In coding agents, /compact replaces chat history with a summary built from the existing cache, then invalidates that cache. That’s the downside, but it’s still a net win, future prompts send far fewer tokens. Manual compactions at natural breaks work better than auto-compaction.

Been looking into LLM token costs in production and noticed something small that adds up. Spelling mistakes in prompts can quietly increase token usage because tokenizers break typos into many more sub-tokens than clean words. One piece showed this can push usage up 50–70% in some cases, plus extra clarification rounds. medium.com/@himanshusin/typo…

“Fairness” in queues sounds like one portable idea. It isn’t! RabbitMQ fairness and SQS fair queues solve two different problems: 🐇 RabbitMQ is fair to workers → don’t send more messages to a consumer already holding unacked work ☁️ SQS is fair to tenants / message groups → don’t let one noisy tenant increase dwell time for everyone else Same word. Different layer. This matters when designing shared queues: ✅ you can fix worker overload and still have noisy-neighbor latency ✅ you can protect tenants and still under-provision consumers I wrote up the mental model here: alagu.dev/blog/4_fair_queues…

Most deep links in modern consoles are not just navigation. They restore enough UI state to reopen the working view. That makes them useful as an agent-to-human handoff after API/CLI work. alagu.dev/blog/3_deep_links_…

We’re also launching Immersive Navigation - our biggest navigation upgrade in over a decade! A new vivid 3D view better reflects your surroundings, with helpful road details like lanes, crosswalks, traffic lights etc. Gemini models analyze real world imagery from Street View and aerial photos to give you an accurate view of landmarks along your route. Starts rolling out in the US today.

A few weeks ago, I wanted to learn something new and decided to explore building Chrome extensions. thought of creating one to help with my distraction issues where i was jumping between social media like Twitter, Facebook, Instagram while working. i was initially exploring whats available in the market but all extensions seemed to be forced and strict blockers. But that approach can backfire, leading to quick frustration. What i did : Not everything requires force. A bit of leniency with thoughtful friction can be more effective. My extension asks: "Why do you need this right now?" You give an honest answer, it analyzes it (simple check), and grants 1-5 minutes max. Or solve a quick math puzzle for access. That pause often makes me rethink and return to work. What if you added a small friction to your distractions?Share your thoughts. Link : chromewebstore.google.com/de…

Unpacking 7 levels of tech design. On the surface, this might look specific to frontend, but there's a lot of backend innovation behind the scenes. youtu.be/FQVRfadJkJk?si=JOg4…

Why does your client need rate limiting if the server enforces quotas? Because retrying after hitting a limit wastes resources. Proactive rate limiting is simpler. Read about the difference between exponential backoff and client-side throttling: alagu.dev/blog/1_client_side…

Brilliant post on prompt caching! This is one of the most effective and underutilized techniques for reducing LLM usage costs.

crazy how akinator was just decision trees and binary search, people underestimate the kinda things they can build without plugging in an llm in every project.

It's been about a year since my team has fully adopted all the AI coding tools (Cursor, Claude Code) And day to day I am feeling the added cruft in the code base. Unit tests are not catching regressions. Unneeded mocking, comments, are left in between. More refactoring is needed to add new features. I find myself sitting down and rewriting files to ensure completeness, correctness, and ease for future developers more than I ever have before.

very requested feature!