Skip to content
Enterprise AI Agents: What the Top 40% Are Actually Building
EconomicsSeptember 5, 20269 min read

Enterprise AI Agents: What the Top 40% Are Actually Building

McKinsey 2026: 40% of large enterprises scale AI agents. Here's what they're actually deploying -- and why productivity gains aren't hitting earnings yet.

McKinsey's August 2026 survey of 1,719 organizations across 97 countries found that 40% of large enterprises (over $1B revenue) are now scaling AI agents in production -- up from 27% a year ago. Smaller firms stayed flat at 22%. The gap between those two numbers is the story.

I've been watching the enterprise AI wave from the outside -- running agents in production on a small scale -- and the McKinsey data finally makes the split legible. There's a two-speed race happening, and the gap between the leaders and the laggards is widening fast.

What does "scaling AI agents" actually mean in practice?

According to McKinsey's 2026 State of AI survey, scaling AI agents means deploying them across at least one business function at meaningful volume -- not running a pilot, not an internal demo, not a proof-of-concept. The 40% figure represents organizations where agents are doing real production work that shows up in operational metrics.

The survey covered 1,719 respondents from May 4 to June 8, 2026. The breakdown by company size is stark: 40% of organizations with over $1 billion in annual revenue are scaling agents, while only 22% of smaller firms are -- and that smaller-firm number hasn't moved year-over-year.

Scale advantages compound here. Larger enterprises have the budget to absorb failed experiments, the IT infrastructure to support agentic workloads, and legal and compliance teams already thinking about AI governance. The 60% still on the sidelines aren't slow -- they're facing genuine structural barriers that money alone doesn't solve.

Free Newsletter

Get the daily AI agent signal in your inbox.

One email, every morning. The builds, tools, and frontier research that matter — no fluff, no AI hype cycle noise.

Subscribe free

Where are they actually deploying?

McKinsey found that the top deployment areas in 2026 are IT operations, knowledge management, and software engineering. About 31% of large enterprises are now scaling coding agents specifically -- tools like Claude Code, GitHub Copilot, and custom agentic pipelines that write, review, and debug production code.

The knowledge management case is probably the most underappreciated. Corporate knowledge is scattered across SharePoint, Confluence, Notion, Slack archives, and years of email threads. Knowledge agents connect to all of those simultaneously, build semantic indexes, and return cited answers in seconds instead of sending employees on a document hunt that takes hours.

IBM's AskHR deployment is a useful benchmark. Running on watsonx Orchestrate, it logged over 11.5 million employee interactions and achieved a 94% containment rate on common HR questions -- meaning agents resolved the query without human escalation. That translated to a 75% reduction in support tickets and a 40% cut in HR operational costs over four years. That's not a pilot result. That's what production scale looks like after years of iteration.

At the extreme end, JPMorgan's LLM Suite reaches 200,000 employees daily and supports over 450 AI use cases across the organization, backed by an $18 billion annual technology budget. The lesson from JPMorgan isn't "spend $18B" -- it's that scale in enterprise AI comes from many small, well-governed applications, not from betting everything on one giant agent that does everything.

The build-vs-buy flip: 32% are skipping SaaS purchases

One of the most economically significant findings from McKinsey's 2026 survey: 32% of respondents said their organizations decided against buying at least one software product or feature because they could build it with agentic coding tools instead. That number jumps to nearly 50% among the 6% of organizations McKinsey classifies as "high performers."

Gartner put a dollar figure on this shift in July 2026: $234 billion of enterprise application software spending is exposed to what they call "agentic arbitrage" by 2030. That's not all at risk immediately, but it signals a structural threat to the SaaS model that vendors are scrambling to respond to.

To be clear about what this means and doesn't mean: a company using Claude Code to build an internal reporting tool is very different from replacing Salesforce or SAP. The 32% are filling gaps and killing point solutions, not ripping out their ERP. But the long-term trajectory matters. If 32% are doing this today, what happens to the niche SaaS market in 2027 when the tooling gets better and cheaper?

