R00mi

4 ms·

1. 1.▲
   by R00mi 5mo ago

   The 56% conversation vs 21% coding split is a really interesting finding — it lines up with trajectory studies on SWE-bench where ~38% of an agent's actions are pure exploration (grep, find, file reads). The remaining "no-tool&quo
2. 2.▲
   by R00mi 5mo ago

   Bottom-up preparation :) - Plate — your product - Pancake 1 — the model - Syrup #1 — how the model emits the tool call (Harmony, JSON, XML-ish). Different flavor at every table. - Pancake 2 — the client/agent - Syrup #2 — MCP. Same bra
3. 3.▲
   by R00mi 5mo ago

   I agree with the core of what you're saying, but I think the real split isn't "Anthropic trustworthy or not" — it's: what's proprietary vs what's open in the stuff you're actually building on. Routine
4. 4.▲
   by R00mi 5mo ago

   MCP is the wire format between agent and tool, not the format the model itself uses to emit the call. That part (Harmony, JSON, XML-ish) is still model-specific. So the M×N the article describes is really two problems stacked — MCP only sol