Agent Ops: autoresearch, software factories, MCPs, and governance

Autoresearch: The feedback loop behind self-improving agents. Introspection champions autoresearch loops where agent recipes, evals, and human feedback let systems self-improve in production. Outcome engineers should build closed-loop evals and feedback channels to let agents iterate safely and measurably — Principle 09.

Warp CEO Zach Lloyd on why software factories are the next phase of coding. Warp’s Oz automates software engineering loops, treating agents as continuous delivery lanes that integrate with developer workflows. Adopt the software-factory pattern to design agent orchestration, CI controls, and artifact flows that scale teams — Principle 09.

SnapLogic MCP Builder eases creation of MCP servers. SnapLogic converts integration pipelines into agent-ready MCP servers with governance, observability, and identity propagation. Use MCP builders to reduce integration friction while preserving auditing and access controls for production agents — Principle 06 & 11.

Safari’s new MCP server lets coding agents inspect and debug websites. Apple exposes page content, console logs, network traces and screenshots to coding agents for automated debugging. Browser-level MCPs change what agents can do on the web, so engineers must design secure capability boundaries and test harnesses around in-browser agent actions — Principle 06.

Preventing agent-generated infrastructure bloat through spec-driven governance. The piece argues embedding sustainability and deployment constraints directly into agent-executed specs to avoid uncontrollable post-deploy bloat. Outcome engineers must bake cost, resource, and approval gates into agent specs to prevent runaway infrastructure and enable auditability — Principle 10 & 15.