My MCP Stack: How I Run My Own Agent Tools
Autonomous coding agents need local discipline: honest token accounting, governed skill drawers, and repeatable quality gates. Here is how I run Thermal, Skill Cabinet Go, and a...
Archive
Every published piece in chronological order, now with topic and year filters for readers who want more than a linear scroll.
Index
16 posts
Use this page when you want the full chronology, filtered by theme, year, or a specific phrase.
Discovery
16 essays currently visible.
Autonomous coding agents need local discipline: honest token accounting, governed skill drawers, and repeatable quality gates. Here is how I run Thermal, Skill Cabinet Go, and a...
Centralizing agent skills in global vaults creates a brutal trilemma: catastrophic prompt token bloat, silent alphabetical truncation past fifty tools, and the total blindspot o...
We handed our terminals over to autonomous coding agents, only to discover our token bills were compounding in silence. Here is how I built Thermal to track 14 agent harnesses w...
Skills help agents by stabilizing action, not by injecting facts. But the same abstraction that makes them useful also creates a new failure surface. A deep dive into the Prince...
Reading Microsoft's modern stance on open weights feels surreal when you remember the 2001 WhiteHouse.gov Linux migration and the 2006 Novell deal. Here is why the modern 'open...
A memory of the 1991 Gulf War in a quiet Jabal Amman stair-alleyway, and how screens have atomized our collective response to regional crisis into solitary background noise.
Latency is usually described as a performance metric. From the edge of the network, it starts to feel more like a power relationship: a reminder of who gets to live close to the...
From Amman, the current open-source AI wave feels uncannily familiar. The same builder energy, the same elite gatekeeping, and the same fight over who gets to shape the future o...
Working with Jules gave me a concrete look at what AI coding agents are actually good at: absorbing context, handling the tedious middle, and forcing human engineers to operate...
A grounded look at the AI experiments that have been occupying my evenings: private proxies, Arabic-language reality checks, and an early podcasting setup powered by equal parts...
Arabic sentiment analysis sounds straightforward until dialects, morphology, and limited datasets enter the picture. This piece breaks down why the problem remains both difficul...
Long before ChatGPT, ELIZA showed how quickly humans project empathy, intention, and understanding onto software. That instinct still shapes both the appeal and the risks of mod...
A late-night dialogue with Claude 3.5 Sonnet pushed me past the usual AI hype and into the harder questions: opacity, variability, auditability, and how to use these systems wit...
What began as technical curiosity turned into something more personal: AI art gave me a low-friction way back into visual creativity and changed how I think about who gets to ma...
Watermarking systems like SynthID will not solve authenticity on their own, but they may become one of the few practical tools we have for separating provenance from pure guessw...
After two decades away from regular blogging, I found myself returning to the medium that first taught me the value of owning my voice, my perspective, and my little corner of t...
No essays matched that combination. Try clearing the filters or broadening the search terms.