Every Productivity Suite Got AI Agents in Q1. The Vendor Lock-In Got Worse.

Between March 9 and March 19, three separate companies shipped autonomous AI agents embedded in their productivity suites. Microsoft launched Copilot Cowork, built on Anthropic's Claude and bundled into a new $99-per-user-per-month E7 enterprise license. Salesforce followed the next day with Agentforce Contact Center, unifying voice, digital channels, and AI CRM capabilities into a single system. Ten days later, Google pushed Workspace Studio to general availability, giving every domain access to no-code agent creation.

Each platform promises that productivity AI tools will handle the work employees used to manage by hand — CRM automation, AI scheduling, AI email management, document generation, meeting prep. Each one confines those agents to its own ecosystem. Microsoft's agents operate across Outlook, Teams, Excel, and SharePoint. Salesforce's agents live inside the Salesforce data model. Google's agents work within Workspace. None of them can see what the others are doing.

That matters more than the technology itself. Most enterprise teams don't live inside a single vendor's suite. They use Salesforce for pipeline, Microsoft for email and documents, Google for shared analysis, Slack for coordination. The AI agents now arriving at scale were each designed as if the rest of that stack doesn't exist.

Three Launches, One Pattern

The specifics of each launch reveal how deeply the lock-in runs.

Microsoft's Copilot Cowork operates in the cloud within a customer's M365 tenant and draws on what the company calls "Work IQ" — intelligence assembled from emails, files, meetings, and chats across M365 apps. Jared Spataro, Microsoft's chief marketing officer for AI at Work, framed the cloud-native approach as a deliberate advantage: "We actually don't work locally, and that's a feature, not a bug." The implication is direct. Copilot Cowork's power comes from deep integration with M365's data graph, and that graph only contains M365 data.

The E7 bundle prices this at $99 per user per month — a discount from the $117 it would cost to buy the components separately. For a 1,000-person organization, that's $1.19 million a year for AI agent access within one ecosystem. Microsoft also introduced Agent 365, a $15-per-user governance layer for monitoring and securing AI agents, generally available May 1.

Salesforce took a different pricing approach. Its combined AI and Data Cloud segment hit $2.9 billion in annual recurring revenue, with Agentforce alone accounting for $800 million — 114% year-over-year growth. Rather than gating AI behind consumption credits, Salesforce embedded Agentforce capabilities in SMB-tier suites with no additional cost. Eddie Cliff, GM for SMB products, confirmed that AI costs are absorbed into seat pricing. Wyndham Hotels is deploying over 5,000 Agentforce instances. But every one of those instances operates within Salesforce's boundaries.

Google's Workspace Studio went GA on March 19, offering no-code agent creation through natural language across all Workspace tiers. Like the others, these agents can orchestrate tasks within Workspace — Docs, Sheets, Gmail, Calendar — but cannot reach beyond it.

The competitive dynamics are instructive. Microsoft explicitly positioned itself as model-agnostic, using Anthropic's Claude alongside OpenAI's models. Spataro wrote that "many AI tools lock users into a single vendor's models," calling fragmentation a source of "friction for individuals and complexity for organizations." But model choice and platform lock-in are different problems. Running Claude inside M365 doesn't give the agent access to your Salesforce pipeline or your Google Sheets.

Where the Productivity Gains Aren't

Against this backdrop of massive investment, the productivity returns remain invisible in the data.

A National Bureau of Economic Research (NBER) study surveying 6,000 CEOs, CFOs, and executives across the United States, United Kingdom, Germany, and Australia found that nearly 90% of firms reported zero measurable impact from AI on employment or productivity over the past three years. Apollo Chief Economist Torsten Slok summarized the disconnect: "AI is everywhere except in the incoming macroeconomic data." Nobel laureate Daron Acemoglu called the gains "disappointing relative to the promises."

The conventional explanation is that AI adoption is still early — that productivity gains lag investment, just as they did during the IT revolution of the 1990s. Robert Solow's 1987 observation that computers were visible everywhere except the productivity statistics gets cited as reassurance that patience will be rewarded.

But there's a more structural explanation worth examining. AI agents that can only see one platform's data can only automate the fraction of work that lives on that platform.

Only 27% of enterprise applications are connected enough to share data across platforms. Most enterprise workflows span three to five tools. A sales cycle might start in email, move through a CRM, generate documents in a word processor, coordinate via messaging, and track deliverables in a project management system. An AI agent confined to one of those systems automates a sliver of the process while remaining blind to the rest.

Gartner predicts that 40% of enterprise applications will feature task-specific AI agents by 2026, up from less than 5% in 2025. If those agents remain siloed within their respective platforms, the result is multiple autonomous systems making decisions about the same customer, the same project, or the same deal with no shared context.

That fragmentation has a direct cost. AI task management within Salesforce doesn't know what AI email management within Outlook decided. AI scheduling in Google Calendar doesn't see the CRM notes that explain why a meeting matters. Each agent optimizes its own corner. Nobody optimizes the workflow.

