Find a thread to follow

What are you curious about?

Search stories and experiments by topic, tool, or idea.

Browse all 10 stories and experiments

AI
Next-Concept Modeling: a model that doesn't write word by word
A Stanford paper sketches the whole answer first, then fills in detail layer by layer. How NCM works, what it achieved and the catch.
AI
The AI bubble could burst. The disruption could stay.
An AI investment crash could leave cheaper software, smaller teams, and a harder market for engineering work. The tools would still exist.
Engineering
How the new Tesla Roadster stays upside down
Fans, sealing skirts and cold-gas thrusters do different jobs. Follow the forces behind the Roadster's inverted ramp and airborne demonstrations.
AI
The interesting part of AI is the system around it
A model can generate an answer. Useful AI also needs context, tools, feedback, and a clear boundary around what it can do.
Engineering
Latency is a budget, not a single number
Network time, queueing, and service work all spend the same budget. Explore what changes when independent work runs in parallel.
Software
A cache is a promise about freshness
Caching saves work by reusing an answer. The hard part is deciding how long that answer is allowed to stay true.
The web
From HTML to pixels: what the browser waits for
A page is more than downloaded bytes. Follow the work between receiving HTML and putting something useful on the screen.
Engineering
Retries need a budget, too
Trying again can recover from a brief failure. Without limits, backoff, and a safe operation, it can also make the failure worse.
Software
An interface is a boundary you can reason about
Clear inputs, explicit failures, and stable expectations make a software boundary useful long after its first implementation.
Interactive lab
The latency budget lab
Change the inputs and compare sequential and parallel requests in an interactive waterfall.