Most enterprise digital workspaces today were architected for one operator pattern: a human at a browser, signing in, clicking through interfaces, and producing work one keystroke at a time. That assumption is breaking down. Autonomous agents now compose emails, reconcile expenses, draft contracts, run code reviews, and operate internal systems on behalf of knowledge workers. The workspace has become a multi-agent execution environment running on cloud infrastructure that was not designed for it.
CEO and Co-Founder
Island Computing
The workload underneath
The first shift is in the workload itself. Through 2024, the dominant AI workload inside the workspace was model inference: a short, stateless, latency-sensitive request that produced a token-stream response. From 2025 onward, the dominant workload is agentic: long-running processes that hold context across steps, call tools, query systems of record, and recover from failures. A customer support agent in production today executes hundreds of tool calls over a session that can last minutes or hours. A code agent reviews diffs, runs tests, and proposes deployments. A knowledge agent reads documents, queries databases, and writes back to enterprise systems. Each workload exhibits long-running state, non-deterministic resource consumption, and access to data the original cloud security model never anticipated.
What the security model has to absorb
The security model has to absorb three structural changes. The first is identity. The IAM frameworks that enterprises run on today were designed for two principal types: humans and services. An autonomous agent is neither. It acts on behalf of a human, with delegated authority, across multiple systems, in sessions that can outlast the human’s login. The cloud control plane has to issue scoped, time-bound credentials that the agent can use within a narrow blast radius, and the credential lifecycle has to be managed automatically. The second is audit. A single agent run produces a tree of model calls, tool invocations, memory reads and writes, retries, and human-in-the-loop checkpoints. Audit infrastructure designed for request-response logging cannot reconstruct what the agent did or why. The third is the deployment posture for agent-initiated change. When an agent proposes a configuration change, a data update, or a code deployment, the safety boundary moves from human approval to policy enforcement at the platform layer. Progressive rollout, canary analysis, and policy-as-code stop being best practice and become required controls.
The data and sovereignty layer
The data and sovereignty layer is the second structural change, and the one most consequential for Indian CIOs. Agent quality scales with the richness of the data the agent can access. For Indian enterprises, that data sits inside regulated systems: banking transactions, patient records, citizen identity, payments, customer correspondence, internal financials. The DPDP Act, RBI’s data-localization circulars, SEBI’s cloud advisories for capital-market firms, and CERT-In’s six-hour incident-reporting directive each tighten the requirement that this data stay in India, under Indian law, on infrastructure where the operator can be reached during Indian working hours. Foreign-controlled stacks cannot fully meet that requirement, and 87% of India’s cloud market still runs on them today (IDC, 2025). The sovereignty question for digital workspaces is settled. What remains is what to do about it before the next compliance audit.
The architectural answer
The architectural answer has three components, and all three need to be designed in from the start. The first is a sovereign cloud substrate with control plane, identity service, and audit log designed and operated within Indian jurisdiction. Data residency lives in the platform itself. Compliance does not require a contract addendum. The second is an agent-native identity and access framework: scoped credentials with time-bound permissions, automatic credential rotation, and audit traces structured around agent runs rather than request logs. The third is a cloud substrate purpose-built for agentic workloads: long-running execution environments, durable state stores, tool sandboxes with admission control, and orchestration layers that schedule across compute, cost, and tool throughput at once. The serving stack designed for short stateless requests cannot carry these workloads at production scale.
The role of cloud computing
Cloud computing has shifted from being the cost-efficient hosting layer for SaaS applications to being the determinant of how secure and capable an AI-ready digital workspace can become. The cloud substrate now defines whether agents can be granted scoped authority and audited end-to-end, whether sensitive workspace data stays inside the jurisdiction that regulates it, and whether the workload patterns of the next five years run at production economics. The conditions for getting this right are aligned now: the senior engineering talent that operated cloud and AI platforms at global scale is starting to return to India, capital is moving into sovereign cloud and deep tech, and regulatory frameworks are arriving in a coherent direction.
Takeaways for technology leaders
For enterprise technology leaders, three takeaways follow directly. The cloud underlying the digital workspace will determine the security posture more than any application-layer control. Agentic workloads require execution and identity primitives that legacy cloud architectures do not provide, which makes the cloud platform a strategic capability decision rather than a procurement decision. Sovereign infrastructure has moved from a compliance hedge to a design requirement for any workspace touching regulated data.
The cloud architecture for the AI-ready workspace is being defined now. Indian enterprises that align with sovereign, agent-native cloud infrastructure today set the security and performance baseline for the next decade of knowledge work. The role of the cloud has become the role of the workspace itself.
–Authored by Vishal Sirohi, CEO and Co-Founder, Island Computing