Insights
Engineering perspectives on architecture, product development, AI, and the everyday decisions behind useful software.
From the engineering desk
Browse practical articles or follow the latest technology updates.
Subscribe via RSS →Curated links from external sources — not 360Softy original articles.
Google is rolling out new privacy controls for Search services and Google Play, giving you more control over saved history and personalized recommendations. [...]
On the morning of June 3rd, a customer on a live call sat through 1.4 seconds of dead air after she finished a sentence, long enough that she said "hello?" before the agent answered. I had the trace open in Honeycomb forty seconds later. Every span was green. End-to-end p95 read 980ms, comfortably under our budget, and not one span in that waterfall was longer than 400ms. The dashboard told me everything was fine while the customer was, in fact, talking to silence. TL;DR: End-to-end voice latenc
Originally published on my site, arafatcro.dev/guides. You copy the exit-intent snippet, it fires nicely when the cursor heads for the close button, and you ship. Then you check the data and half your traffic never triggered it, because they were on phones and a phone has no cursor to chase. Exit intent is not one trick. It is a different signal on each device, fired once, and built so it does not become the thing people leave over. The standard snippet listens for the mouse leaving the top of t
I spent three days last month building a specialized API wrapper for a simple Scikit-learn model. Not because the logic was hard—it wasn't. Because I wanted Cursor to be able to run inference on our churn prediction data without me having to manually copy-paste JSON results into the chat. It is a classic engineering trap: the 'Integration Tax.' You have a working Modelbit workspace, you have your weights deployed, and you have an AI agent that could theoretically use it. But instead of using the
Most revision rounds don't spiral because the client is picky. They spiral because the feedback lost context on the way to you. Here's why that happens, what agencies try, and how to make each note clear enough to fix the first time. A revision round is supposed to be one loop: client looks at the site, says what's off, you fix it, they look again. The trouble is that most of the feedback you get can't be acted on as-is. "The hero feels cramped on mobile." Which phone. Which section. Cramped how
You want your app to have its own copy of a user's email: to search it fast, run analytics over it, or feed it to a model without hitting the provider on every query. The naive version is a one-time pull that's stale the moment it finishes, because new mail keeps arriving and old mail gets read, moved, and deleted. A real sync is two jobs working together: backfill the history once, then track changes forever after. This post builds that out with the Email API and shows the CLI shortcuts for tes
In my last post I described graphlens — what it does, how it works — and along the way I casually claimed that an agent "burns tokens grepping around a repo." I gave exactly zero numbers to back that up. This post fixes that. Here are the measurements, the data, and a reproducible harness. Spoiler: the conclusion is not the one I expected going in, and that's the interesting part. I took one agent (Claude Code), changed exactly one thing — which MCP server feeds it code context — and ran it over
Let’s start with a conversation
An idea, a challenge, or a system that needs to work better. We’ll help you understand the next step.
Prefer email? [email protected]
