Notes from the build.
What we are building, how we measured it and what we got wrong, written as it happens. Every post names its sources and carries its date. A number carries its tag where its source does. The tags are measured, target and provisional.
- Bench
Two clocks over the field
8GEN Bench runs two long-horizon clocks. One stands over every frontier model an agent could think with and asks whether it holds up over days. The other stands over every platform an agent could live on and asks how long the first thought takes. Here is how each clock is read, and why none of its figures is ours to round.
3 min
Markdown - Trust
Seven of eight witnesses
An agent’s own account of what it did is one witness. Seven more stand on the host side of the kernel boundary, where the agent cannot reach. What they see goes into a hash-chained ledger, stamped with its epoch, session and policy. This post walks the eight and explains why the record holds even when the agent is compromised.
2 min
Markdown - The category
Beyond sandboxes
Sandboxes rent execution by the second and are built to end. An agent that is meant to work for weeks needs a home, not a rental. This is the argument for the category 8GEN is building in, and for the word we use for its unit.
2 min
Markdown
Follow the blog
RSSJSON FeedMarkdown indexJSON indexevery post also at its address with .md and .json