Anthropic is betting that the main bottleneck for enterprise AI is not model capability but the fact that most people still use AI solo. In an exclusive interview with VentureBeat, Scott White, Anthropic's head of product for enterprise, described the company's case for "multiplayer AI": moving away from single-user chatbots toward agents that operate across teams, read organizational context, and proactively insert themselves into work—sometimes without being asked.
What changed with Claude Tag
This strategy became concrete earlier this month when Anthropic updated Claude Tag, its Slack-resident agent, to read context across entire conversations rather than evaluating messages one at a time. According to the company, the change makes Claude about 30% better at deciding when—and importantly, when not—to intervene. White says this is more than an incremental product update: it's the point where AI stops being a personal tool and starts acting like a colleague deployed by the organization.
"Claude used to feel like your personal chief of staff," Scott White said. "Now Claude, in the context of organizational deployment, feels like the company's chief of staff."
How enterprise AI has evolved
White frames the industry evolution in three phases. At first AI handled part of one task—answering a chatbot question or completing a line of code. As models improved, they handled whole tasks—writing a function, producing a report, drafting documents from multiple sources. Now, White argues, models plus system connectivity enable higher-order, goal-oriented work: keeping a product bug-free or accelerating NDA review, with models operating in the background and connecting data sources agentically.
Goals tend to involve multiple people, and knowledge work is messier than software engineering, which benefits from decades of collaboration infrastructure like Git. "Knowledge work is much messier than software engineering," White said. "It's multi-person. It's multi-job function. It's multi-connected systems that have different permissions... The output is also not as easily verifiable as code."
Three technical pillars behind proactive agents
White identified three breakthroughs from the past two years that enabled the multiplayer shift:
- Connectivity: Anthropic's Model Context Protocol (MCP), a public standard the company introduced in late 2024 and that OpenAI and Google adopted in 2025 according to White, has matured into what he called "the USB-C for AI connectors," giving Claude governed access to enterprise data systems.
- Model intelligence: a level of capability that makes proactivity useful rather than annoying, so the agent can connect dots across diverse data sources and identify problems it can fix.
- Form factor: placing Claude where collaboration already happens. The Slack integration gives Claude a federated agent identity with its own permissions, channel awareness, and MCP-connected context.
The Claude Tag update ties these elements together. Previously, a lightweight classifier evaluated each Slack message in isolation and made a binary decision about responding. Anthropic removed that classifier: Claude now reads the channel's full context, plus its memory and standing instructions, and chooses among four moves—reply inline, start deeper work in a thread, route the message into an existing workstream, or say nothing. Anthropic illustrates this with two engineers chasing the same bug from opposite directions; individually neither message warrants a response, but read together one has a theory and the other has evidence—Claude opens a thread and begins the investigation. The company also built explicit restraint: Claude goes dormant in channels where it repeatedly has nothing to add.
Why collaborative AI can beat solo high performers
Skeptics note that top performers often move fastest alone and coordinating AI use across people is hard. White points to his own experience inside Anthropic: where he once needed a data scientist to run a day or two of analysis, Claude can now do the analysis and the data scientist has built infrastructure that makes Claude reliable at that task.
That shortens handoffs and shifts time toward judgment: "More of my time is spent with my collaborators at my level, jamming with them about what we should actually do and what we should change about our strategy... The multiple handoffs that we saw to get a piece of work done to then try to make a decision — all of that has been shortened dramatically," White said. He cited site reliability engineering as a clear example: Claude collects error logs, connects them to recent code changes and related Slack conversations, synthesizes the information, and pulls in the right people to solve larger, goal-oriented problems.
Market context and adoption metrics
Adoption remains mixed. McKinsey's latest State of AI survey found 88% of organizations report using AI in at least one function and 62% are at least experimenting with AI agents, but only 39% attribute any earnings impact to AI and just 6% are "high performers" seeing significant value. Deloitte predicted that 25% of enterprises using generative AI would deploy agents in 2025, doubling to 50% by 2027. Anthropic aims to win on redesigned workflows rather than marginal model improvements.
Defending always-on agents against attacks
Giving an agent standing access to Slack channels, documents, and connected systems raises prompt-injection and other risks. White described a layered defense: models are trained with classifiers to reduce prompt injection; Anthropic delayed the general availability of its Chrome extension and placed it on a waitlist while they studied browser-context risks and performed novel classifier training from that dataset.
Beyond model-level defenses, Anthropic exposes compliance and analytics APIs so customers can set risk thresholds and integrates with third-party security vendors and data loss prevention. Claude's permissions collapse to the most restrictive intersection of what the agent can see and what the requesting user can see, preventing cross-channel leakage of context.
Pricing and the question every CIO should ask
A notable detail in the Claude Tag announcement is that the expanded channel context Claude now holds does not currently count toward usage or spend limits on any plan. When asked whether that context will become billable, White declined to commit, calling the phase an active experimentation and co-design period with customers. Anthropic emphasizes controls—budget caps tied to agent identities or role-based groups and model entitlements that let teams run different cost-performance profiles.
The honest translation is that the unit economics of always-on, context-heavy agents remain unsettled and Anthropic is effectively subsidizing the learning phase. Enterprises should architect deployments assuming the meter could eventually be turned on.
Competition with platform owners
Competition is real: Salesforce owns Slack, Microsoft is embedding agents in Teams, and Google is doing similar work in Workspace. White argues that value comes from connecting multiple systems—pulling call transcripts from Salesforce, internal debate from Slack, usage analytics from product tools, and code from development environments—so a neutral orchestration layer built on a standard like MCP can have structural advantages over a walled garden.
Platform owners can still bundle agents for free and control integration points Anthropic depends on. Anthropic's bet is that frontier intelligence plus cross-system neutrality will beat distribution; that thesis has worked in coding but remains unproven for general knowledge work.
What's next: giving Claude company OKRs
Asked about the future, White sketched a trajectory in which delegation flows both ways: "We're now giving Claude projects to accomplish. Soon, I think, we'll give Claude its OKRs, and it will figure out which projects the organization needs to work on and how to connect people to do them—improving itself, and the people around it, over time." He acknowledged the idea is abstract and raises governance questions that current controls only begin to address.
For now the immediate change is already visible in Slack channels: an AI that listens to the whole room, decides when to speak, and increasingly does connective work that used to fall to managers, analysts, and project leads. White has seen that compression reduce his role to its most human remainder—judgment, debate, and decisions—and he expects similar shifts elsewhere.
Two years ago the industry's favorite AI metaphor was the intern. Anthropic has effectively promoted that role to chief of staff—and made it report to the whole company.
