Latest Articles
- 24 days agoAI Coding
Wrong Tool Calls Are a Retrieval Problem
Agents do not reason over a menu of tools. They match against descriptions competing for attention, which means the fixes are indexing fixes and not prompt fixes.
- 27 days agoProgramming
Driving a Browser With an Agent Is Still Specification
The natural-language prompt is usually longer than the test it produces. What an agent actually saves you is locator work, and what it costs you is a stable result.
- about 1 month agoAI Coding
Eval Vocabulary for People Who Already Test Software
Most AI evaluation terms are QA concepts with new names. Here is the translation table, and the three places where the analogy quietly stops holding.
- about 1 month agoAI Coding
Your Import List Is Not Proof That You Reused Anything
One module imported four others and executed zero lines from one of them. What it costs to assemble nine independently built parts, measured instead of claimed.
