The enterprise software landscape is undergoing a quiet but decisive shift. While multi-user support has long been a baseline feature—something as expected as encryption or cloud hosting—the past six months have seen it evolve from a technical checkbox into a strategic differentiator. As organizations deploy AI agents, embrace asynchronous work, and grapple with regulatory pressure around data sovereignty, the definition of "multi-user" is being rewritten in real time.

The Agent Problem: When Non-Human Users Join the Roster

One of the most significant developments in late 2025 is the formalization of "non-human identities" (NHIs) within multi-user architectures. Major platform vendors—including Salesforce, Notion, and Atlassian—have begun shipping updates that allow AI agents to exist as first-class users with their own permissions, audit trails, and resource quotas. This is not a cosmetic change. In October, Salesforce’s Agentforce 2.0 introduced a dedicated "agent workspace" that separates human sessions from automated workflows, preventing the kind of session-collision bugs that plagued early AI integrations.

Industry analyst Dana Whitfield of RedMonk notes, "The old model assumed one human, one session, one device. Now we have a single human orchestrating five agents across three platforms, each agent needing read-write access to different data silos. If your multi-user support doesn't account for that, you're building on sand." Whitfield’s point is echoed by a recent Gartner survey, which found that 68% of organizations using AI copilots report "permission fatigue"—the inability to quickly grant or revoke granular access for machine users without disrupting human workflows.

Real-Time Collaboration Hits a Ceiling: The Shift to "Co-Presence"

Meanwhile, the consumerization of enterprise tools has pushed real-time collaboration beyond simple document co-editing. Google Workspace and Microsoft 365 both rolled out "presence-aware" features in Q3 2025, showing not just who is viewing a file, but which specific paragraph or cell they are interacting with. This granular telemetry, while useful, has exposed a critical flaw in legacy multi-user systems: lock contention. When 40 users simultaneously interact with a single spreadsheet, traditional row-level locking causes latency spikes.

To address this, a new wave of "conflict-free replicated data types" (CRDTs) is moving from academic papers to production. Startups like Croquet and Liveblocks have reported a 300% increase in enterprise adoption this year, with their CRDT-based engines now powering collaborative dashboards for logistics and financial trading firms. The key innovation is "operational transformation without a central server"—each user maintains a local replica, and changes are merged asynchronously. This not only improves speed but also enables offline-first multi-user support, a feature that has become critical for field workers in low-connectivity regions.

Regulatory Pressure: The EU AI Act’s Hidden Clause

Perhaps the most underreported driver of change is regulatory. The European Union’s AI Act, which entered its full enforcement phase in August 2025, includes a rarely discussed provision: Article 52 mandates that any system with multi-user access must maintain an "unbroken, immutable log of all human and machine interactions" for systems classified as high-risk. This has forced vendors to rethink how they store session data.

In response, several mid-tier enterprise platforms—including Germany’s Nextcloud and France’s Linagora—have introduced "multi-tenancy with cryptographic separation." Instead of a single shared database, each user (human or agent) gets an encrypted shard, and queries are executed via homomorphic encryption. While this is computationally expensive, it satisfies the audit requirement without exposing user data to other tenants. "We are seeing procurement teams explicitly ask for 'AI-Act-ready multi-user support' in RFPs," says Elena Marchetti, a compliance consultant at TÜV Rheinland. "It’s a new checkbox, but it’s reshaping architecture decisions."

The Trend: From "Concurrent Users" to "Concurrent Contexts"

Looking ahead, the most forward-thinking vendors are abandoning the concept of "concurrent user count" entirely. Instead, they are adopting a model called "concurrent context"—the number of distinct workflows, projects, or data domains that are simultaneously active. This shift is driven by the rise of "contextual AI," where a user’s AI assistant needs to switch between different project contexts without losing state.

For example, Coda’s November update introduced "Context Spaces," which allow a single user to maintain up to 12 isolated workspaces within one login, each with its own AI agent, permission set, and version history. This effectively multiplies the number of active "users" from a system perspective, but without the overhead of creating new accounts. "We realized that a power user is not one person—it’s a constellation of roles," says Coda’s VP of Product, James Park. "Multi-user support should be about supporting those roles, not the physical human behind them."

Expert Verdict: A Fragmented Road Ahead

Not everyone is convinced the industry is moving in the right direction. Dr. Anika Sharma, a distributed systems researcher at ETH Zurich, cautions that the rush to support agents and contexts may be premature. "We are seeing a proliferation of custom protocols for agent-to-agent communication, but there is no standard for cross-platform multi-user agent sessions. If you have a Slack bot that needs to fetch data from a Salesforce agent, you still have to write brittle glue code." Sharma points to the rise of "agent orchestration layers" like LangGraph and AutoGen, which are attempting to standardize this, but notes that adoption remains limited to technical early adopters.

Market Implications

The commercial stakes are high. Pricing models are shifting from per-seat to per-agent or per-context, a move that has drawn criticism from enterprise buyers. A November report from Forrester found that 54% of IT decision-makers expect their multi-user software spend to increase by 20% or more in 2026, but they are demanding more transparency in how "user" is defined.

"This is a golden opportunity for open-source alternatives," says Whitfield. "If proprietary vendors confuse customers with opaque usage metrics, we will see a resurgence of self-hosted tools like Mattermost and Zulip, which offer clear, predictable multi-user pricing."

Conclusion

The next 12 months will likely bring further consolidation. Expect to see major cloud providers (AWS, Azure, Google Cloud) introduce "multi-user-as-a-service" APIs that abstract away the complexity of agent identity, CRDTs, and audit logging. The winners will be those who recognize that multi-user support is no longer a technical detail—it is a business model, a compliance obligation, and a competitive weapon. For now, the message to CIOs is clear: audit your current platforms not just for "how many users," but for "how many kinds of users." The future is not about more seats. It is about more roles, more contexts, and more machines—all coexisting without chaos.