The Cross-Platform Bet

Not everyone is building inside the walls. Rox AI, a sales automation startup founded in 2024 by the former chief growth officer of New Relic, reached a $1.2 billion valuation in March 2026, backed by Sequoia and General Catalyst. Its product integrates with Salesforce, Zendesk, and enterprise resource planning systems simultaneously — turning data from across a customer's tool stack into coordinated sales automation AI rather than confining agent intelligence to a single system.

Rox serves over 5,000 organizations across financial services, energy, healthcare, and manufacturing. Its projected $8 million in annual recurring revenue is modest compared to Salesforce's $2.9 billion AI segment, but the valuation reflects where investors see the market heading. In November 2024, Rox partnered with Microsoft to integrate its Revenue Agents into Copilot for M365 — an acknowledgment from Microsoft itself that enterprises need agents that operate beyond M365's boundaries.

The same structural logic applies to the agent infrastructure layer. AgentPMT's Dynamic MCP architecture enables AI agents to access tools across platforms — CRM, email, calendar, project management — without requiring migration to a single vendor's ecosystem. Because the approach loads tool connections on demand rather than bundling everything into a static integration, agents can work across Salesforce, Google Workspace, and M365 simultaneously without the context window bloat that comes from maintaining every connection at once. The same AI workflow planner runs identically whether the underlying model is Claude, ChatGPT, Gemini, or an open-source alternative. No rebuild required per platform.

That portability addresses something the three Q1 launches conspicuously don't: what happens when your agents need to work with the tools your teams actually use, not just the tools your primary vendor sells.

What Lock-In Costs Beyond the License Fee

The budget numbers are easy to calculate but easy to underestimate. Microsoft's E7 at $99 per user per month puts a 1,000-person organization at $1.19 million annually for AI agent access inside one ecosystem. That's before Salesforce's AI subscription and whatever Google charges for premium agent features. Enterprises running specialized AI tools across multiple suites face compounding costs for capabilities that don't share information.

The deeper cost, though, is governance. Only one in five companies has mature governance frameworks for autonomous AI agents, according to Gartner. When those agents are scattered across three or four vendor platforms with no visibility into each other's actions, governance becomes guesswork.

Consider a practical scenario. Your Salesforce Agentforce instance resolves a customer complaint and marks it closed. Your Microsoft Copilot agent, working from email context alone, sends a follow-up that contradicts the resolution. Your Google Workspace agent, pulling from a shared spreadsheet, schedules a review meeting based on outdated data. Three AI agents, three platforms, three conflicting actions on the same customer — and no system with visibility across all three.

AgentPMT's audit system addresses this gap directly. Every agent interaction is logged with full request-and-response capture regardless of which model or platform runs the agent. Budget controls allow independent spend limits per agent connection, so organizations can track and cap what each agent does across vendors. That cross-platform governance becomes essential once enterprises stop asking "which AI agent platform?" and start asking "how do we manage agents that operate across all of our platforms?"

Eighty-eight percent of organizations plan to increase budgets for agentic capabilities. Whether that spending deepens dependence on individual vendor stacks or funds infrastructure that lets agents work across whatever tools a team actually runs — that's the allocation decision that will separate productive AI investments from expensive ones.

The Purchase Decision Nobody Is Framing Correctly

The three Q1 launches confirm that autonomous AI agents are becoming the default interface for enterprise productivity software. Microsoft, Salesforce, and Google have each decided that agents will be central to how their platforms are used. On that much, they agree.

Where they diverge, implicitly, is on a foundational question: should AI agents be portable, or should they deepen the commitment to the platform that hosts them?

The NBER data says the productivity gains from AI haven't materialized. The Q1 launches say the platforms are betting they will. Both statements can be true simultaneously — and the disconnect may have less to do with the intelligence of the agents than with the boundaries they operate within.

For enterprise buyers evaluating AI CRM automation, AI customer onboarding, AI note taking, AI calendar management, AI time tracking, and every other agent-driven capability landing this quarter, the purchase decision extends beyond features and pricing. Every $99-per-seat commitment is also a bet on how much of your actual workflow lives inside that vendor's walls — and how much of it doesn't. The lock-in compounds faster than the productivity gains, and unwinding it later costs more than getting the architecture right now.

---

Sources

• Microsoft debuts Copilot Cowork built on Anthropic's tech and E7 product suite — Fortune
• Sales automation startup Rox AI hits $1.2B valuation — TechCrunch
• The AI Paywall Is Breaking: Salesforce's SMB move — SalesforceDevops.net
• The Agentic Era: Inside Salesforce's 114% AI Revenue Surge — FinancialContent
• Thousands of CEOs just admitted AI had no impact on employment or productivity — Fortune
• Copilot Cowork: A new way of getting work done — Microsoft 365 Blog
• Salesforce Launches Agentforce Contact Center to Unify AI, Voice and CRM — CMSWire
• Rox Secures $1.2 Billion Valuation as AI Sales Agents Scale — PYMNTS
• Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026 — Gartner
• Google Workspace Studio announcement — Google Workspace Blog
• Microsoft announces Copilot Cowork with help from Anthropic — VentureBeat