I Thought OpenClaw Would Choke on My Research Project. It Didn't.

Published: 2026-02-08 23:04

I've been running OpenClaw on a new API key funded by work. The use case is different from my coding setup: Open Code handles the coding, OpenClaw handles administration tasks for my job.

Today I was thinking about a research project that needs to get done. Lots of steps involved. URL lookups, screenshots, information gathering across multiple sources. The kind of task that eats context for breakfast.

Here's the thing: OpenClaw loads a bunch of files from the start. Identity file, memory file, soul file, and more. So I was concerned the context would rot quickly. All that baseline overhead plus a multi-step research task? Seemed like a recipe for hitting the dumb zone fast.

But after many turns against Claude Opus 4.5 (asking questions, feeding it info, building out a file I could actually use for work) the context was sitting at just 61k tokens.

Well within the dumb zone.

I expected worse. Pleasantly surprised.