OpenAI's recent ChatGPT Work release, running on GPT-5.6, joins Microsoft, Anthropic, and others in deploying AI agents targeted at non-coder knowledge workers and developers. While consensus exists on who manages the agent runtime and policy, significant differences remain around user data custody and persistence, impacting cloud cost, reliability, and observability.
- Agent runtime control centralized by vendors with divergent data ownership models
- Four deployment archetypes influence cloud service costs and developer experiences
- Enterprise agents emphasize governance and deep workflow integration
Infrastructure signal
The deployment of AI agents across industry leaders such as OpenAI, Microsoft, Anthropic, Amazon, and others presents clear infrastructure patterns differentiated by user archetypes. Cloud-hosted runtimes dominate for knowledge workers and enterprise users, consolidating operational responsibility and state persistence within vendor-managed environments. This centralization encourages consistent reliability and streamlined observability across multi-tenant infrastructures.
In contrast, power users who self-host agents introduce hybrid infrastructure models where runtime control resides partially or fully on user hardware, such as personal servers or laptops. This approach increases operational complexity but offers potential cost savings by offloading compute from the cloud and enables tighter credential management. Developer archetypes utilize both local IDE-based runtimes and cloud sandboxes, further diversifying infrastructure footprints and necessitating robust API gateways to harmonize local and remote executions.
Developer impact
Developers working with these AI agents must adapt to varying deployment and control paradigms. For knowledge-worker-oriented tools like ChatGPT Work or Claude Cowork, the developer’s role is mostly extensibility and integration within cloud ecosystems, relying on vendor-managed runtimes for scaling and observability. This reduces direct operational overhead but limits deep customization or data residency control.
Conversely, the self-hosting and developer archetypes demand a more hands-on workflow involving runtime setup, credential orchestration, and memory persistence. Developers managing these environments gain granular control over API integrations and data flows but must invest in maintaining reliability and deployment tooling. Observability strategies shift toward distributed tracing across local development environments and cloud sandboxes, requiring enhanced telemetry frameworks.
What teams should watch
Teams must closely evaluate agent deployment archetypes in relation to their cost, compliance, and operational priorities. Cloud cost management differs substantially when runtime and state reside in vendor clouds versus on-premises or user-owned infrastructure. Anticipating these variances informs budgeting and capacity planning, especially as long-running agent sessions with persistent state become normative.
From a reliability and observability perspective, teams should monitor vendor roadmaps for agent frameworks supporting hybrid runtime models and better policy enforcement mechanisms. API and platform decisions will increasingly drive integration ease and security postures, especially for enterprise workflows embedding AI agents. Staying current on evolving control plane ownership and data export capabilities ensures flexibility and alignment with organizational data governance standards.