Getting Freaky in the Age of AI
The only way to have a competitive edge in the age of AI is to be aggressively weird with what you build.
Ian Duncan
Software engineer, open source contributor, and occasional writer.
Latest posts
The Mathematics of Build Queue Optimization and Auto-Scaling
A deep dive into queueing theory and auto-scaling for CI/CD build fleets, exploring when and how to optimize build infrastructure for both cost and developer productivity.
Deconstruct the Monolith: Composable App Bootstrapping with Fractal Layer
How fractal-layer attacks the compilation bottleneck and initialization chaos of the monolithic App datatype by treating startup as a composable graph of resource-managed layers.
Brief aan de Eerste Kamer over box 3 · A letter to the Dutch Senate on Box 3
Waarom de vermogensaanwasbelasting slecht uitpakt — voor de belastingplichtige, de schatkist en de economie. · Why the accrual tax is bad for the taxpayer, the Treasury, and the economy.
Essential Complexity
Brooks's distinction between essential and accidental complexity in software systems
Phronema
Phronema as engineering judgment: mentorship vs dead rules
Premature Optimization
Knuth's full quote on premature optimization and how the slogan gets misused
YAGNI
YAGNI as engineering principle and when it becomes a thought-terminating cliché
What's in the Box? A Field Guide to AI Models
A beginner-friendly tour of parameters, quantization, MoE, context windows, and other LLM jargon.
Stealing from Biologists to Compile Haskell Faster
The trick that makes GHC’s ApplicativeDo optimization faster is structurally identical to how biologists predict RNA folding.
Making Haskell Talk to PostgreSQL Without Suffering
How to eliminate the three performance taxes that make database queries slow: encoding overhead, round-trip latency, and N+1 query patterns.
Archive: Automatic Optimal Pipelining of Redis Commands
Archived from Informatikr (originally published January 18, 2012). Describes how the Haskell Redis client Hedis achieves automatic and optimal pipelining by combining lazy I/O with a synchronous-looking API, getting the best of both synchronous and asynchronous approaches.
TIL: HTTP/3 Is Not Always Faster Than HTTP/2
Local benchmarks revealed HTTP/3 can be 50-100x slower than HTTP/2. QUIC's userspace implementation loses to decades of kernel TCP optimizations on high-bandwidth, low-latency networks.
What Functional Programmers Get Wrong About Systems
Type systems verify properties of programs. Production correctness is a property of systems. The gap between these is where the interesting failures live.
Your Best Thinking Is Wasted on the Wrong Decisions
Some engineering decisions punish you for being wrong. Most don't. Learning to tell the difference and calibrate your deliberation accordingly might be the most valuable skill your team never explicitly develops.
In Praise of Control Planes, or: Why You Need a Place to Stand
The control plane pattern (a coordinator that tells workers what to do) shows up everywhere good infrastructure exists. The concept is simple. Naming it is useful.
No, Really, Bash Is Not Enough: Why Large-Scale CI Needs an Orchestrator
A deeper dive into why shell scripts cannot replace a proper CI control plane, drawn from years of building and operating CI systems at scale.
GitHub Actions Is Slowly Killing Your Engineering Team
Why GitHub Actions is the Internet Explorer of CI, and why Buildkite offers a better path forward for teams that care about developer experience.
Punycode: My New Favorite Algorithm
A deep dive into how Punycode works, and why I think it's so neat.
The CRDT Dictionary: A Field Guide to Conflict-Free Replicated Data Types
A comprehensive guide to CRDTs and their tradeoffs, from counters to sequences. Written in the spirit of the Typeclassopedia, exploring how different CRDTs solve the distributed consensus puzzle.
JSON Schema Demystified: Understanding Schemas, Dialects, Vocabularies, and Metaschemas
An accessible guide to understanding the confusing terminology around JSON Schema. What are schemas, dialects, vocabularies, and metaschemas, and how do they all fit together?
Event Design for Streaming Systems: A Primer
How to use event streams in an effective, scalable, future-proof way.
Accidental or Essential? Understanding Complexity in Software Design
Understanding the difference between essential and accidental complexity in software design.
What You Need to Know About Trump\
A look at the Trump administration\
Actuarial Murder and Other Diffuse Acts of Violence
An exploration of how institutional decisions and systemic policies can cause widespread harm with less accountability than direct violence, examining cases from healthcare to education where bureaucratic choices have life-altering consequences despite being treated as mere business decisions.
The Cadence of Salting
When you salt matters as much as how much you salt. Understanding the timing changes everything.
Embracing Flexibility in Haskell Libraries: The Power of Records of Functions
Improve the robustness and flexibility of your Haskell apps by changing how you design your libraries.
Hi, Barbie! Higher kinded records made easy(er)
We often need to make and manipulate derived types that are similar to each other, but not quite the same. Higher-kinded records provide a way to do this in a generic way, but are clunky. We\
Haskell Best Practices– Wildcards aren\
An argument for using really exhaustive pattern matching.
There and Back Again: Turning Haskell functions into RPC calls- Part 2
In the last post, we figured out an interface for the RPC caller to invoke remote Haskell functions. In this post, we\
There and Back Again: Turning Haskell functions into RPC calls- Part 1
In
Integrating a large Tokio-based Rust library with Haskell
A lot of the Rust Haskell interop examples out there are for small, synchronous libraries. They also tend to omit how to actually package the Rust library. We\
One Serialization Class to Rule Them All
Typical serialization libraries in the Haskell ecosystem target a single format. We explore how to provide composite serialization for data types that chooses the best format for each data type.
OpenTelemetry Gotchas– Phantom Spans
We struggled for several months with a mysterious tracing issue in our production environment. Unrelated web requests were being linked together in the same trace, but we could never see the root trace span. This is the story of how we found and fixed root cause.
The Cruel Part Is
A poem
Vignettes
A collection of memories
Garbage in, Never out
Adventures in streaming algorithms.
Using React with GHCJS
Early explorations in data-driven Haskell applications in the browser.
Introducing Growler, a simple HTTP server toolkit
Sinatra-esque Haskell web apps.