System Architecture

Durable agent-to-agent messaging using a filesystem: inspectable, remote, replayable, and SQL-queryable with db9 fs9.

Intra-domain delegation enables an agent to grant constrained authority to subagents without expanding the system’s trust boundary. In the OpenClaw + Cedar demo, delegation is modeled as data and enforced by a stable policy set, preserving policy integrity while allowing dynamic scoping of power.

If you have built APIs, then by now, you are already familiar with the practices of how we typically build APIs:

The Sequel Nobody Asked For (But AI Agents Demanded)

For the next paper in my Paper Implementation series, I picked the Dynamo 2007 paper by Amazon. I had a really good time with this one. Large scale distributed systems always fascinate me , how do you handle that much data, across that many machines, and keep it working when things inevitably break? The Dynamo paper has a clear answer, and it keeps repeating it in case you miss it: eventual consistency.

Modern verification systems force individuals to rely on institutions to prove facts about themselves, creating a

A practical, SQL-first workflow: file artifacts + Postgres state + fs9 (files as relations) for retrieval, governance, and debuggability.

The famous article Choose Boring Technology lists two problems with using innovative technology: There are too many "unknown unknowns" in a new technology,...

Privacy-first email infrastructure for fast growing companies. Run agents and automations securely on your email and proprietary data.

The following article originally appeared on Q McCallum’s blog and is being republished here with the author’s permission.Generative AI agents and rogue

I deployed a standard ECS architecture for Lush Aural Treats and didn't watch the bill. NAT gateways and ALBs quietly ate $1,000. Here's how I redesigned the whole thing to cost almost nothing.

Raft is a consensus algorithm that allows a group of servers to agree on the order of a sequence of commands. The objective of this article is not to re-explain RAFT, as this work has already been done in several other places, and the RAFT paper gives enough details to implement it. Instead, my focus will be on two problems that I faced when I voluntarily studied the open distributed systems course from MIT (course code 6584, I didn’t enroll in the course officially, and I wasn’t a student of MIT).

I introduce Memory-Continuous Architecture (MCA), a runtime model that decouples deployment boundaries from memory boundaries in function composition. KubeFn implements MCA across JVM, Python, and Node.js runtimes, demonstrating 4-100x latency reduction in full HTTP benchmarks.

Learn how to use AI as an intent layer in Phoenix apps to build reliable, structured filtering without replacing your existing UI.

The database market is harsh, particularly for newcomers. It’s very hard to launch a new product and differentiate yourself from the incumbents. Even …

Funny notes on niche optimizations in Rust, and a few updates.

A practical deep dive into container image internals that will help you build a clear mental model of how images are composed, identified, stored, and distributed across registries.

Creating web standards for AI models means wrestling with questions that didn't exist two years ago. Here's what I've learned building the prompt API and its siblings.

Eight Weeks to a Working Key-Value Storage Engine

wingolog: article: nominal types in webassembly

Inspired by comments under my recent post where I shared my painful experience from reactive code, I'd like to show how simple it is to write streaming code with plain Java API, avoiding reactive streams API altogether. Handling streams of events is the main concern when people discuss whether react

Learn how Amazon built the backbone of the modern web that scales to 1 PB/s and 150M QPS on commodity hard drives

One of my favorite SDF techniques is domain repetition:

A practical approach to running expensive accessibility and Lighthouse tests in CI without ballooning your time budget.

Learn how to design an OTel-native backend that seamlessly exports logs, traces, and metrics to any observability stack via OTLP.

Exploring a minimal implementation of a time-sharing kernel on RISC-V, implemented in Zig, on top of OpenSBI.

Overview arguing that OpenTelemetry eBPF Instrumentation removes barriers to full observability by enabling zero-code, kernel-level telemetry

One of my favorite YouTube channels is Reject Convenience. The core premise of the channel is; to engage with the modern technological landscape, consumers need to contend with the tradeoff between convenience and privacy. To be a consumer of any tech product, be it for navigation, entertainment, productivity, or even health, is to cede a treasure trove of data about your lifestyle, habits, and personal thoughts to the Charybdis that is Silicon Valley. The channel focuses on educating consumers about how their privacy is at risk online, available alternatives, and evaluating privacy tools, such as VPNs.

How do you measure product-market fit for a developer tool? A PMF scoring model from Evil Martians—a product development consultancy for developer tools startups—built on data from 37 devtools companies across AI, infrastructure, and cybersecurity. Five metrics, real benchmarks, and a dual score that tells you whether to invest in product or go-to-market.