Fixed teams -> executable mandates

Work has traditionally inherited the topology of the organization. A request enters a functional queue, moves through stable teams, and accumulates context inside the people and systems that own that route. Goals coordinate the work, but they remain comparatively thin objects: a target, document, ticket, or metric.

The emerging alternative gives the goal itself more operational substance. A bounded mandate can hold addressable state, task-scoped authority, a resource envelope, attachable capabilities, a resumable workspace, acceptance checks, and a durable record. It can then assemble people, agents, tools, and machines for the current stage of work rather than inheriting one fixed execution path.

What changed

The individual primitives now connect. Background agent runs have stable IDs; shared agents can operate on schedules; task-scoped credentials can expire; sandboxes and machine pools can be created for a request; workflows and skills can package reusable expertise; and accepted artifacts can be bound to tests, approvals, and audit history.

Recent organizational cases add a second layer of evidence. Asana describes parallel agents compressing a migration campaign while humans reviewed each change. loveholidays describes domain practices and validations packaged into workflows that non-engineers can use against a real codebase. The .NET team reports useful agent-authored pull requests, but also shows that maintainers, task selection, repository instructions, tests, and review remain decisive.

What stays rigid

The shift does not make the whole company fluid. The execution topology can be temporary; the record cannot. Goal identity, authority, provenance, evidence, acceptance, and institutional writeback must survive changes of model, worker, session, and team membership.

This is the key inversion: flexibility at the execution layer requires more rigidity at the mandate layer. The goal behaves like an organizational unit only because its autonomy is bounded by external memory and controls.

Current boundary

The category is still a fragile tool, not boring infrastructure. Public agent tooling is broad, but Cohere’s strict evaluation found little end-to-end occupational coverage. The closest formal architecture, Fluid Structure, Rigid Record, remains a prototype supported by small-sample experiments. Current case studies are concentrated in software and frontier organizations.

Raise this shift to accelerating when several independent organizations can show that the same mandate survives executor changes, uses scoped authority, produces accepted work with bounded review cost, and improves later mandates through governed memory. Lower confidence if review queues, permission administration, or stale goal state grow faster than accepted organizational change.