Chris Smith
a7c05104af
Moves the Next.js app's contents from new-site/ to the repository root and deletes the previous Hugo site (assets/, content/, themes/, hugo.toml, etc.). Also retires the AWS Amplify config and old Netlify _redirects file — the new site deploys to Vercel. Updates STRATEGY.md path references to drop the new-site/ prefix. LICENSE preserved. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
45 lines
3.1 KiB
Plaintext
45 lines
3.1 KiB
Plaintext
export const metadata = {
|
|
title: "Why I take 'glue work' seriously (and you should too)",
|
|
description:
|
|
"The unsexy code that lives between your store, your warehouse, and your books is doing more for the business than the cool new feature ever will.",
|
|
date: "2026-04-30",
|
|
readingMinutes: 5,
|
|
tags: ["e-commerce", "philosophy"],
|
|
};
|
|
|
|
There's a kind of code I love writing that almost nobody brags about.
|
|
|
|
It's the script that takes a CSV from a 1996-era ERP, normalizes the SKUs, decides whether each line is a real shipment or a return, validates the address against USPS, and hands the result to a 3PL's API in the format that 3PL expects this Tuesday. It's the cron that watches a folder for EDI 940s, parses them, and writes a row in your warehouse system. It's the one-page admin tool that lets a customer service rep look up an order without opening five tabs.
|
|
|
|
Glue work, integration work, ops tooling — whatever you call it, this is the code that keeps small businesses shipping.
|
|
|
|
## It looks dumb. It isn't.
|
|
|
|
The first time someone sees an integration like this, the reaction is usually some version of "wait, that's it?" A few hundred lines of PHP, a queue, a Postgres table, a few well-placed log lines. No clever framework. No machine learning. Sometimes not even a class.
|
|
|
|
But here's what that "few hundred lines" is actually doing:
|
|
|
|
- Encoding **how the business actually works**, including the exceptions nobody wrote down
|
|
- Surviving every change a vendor pushes in the next three years
|
|
- Logging enough that when something breaks at 4pm, you can tell *what* in under a minute
|
|
|
|
That's a lot of value packed into something that looks like a script.
|
|
|
|
## The skill is in the boring parts
|
|
|
|
Anyone can write the happy path. The skill is in:
|
|
|
|
1. **Reading the messy reality.** Real input data is never clean. Half the bugs in glue code come from someone shipping a SKU that "shouldn't exist" or an order with a comma in the customer name. Senior glue work assumes your data is messy and proves it before going to production.
|
|
2. **Logging like an adult.** A glue script that runs unattended needs to leave a trail you can read six months later when a CFO asks why a single order shipped twice. Structured logs, idempotency keys, retry semantics that won't double-charge a customer.
|
|
3. **Writing tests against contracts, not just code.** When the 3PL changes their API on a Sunday night, the test suite that catches it is the one that ran against a recorded fixture, not the one that mocked the function call.
|
|
|
|
None of that is glamorous. All of it is the difference between glue that lasts five years and glue you have to rip out in three months.
|
|
|
|
## Why I keep doing this work
|
|
|
|
I get to look at a small business and see, very concretely, where money is leaking out — usually because someone is doing in three hours of manual work what a 200-line PHP script could do in 200 milliseconds.
|
|
|
|
When I write that script, I get to give those three hours back to a real person. Repeat that across a year and the math is easy.
|
|
|
|
That's why glue work is worth taking seriously. The alternative is human beings spending their working lives moving CSVs around.
|