Posts
Aug 17, 2026
Building a bulk upload or download API for structured data usually means picking a file format,
writing parsers and serializers on both sides, coercing strings back into
real types, chunking large transfers, and cleaning up half-loaded data after
failures.
Direct-to-S3 uploads point at a better shape: vend a scoped credential and
let the client write bytes straight to storage, no API in the middle. This
works great for blob uploads, but for structured data, clients typically
still send a CSV or JSON batch (e.g. Stripe’s batch jobs API).
With S3 Tables and Apache Iceberg, we can apply the same
idea to structured data: vend access to a staging table instead of a blob
location, and let the client write rows to it directly. No parsers,
serializers, or large payloads to juggle; the Iceberg ecosystem handles
that for us.
(continue reading)
Mar 10, 2026
After coding with agents, I’ve settled into two habits that have meaningfully improved the experience: keeping a self-documenting AGENTS.md in every repo and doubling down on making everything runnable locally.
(continue reading)
Apr 14, 2025
Special thanks to Tim Cardwell for reviewing a draft of this post.
“Build Server Pong” is what you inadvertently play when you treat your build
server and source control as a REPL (read-edit-print-loop). You make a code change, commit it, push it, kick off a build, grab a coffee, then wait for the results. Rinse and repeat until you’ve got a successful build.
However, even if you eventually score a point by getting a successful build, you’ve lost time in the rally to get it. It’s an easy trap to fall into because most build systems require you to run the workflow on their hosted runners to see the results of your changes.
(continue reading)
Jan 16, 2017
How many bytes are in a kilobyte? 1000 or 1024? The answer depends on who you ask. Google says 1 kilobyte = 1000 bytes. Hard drive manufacturers also say 1 kilobyte = 1000 bytes. However, my computer measures disk space and memory using the ratio 1 kilobyte = 1024 bytes. This discrepancy is well known and has been written about by folks like Jeff Atwood and Randall Munroe in xkcd 394.
(continue reading)
Jan 10, 2017
A plethora of books, articles, and research papers have been written about software engineering quality measurement. Over the course of the last half-century, different methods of measurement have been developed, adopted, abused, and discarded for newer methods that aim to solve the problems of the previous methods.
(continue reading)
Dec 03, 2016
What is the correct way to validate an email address? Technically, a valid email address is described in IETF RFC #822 and can be approximated by this regex (courtesy of Paul Warren):
(continue reading)
Jul 17, 2016
Bitcoins are in vogue. Distributed ledgers, fiat currency, and the dramatic impact cryptocurrencies will have on financial systems are topics of discussion on trains, in board meetings, and at the dinner table. As of writing, the exchange rate is 1 BTC = 679.99 USD.
(continue reading)
Dec 10, 2015
Originally posted on the Pangea Money Transfer Engineering blog.
We get asked “Why .NET?” a lot. It’s rare to find .NET startups these days and for good reason. Historically speaking, the .NET ecosystem was unattractive due to the high licensing cost of Windows and Visual Studio. In addition, .NET had poor interoperability with non-Microsoft products, which meant additional licensing costs for compatible products (like SQL Server). Lastly, Microsoft’s aggressive anti-competitive stance towards its competitors and open source software created a rift between their products and developers.
(continue reading)
Nov 15, 2013
ByteSize is a utility class that makes byte size representation in code easier by removing ambiguity of the value being represented.
(continue reading)
Feb 10, 2013
I’ve come across more than one occasion where a link on a page only needs to modify one or more query string values. Take for example a simple pagination list:
(continue reading)