even matters for most applications. What is not debatable: users notice when things are slow, and they increasingly have alternatives.
For content teams, this applies to your publishing tools as much as your public-facing sites. A CMS that makes you wait is a CMS that slows down your whole operation.
Felony Bench: Cataloging Hostile Architecture
Felony Bench documents hostile architecture across cities. The project catalogs benches designed to prevent sleeping, spikes to discourage sitting, and other design choices that prioritize exclusion over utility.
The discussion goes beyond the obvious ethical debates into questions of public space, design philosophy, and who cities are built for. Not directly a tech story, but the documentation approach and the questions it raises about design decisions resonate.
OTel Adoption Reality Check
Mat Duggan wrote OTel isn't going well, documenting the current state of OpenTelemetry adoption with actual data. The short version: the standard is winning the spec war but the implementations remain fragmented.
The discussion includes war stories from teams trying to adopt OTel. If you are evaluating observability tools, this is worth reading before you commit to a strategy.
Quick Hits
Border privacy under pressure. A citizen faces felony charges for deleting phone data at the US Border. The discussion covers the legal implications and what this means for travel with digital devices.
Kagi filters paywalls. Kagi added a setting for removing paywalled links from search results. The thread debates whether this helps or hurts quality journalism.
Meta trial coverage. The Guardian reports on Meta's alleged strategy for engaging children, characterized as "hook, hold, harvest and hide." The discussion covers the trial's implications for platform regulation.
Kobo runs apps now. Cobalt lets Kobo e-readers run third-party applications. The thread covers what this opens up for e-ink enthusiasts.
AI-blind fatigue. A post about becoming AI-blind describes the growing tendency to mentally filter out AI-generated content. The discussion reflects a sentiment many are starting to share.
Zig's threading model. An analysis of Zig's Io.Threaded explains why the approach is interesting. Worth reading if you are following systems programming evolution.
What This Means for Content Teams
The MCP roadmap and Rust Glancer release both point to the same trend: the tools developers use to build things are getting more capable and more efficient. When your AI assistant can read your content model directly through MCP, and your IDE uses a fraction of the memory it used to, the friction of building drops.
For content operations, less friction means faster iteration. You can experiment with AI-assisted workflows, automate repetitive tasks, and still maintain control over what gets published.
Cosmic's AI agents handle the automation while you set the rules. Connect through our MCP server to let Claude work with your actual content model, or start with a free account to see how AI fits into your stack.
Give your AI agents a content backend they can write to
Structured, versioned content objects, a REST API and TypeScript SDK, and an MCP server your coding agent connects to directly. The Free plan includes 1 Bucket, 1,000 Objects, and 1 agent. No credit card required.
