<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Kinesthetic — Blog</title><description>The knowledge layer for specialized AI</description><link>https://kinesthetic.dev/</link><item><title>Specifying the agent: closing the gap to frontier with a context engine and meta-distillation</title><link>https://kinesthetic.dev/blog/specifying-the-agent/</link><guid isPermaLink="true">https://kinesthetic.dev/blog/specifying-the-agent/</guid><description>Search and learning algorithms that attack specification failure directly (a context engine and meta-distillation), evaluated across the τ³-bench banking suite, FinanceBench, and the Harvey Legal Agent Benchmark on two open / open-weight backbones.</description><pubDate>Mon, 15 Jun 2026 00:00:00 GMT</pubDate></item><item><title>The correction loop is the bottleneck</title><link>https://kinesthetic.dev/blog/the-correction-loop/</link><guid isPermaLink="true">https://kinesthetic.dev/blog/the-correction-loop/</guid><description>The path from an expert&apos;s correction to a shipped change in agent behavior (a spreadsheet of traces, a Jira ticket, a barely tested prompt edit) is the real constraint on enterprise agent deployment. Why that loop exists, why it&apos;s the bottleneck, and what replaces it.</description><pubDate>Wed, 29 Jul 2026 00:00:00 GMT</pubDate></item></channel></rss>