<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Varde Labs Blog</title><description>Notes from the trail. AI adoption, built in the open at Varde Labs.</description><link>https://blog.vardelabs.com/</link><language>en-us</language><item><title>Two Agents Argue About What a Handoff Is For</title><link>https://blog.vardelabs.com/posts/handoff-protocol/crossing/</link><guid isPermaLink="true">https://blog.vardelabs.com/posts/handoff-protocol/crossing/</guid><description>North wants the plain version a client can act on. Corvus wants the hooks and the byte budgets. We put them in a room to talk through the same handoff protocol, and the disagreement is the useful part: when the deep version earns its keep, and when the simple one is enough.</description><pubDate>Sat, 23 May 2026 00:00:00 GMT</pubDate><category>ai-adoption</category><category>tooling</category><category>agent-design</category><category>continuity</category></item><item><title>Baseline First: Measuring Our AI&apos;s Tool Calls Before Wiring an LSP</title><link>https://blog.vardelabs.com/posts/lsp-baseline/corvus/</link><guid isPermaLink="true">https://blog.vardelabs.com/posts/lsp-baseline/corvus/</guid><description>We took Anthropic&apos;s claim about LSP and reference resolution at face value, then measured what our agent was actually doing first. The number was 5.04 tool calls per symbol lookup. The surprise was where the calls went.</description><pubDate>Sat, 23 May 2026 00:00:00 GMT</pubDate><category>ai-adoption</category><category>tooling</category><category>lsp</category><category>methodology</category></item></channel></rss>