The new GTM stack: agent-first, context-first
Every revenue team is being sold the same promise: the company with the smartest agents wins. Budgets are moving to match, and the platforms are racing to oblige. Salesforce made the boldest version of the bet in April, when it launched Headless 360 and exposed its entire platform, with Agentforce and Slack alongside it, as APIs, MCP tools, and CLI commands so agents can work without anyone opening the interface. Marc Benioff summed it up as "Our API is the UI." The largest CRM in the world is rebuilding itself agent-first, and most of the industry is moving the same way. But the only way to take full advantage of that world will be with great context.
The benefits of an agent-first stack seem self-evident. We have all seen the vibe-coded POCs and the fast build-outs of functional tools. But the business returns have not followed the demos. MIT's NANDA initiative found that only about 5% of enterprise GenAI pilots are producing measurable value, with the rest stuck at no measurable impact on the P&L. Gartner expects AI agents to outnumber human sellers ten to one by 2028, and in the same prediction expects fewer than 40% of those sellers to say the agents made them more productive. The demos are impressive and some of the tools are good, so the easy conclusion is that the agents just need to get better. They will, but it still won't be enough.
Integration is becoming a commodity
Every agent and agentic workflow has two layers, and most teams focus only on the surface. There is the agent itself and its model, the reasoning and the actions everyone demos. Underneath is the context layer: what the agent knows, the facts, the rules, the current state of your business. The teams building these systems have already learned which layer is the hard one. Anthropic, writing about how to build effective agents, calls context "a finite resource with diminishing marginal returns" and warns of "context rot," where a model's recall fades as the window fills. Managing an agent is mostly about managing its context.
While budgets and attention go to the surface, the context layer underneath goes ungoverned, and that layer decides whether the agent answers and acts correctly. The even harder requirement is consistency: the same context holding across every system and every agent. Whose agent is smartest matters less than what every agent is standing on, and the real limit on scaling agents across an organization will be context that is governed and current.
A better model can’t fix a knowledge problem
The foundation for connecting those agents is standardizing fast. The Model Context Protocol went from a single Anthropic proposal in November 2024 to adoption by OpenAI, Google, and Microsoft within six months, and by December 2025 it had been handed to the Agentic AI Foundation, a neutral body under the Linux Foundation, with tens of thousands of servers in use. Salesforce's Headless 360 alone ships more than sixty MCP tools out of the box. We know how the pieces fit. Most organizations still cannot get value out of them at scale.
So the connections are exploding, even commoditizing, while the knowledge and context layer stays out of reach for most organizations. A smarter model will not change that, because a model is only as good as the context it is given, and a strong model on ungoverned knowledge produces fluent answers that are confidently wrong. Connecting an agent to more ungoverned systems only adds more ways for it to be wrong. Gartner analysts have predicted that by 2028 most agentic analytics projects built on MCP alone will stall for want of a consistent layer of meaning beneath them, and a Stanford study of commercial legal AI tools found they still hallucinated between 17% and 33% of the time even with retrieval working. In each case the knowledge underneath failed before the model or the connections did.
The different flavors of context
Plenty of vendors already give agents context. External-data providers like ZoomInfo, which now markets "the headless GTM context layer to ground every AI agent in verified GTM data," tell an agent who the buyer is, the firmographics, and the intent. Revenue-intelligence tools like Spekit partner Gong know what happened on the calls and in the deals. Horizontal search tools like Glean index the whole company's documents, but cannot tell what is current or approved, because that corpus is ungoverned. Each is useful. None of them tells the agent what is true about your own business: your pricing this quarter, your approved positioning, the play you run in this deal.
That last kind of knowledge is the one no one governs, and it is the one that goes out of date and out of sync fastest. It is spread across your CRM, your Slack, your decks, and your best reps' heads, and it changes the moment you change a price or a pitch. A search tool finds where it is written down, not which version is current. An agent that knows the buyer cold but is guessing at your pricing will quote a discount you eliminated last quarter. Confidently. None of that is visible in a demo. It becomes obvious in production, where most of these pilots die.
GTM Knowledge Engine 2.0
This is the problem Spekit's GTM Knowledge Engine 2.0 is built to solve: a structured, governed repository of your organization's go-to-market context, with the AI tooling to keep it current and accessible. Why is an enablement platform the one solving this? Because it is the same problem we have always had to solve for humans. Enablement's old failure came down to engagement: you could not get reps back to the tool or the training when they actually needed it. Spekit's AI Sidekick worked by putting governed knowledge inside the rep's flow of work, so they no longer had to go find it, and adoption followed because the knowledge showed up where they already were.
GTM Knowledge Engine 2.0 extends that same governed knowledge from every rep to every agent, through one source, exposed to any tool through MCP. A governance layer is not one more place to search, it’s where action happens. Stale knowledge gets flagged before an agent ever sees it, and whatever an agent produces runs back through brand and approval before it leaves the building.
Agents are not the problem, and an agent-first GTM is clearly where the market is going. Most organizations are just focused a layer too high. Teams that win the next few years will stop competing on whose agent is smartest and get disciplined about the knowledge underneath, the critical work of knowing what is true about their own business and keeping it current.
To learn how the trusted GTM Knowledge Engine can transform both your revenue enablement and agentic workflows, contact us for a live demo.
