DIGNALEGI

The reading room · Digna Legi

Harnesses are Situated Agents

A personal relevance score

80–100: high value. 70–79: worth the time. Below 70: below the usual publication threshold.

Evidence-reviewed score based on available publisher text. Evidence appears to be the source page, but the discussion includes many current product examples that may date quickly.

Scores reflect one reader’s profile, not an objective quality rating. Best is a separate personal selection.

How scoring works →

This brief · about 1 min with detail

Original article ↗

Why read this

Coding harnesses may be better understood as situated agents: infrastructure that shapes context, execution, memory, teams, and persistence.

AI brief · Checked against source text

The main idea

The author argues that a harness is best understood as a situated agent: the agent’s core loop remains prompt, planning, files, and subagents, while the harness manages the surrounding world. That surrounding world includes sessions, environments, repositories, memory, skills, teams, organizations, and models, shaping what enters context, how work runs, and what persists.

Some background helpful. Some familiarity with coding agents.

Go a little deeper

Stickiness moves outward

The useful distinction is that inner agent components are easier to swap than the surrounding operational layers. Once a team has configured sessions, permissions, sandboxes, shared traces, organizational policy, and accumulated memory, changing the harness means disturbing work practices, not just replacing a model or editor surface.

How the case is made

The case is made through observation of recent coding harnesses and a layered taxonomy of what they manage.

Where the idea has limits

The stickiness claim is argued by analogy to software platforms, not demonstrated with adoption or switching data in the supplied text.

What the original adds

The source names concrete harness experiments across multiplayer channels, policy control, modular components, and model co-training.

About this brief

AI-written, then separately checked for source support, useful detail and clarity. The author’s claims and our editorial question are kept separate. The original remains the author’s work. How we select and summarise →

Digna legi. Worth reading.