Tech

How systems, interfaces, and the tools we build with actually work — the tradeoffs hiding inside a feature flag, a cache, an API, an AI eval. For the parts of engineering that shape the product, not just the parts that pass code review.

Tech

How to say "definitely not" without looking.

The trick to fast writes is to stop updating in place.

Every request takes a token — or waits for one.

They all have to agree — and any of them can disappear.

The system heals itself while nobody's looking.

The model can say anything. Your schema decides what ships.

Fail in one room, not the whole house.

Your agent will believe the webpage.

The agent doesn't want your cursor. It wants a worktree.

You can't test for behavior you never wrote down

One plug for every tool

The biggest model is usually the wrong default

Your agent needs a key, not your keys

Your context window is a budget, not a bucket

You can't undo it, so plan the apology

You can't save the row and send the email in the same breath

Let people see their own writes

At scale, your average doesn't matter — your tail does

Stop calling a service that isn't answering

Write it down before you do it

When you run out of connections, you don't run out slowly

The better your AI gets, the harder the human's job becomes

Exactly once is a story you tell yourself

When the cache expires, everyone shows up at once

You can't have all three when the network breaks

A feature flag is a loan against your future simplicity.

You can't just change the table

You can't fix what you can't ask

Evals Are the Unit Tests of the AI Era

A queue is a loan, not a fix

Everything you cache is a bet that the world will hold still

Don't tear down the fence until you know why it's there

Undo is the most underrated feature you can build

Your architecture is your org chart

Calibrated confidence: why AI products live or die on knowing when they don't know

Every abstraction leaks. The skill is choosing which leaks you'll live with.

Your stack is a liability, not an asset