Gartner projects the AI agent software market at $376.3B by 2027 and only about 130 real vendors behind it. The gap between the market number and the reality is what interests me.
The Gartner April 2026 forecast has a number I genuinely had to re-read: the AI agent software market is projected to go from $86.4 billion in 2025 to $206.5 billion in 2026 to $376.3 billion in 2027. That is more than a four-times increase in two years. Gartner also expects agentic AI to appear in 40 percent of enterprise applications by the end of 2026, up from less than 5 percent now. Those two figures together describe something closer to a phase transition than a product trend. And sitting right next to them in the same report is a detail that stopped me: Gartner estimates only about 130 real AI agent vendors exist. The rest, they say, is "agent washing."
That pairing is what I want to think through. A $376 billion market with 130 real vendors and an unknown quantity of agent-washed products is not a normal market dynamic. It tells me that the definition of "AI agent" is doing enormous work right now, and that the work it is doing is mostly in vendor marketing materials rather than in the technical reality of what these systems can do.
The word "agent" spans a very wide range. At one end, an agent is a system that receives a high-level goal, plans a sequence of steps to achieve it, calls tools and APIs to execute those steps, observes the results, and adjusts the plan when something does not go as expected. That version of an agent requires reliable reasoning, effective tool use, and the ability to handle unexpected states without breaking. It barely exists outside of carefully scoped research settings and heavily supervised enterprise pilots. At the other end of the spectrum, "agent" is being applied to any workflow automation that can take an action based on a trigger without a human explicitly clicking a button. A rule-based email router that sends support tickets to the right queue based on keyword matching is being sold in some vendor catalogs as an AI agent. The technical distance between those two things is enormous. The marketing distance is zero.
DiMaggio and Powell (1983) described institutional isomorphism as the process by which organizations become structurally similar not because the structure works but because it is what other organizations in their field are doing. They identified three mechanisms: coercive pressure from regulators or powerful partners, mimetic pressure from copying successful peers, and normative pressure from professional standards and consultants. The AI agent market right now is a textbook case of mimetic isomorphism. Organizations are buying agent platforms and labeling internal tools as "agentic" because their competitors are doing it, because their analysts are reporting on it, and because their boards are asking about it. The question of whether the specific agent purchase will produce value is secondary to the question of whether the organization is seen to be adopting the category. Gartner's own language about "agent washing" is an acknowledgment that the category pressure is outrunning the genuine capability.
What are the real agents actually doing in enterprises right now? The tasks that work are narrow, well-defined, and high-volume. A code review agent that scans pull requests for known vulnerability patterns and flags them for human review is doing something real. A ticket routing agent that classifies incoming support requests by reading the description and assigning them to the right queue is doing something real. A document summarization agent that ingests policy documents and produces structured outputs for human review is doing something real. The pattern across all of these is that the task has clear success criteria, the inputs are semi-structured, failures are recoverable, and a human reviews the output before it creates downstream consequences. The agent is fast and consistent at a narrow task. That is valuable. But it is not the $376 billion vision of autonomous systems that manage complex multi-step business processes without human intervention.
The tasks that do not yet work for agents reliably are the ones requiring multi-step judgment under ambiguity, coordination across organizational boundaries where accountability matters, and decisions where the cost of an error is high. An agent that autonomously approves expense reports below a certain amount and routes exceptions to a manager is plausible today. An agent that autonomously manages vendor contract negotiations, interprets ambiguous clause language, and signs agreements on behalf of the organization is not plausible today, and an organization that deploys one and calls it an agent is taking on risk that the market size number does not capture.
From an IS research perspective what surprises me about the $376 billion projection is not the size of the number. Large numbers in technology forecasts are not unusual. What surprises me is the combination of that number with the "only 130 real vendors" finding. If there are only 130 genuine agent vendors and the market is $376 billion, the average revenue per real vendor is enormous. That arithmetic suggests most of the market value is concentrated in a very small number of providers, the hyperscalers and the one or two platform vendors who have real agentic infrastructure. The rest of the $376 billion is going to products that are calling themselves agents for commercial reasons, not technical ones. From an IS value perspective, the interesting question is not how much money is flowing into the agent category. It is how much of that money is buying genuine capability versus category membership. I suspect those are very different numbers, and I do not think we have the measurement frameworks to tell them apart yet.
The absorptive capacity argument applies here too, in the same way it applies to any technology adoption wave. Cohen and Levinthal (1990) argued that the ability to absorb new knowledge depends on the prior related knowledge an organization has already accumulated. Organizations that have already built reliable automation, that have already managed data quality across systems, that have already designed workflows with structured handoffs, are in a position to absorb agentic AI and get something from it. Organizations that have not done that prior work cannot simply buy an agent platform and expect autonomous value. The agent will fail in the places their existing automation already fails, just faster and at higher volume. The $376 billion market will produce very different outcomes for those two types of organizations, and the aggregate forecast number tells us nothing about that distribution.
---
claims_checked:
- "AI agent software market: $86.4B (2025), $206.5B (2026), $376.3B (2027)": "https://www.gartner.com/en/newsroom/press-releases/2026-04-07-gartner-forecasts-worldwide-it-spending-to-grow-9-8-percent-in-2026"
- "Agentic AI in 40% of enterprise apps by end 2026, up from less than 5% now": "https://www.gartner.com/en/newsroom/press-releases/2026-04-07-gartner-forecasts-worldwide-it-spending-to-grow-9-8-percent-in-2026"
- "Only ~130 real AI agent vendors, rest is agent washing": "https://www.gartner.com/en/newsroom/press-releases/2026-04-07-gartner-forecasts-worldwide-it-spending-to-grow-9-8-percent-in-2026"
claims_unverified:
- "DiMaggio and Powell (1983) institutional isomorphism: drawn from IS theory background, not re-fetched this session"
- "Cohen and Levinthal (1990) absorptive capacity: drawn from IS theory background, not re-fetched this session"
- "Specific agent use case examples (code review, ticket routing, document summarization): described from industry observation, not a single primary source"
sources_used:
- "https://www.gartner.com/en/newsroom/press-releases/2026-04-07-gartner-forecasts-worldwide-it-spending-to-grow-9-8-percent-in-2026"
word_count: 1090
About the author
Share
More notes
Related notes