KEY HIGHLIGHTS

  • MCP (Model Context Protocol) is emerging as the universal adapter for enterprise AI — the connective layer that lets AI systems work across tools, data, and workflows without vertical lock-in.
  • Smartsheet launched its MCP integration with 36 tools, built for token efficiency, tool chaining, and full parity with its internal AI, not a checkbox wrapper.
  • The shift in enterprise AI is from ‘who has AI’ to ‘who is actually using AI’ — deployment metrics are giving way to usage and impact metrics.
  • Governance must be baked into architecture from the start, not bolted on after the fact, with every AI action auditable, reversible, and explainable.
  • In 18 months, knowledge workers will operate primarily through personal agent clusters, with AI handling routine workflows and humans focusing on judgment, creativity, and collaboration.
  • Token efficiency is the new cloud cost conversation, companies that optimize for surgical, precise AI usage will have a structural cost and performance advantage.

Enterprise AI investments are multiplying. But for many organizations, the returns are not keeping pace, and the reason is rarely the model. It’s the gap between where AI lives and where work actually happens. In a recent episode of our Age of AI series, I sat down with Drew Garner, SVP of Engineering at Smartsheet, to explore how MCP — the Model Context Protocol — is emerging as the connective infrastructure that closes that gap, and what it means for CIOs, enterprise architects, and the workers trying to get real things done.

The Connectivity Gap Is Real — and Getting More Expensive

Garner has been in and around enterprise technology for 25 years, from information security and social engineering pen tests in the early 2000s (with a 100% success rate, which says something about physical security that still holds today) to building global teams across Europe and telehealth, and now leading engineering at Smartsheet. He came to the company as a customer first, nine years of using the platform before joining, and brought with him a clear-eyed view of both what works and what needs to change.

His diagnosis of where enterprise AI is stuck aligns with what I see constantly in the market: the problem isn’t a shortage of AI tools. It’s the fragmentation between them. Every major platform — from Microsoft to Salesforce to your vertical SaaS of choice — has shipped its own AI layer. But those layers don’t talk to each other. They operate in silos, each with different terminology, different agent constructs, different interfaces. The result is context switching, cognitive overhead, and AI that can’t follow work across the systems where it actually lives.

“It feels sort of like the early days of SaaS and web services,” Garner said, “where everyone was still just trying to find common ground.” The parallel is apt — and instructive. What resolved that chaos in the web services era wasn’t one vendor winning. It was the emergence of shared protocols and standards that let systems interoperate. MCP is that moment for AI.

Watch the full episode of the Age of AI here:

What MCP Actually Is — and Why It’s Not Just Another Tech Spec

The Model Context Protocol is, at its core, a universal adapter for AI. It standardizes how AI agents discover and invoke tools, pass context between systems, and take action across applications, all within a structured, token-efficient framework. Garner’s analogy: think of how APIs transformed the SaaS ecosystem when they gained widespread adoption. MCP is doing that for AI agents.

But not all MCP implementations are the same, and Garner was direct about this. Many vendors are shipping what he called “lazy MCPs” — base-level wrappers that check a box without delivering real capability. Smartsheet’s approach was different by design. The team built for token efficiency from the ground up, compressing responses, making errors machine-readable and learnable, and engineering for surgical precision with data rather than brute-force context dumping. The result: the ability to do significantly more complex work across a much larger set of artifacts in a single context window, at lower cost.

That last point matters more than it might seem. “Token cost is the new public cloud,” Garner said, invoking a comparison I found immediately recognizable: the same companies that rushed to cloud and then faced shock at the bills are now loading up on AI infrastructure without clear cost discipline. Token efficiency isn’t just a technical preference. It’s a financial strategy.

From Walled Gardens to Horizontal AI

For a decade, the SaaS playbook was built on lock-in. Vendors endeavored to keep the data in the walled garden, make switching painful, monetize the captivity. MCP represents a philosophical break from that model, and this is where Smartsheet is making it a strategic bet.

Garner was unambiguous about the reasoning: Smartsheet is not going to train frontier models. It’s not going to become the next Anthropic or OpenAI. What it can do, and what it’s betting will be durably valuable, is be the best-connected platform for how teams get work done together. “Great work is done in teams,” Garner said, pointing to Smartsheet’s role in running major events, construction projects, rocket programs. The platform’s value is in orchestrating that complexity, and MCP is how it stays in the flow regardless of which AI a team chooses to use.

The strategic framing is clear: the best-connected work management platform is more durable than any closed ecosystem. Customers who adopt MCP-aligned platforms don’t face the integration tax every time they bring in a new AI tool. New Smartsheet capabilities become automatically available to any connected agent, no reintegration required, no waiting for a separate API release.

Governance Is Not an Add-On

The part of this conversation I found most substantively important was the governance discussion. When AI can take action across enterprise systems autonomously, the immediate CIO question is: what does it have access to, who authorized it, and how do I reverse something that went wrong? These are not hypothetical concerns.

Garner’s position was unequivocal: you cannot bolt governance on after the fact. It has to be baked into the architecture itself. Smartsheet’s approach runs every agent action, every query, every data modification through a policy layer that’s part of the infrastructure, not a monitoring layer on top of it. The platform uses a governance data lakehouse architecture with tiered quality gates: raw data is validated and cleaned before any agent touches it, and every agent action is tracked with data time travel for rollback capability.

The responsible AI principles Garner described: data privacy and security, accountability and control, reliability, and transparency and explainability, are implemented as technical specifications, not paper policies. Every AI-generated action surfaces an explicit confirmation before modifying data. What came from AI versus what came from a human is never ambiguous. Logs are shareable, observable, and compatible with customers’ existing observability tooling.

This is what load-bearing responsible AI looks like. Not a blog post. Not a values statement. Architecture.

What the Next 18 Months Actually Look Like

Garner’s near-term forecast was one of the most grounded I’ve heard: within 18 months, most knowledge workers will operate primarily through a personal cluster of agents they’ve trained and trust, spending 80-90% of their time in prompts while their agent workers handle the routine workflows in the background. Context switching will collapse. Manual toil will decline. The interface will shift from navigating to the tool to the AI coming to you.

His engineers who have adopted these tools are already moving 5-7x faster than before. The productivity differential is real and compounding. Teams that figure out agentic workflows in the next 12-18 months will have a structural advantage that will be hard to replicate — not because the tools are secret, but because the organizational fluency, the trained agents, and the governance infrastructure take time to build.

The implication for CIOs is direct: measure AI success by impact and usage, not by deployment and licensing. And invest now in the connective infrastructure — governance, interoperability, and behavioral adoption — that lets AI actually work where work happens.

Read more of my coverage here:

Smartsheet’s MCP-Enabled Claude Integration Signals a New Era for Enterprise AI Connectivity

 

This article was originally published on LinkedIn.