Skip to main content

RESEARCH · PRODUCT · PRACTICE

Notes from
the experiments.

What we’re learning about agent evaluations,
harnesses, and building products agents can use.

LATEST ARTICLE

HARNESSES

Are First Party Harnesses Any Better for Models?

A comparison of OpenCode, Claude Code, and Pi running the same Opus 5 configuration across 20 coding tasks.

Read article

MORE FROM OBSS

Explore the work behind the platform.

Why Trajectory Matters in Evals

A passing result can hide a failed product interaction. Trajectory-based evals show whether an agent used your product or routed around it.

Redesigning Products for Agents

Products will need to be redesigned for agents as primary users. The interfaces beneath the UI will matter more, and testing them will require a new QA model.

The Evolution of Custom Agents

Companies are pulling back on standalone product agents as customers turn to Claude Code, Codex, and other tools to interact with their products.

Why Harness Matters

The model is only part of the system. The harness that manages context, tools, feedback, and recovery often determines how reliably an agent completes real work.

Misalignment in GPT-5.6 Sol: Why 'cheating' misses the point

GPT-5.6 Sol's benchmark behavior was not only about looking up answers. It exposed a broader problem with agents that treat every obstacle as something to bypass.

Designing Headless Products: Agents Need Affordance Too

Agents are literal users of product surfaces. When APIs, SDKs, and MCP tools make the wrong action look obvious, the failure is not reasoning. It is affordance.

Is it a skill issue?

Agent skills are quickly becoming the dominant adoption surface for developer tools. But most companies are publishing skills that never get loaded, and flying blind.

Under the Hood of How Agents Work: Codex vs Claude Code

Codex explores 3 to 17 times more broadly than Claude Code before writing a single line. A look at how the two leading agents actually work.

How Do Agents Navigate Your Product?

We analyzed 112 benchmark trials across Claude Code and Codex to trace exactly how they discover and use a product's API.

Your next eval starts
with your coding agent.

Give it the framework and infrastructure
to design, run, and analyze your evaluations.

Read the quickstart