Retool's 2026 Build vs. Buy Report adds another data point: 35% of enterprises say they've already replaced at least one SaaS product with custom software built internally. The McKinsey and Retool numbers are converging around the same story.

Want this built for your business?

Venti Scale builds AI automation systems for businesses that want results without the learning curve. One operator, AI-powered, full marketing stack.

See What We Build

The ROI gap: 80% individual productivity, 37% EBIT impact

Here's the uncomfortable number buried in McKinsey's data: 80% of individual workers report that AI has improved their daily productivity. But only 37% of organizations report any EBIT impact from AI -- and that share hasn't increased from 2025. Only 6% of respondents qualify as "high performers" (attributing at least 5% of EBIT to AI), and that percentage hasn't grown either.

This is the productivity paradox of the current AI moment. Individuals feel faster. Companies aren't getting materially richer. The gap exists because most organizations are using AI to do the same things slightly faster -- not to redesign what they're doing.

McKinsey's data is clear on what separates high performers from everyone else. High performers pursue growth or innovation alongside efficiency (not just cost-cutting). They fundamentally redesign workflows that are enabled by AI rather than bolting AI onto existing processes. And they're twice as likely to have senior leaders who are visibly committed to AI adoption -- not just funding it from the budget, but personally using it and demanding it from their teams.

An agent that summarizes your existing sales reports faster is a productivity tool. An agent that monitors signals from 50 data sources and surfaces a deal risk before the sales rep even flags it is a workflow redesign. The first saves minutes. The second changes the outcome.

Gartner's agentic AI spend number: $201.9 billion in 2026

Gartner forecasts total agentic AI spending at $201.9 billion in 2026 -- a 141% increase from 2025. This number counts agentic capabilities embedded across enterprise software broadly, so it's bigger than the pure "AI agent platform" market. But even as a directional signal, a 141% year-over-year growth rate tells you where enterprise software budgets are moving.

By 2027, Gartner expects agentic AI to overtake chatbot and assistant spending for the first time. The shift from "AI that responds to questions" to "AI that takes actions" is the defining enterprise technology transition of this decade. The 40% in the McKinsey data are early movers in that transition. The 60% are running out of runway to catch up without paying a premium for vendors who moved first.

There's a sobering counterpoint though. Gartner's same research predicts that 40% of agentic AI projects will be canceled by 2027. Not deferred -- canceled. Production complexity, governance overhead, and the gap between demo performance and real-world reliability are killing projects that looked promising in pilots. The 40% running production agents today have already survived that filter.

What the 60% are actually stuck on

If the economics are this compelling, why are 60% of large enterprises still not there? The honest answer is a cluster of genuine hard problems, not lack of ambition.

Data access and quality is the most common blocker I see mentioned. Agents are only as useful as the data they can reach. If your CRM is a mess, your SharePoint has 15 years of orphaned files, and your databases require 4 approvals to query -- the agent has nothing to work with. The infrastructure debt that organizations accumulated over decades doesn't disappear because LLMs got good.

Governance and liability is the second blocker. When a human makes a decision, you can ask them why. When an agent makes 10,000 decisions per day, accountability gets complicated fast. Regulated industries -- finance, healthcare, legal -- are moving slower because the cost of an agent making the wrong call is higher.

The third is internal capability. Building and operating agents requires a skillset that's genuinely scarce: people who understand both what the agents can do technically and what the business actually needs them to do. That intersection of AI-literate operators is where most organizations are thin.

FAQ

What percentage of enterprises are running AI agents in production in 2026?

According to McKinsey's 2026 State of AI survey, 40% of large enterprises with over $1 billion in annual revenue are scaling AI agents in at least one business function. Among smaller organizations, the number is 22% and hasn't grown year-over-year. Gartner's CIO survey puts actual production deployment across all organizations at 17%.

What functions are enterprises deploying AI agents in first?

McKinsey's 2026 data shows IT operations, knowledge management, and software engineering as the top three deployment areas. Coding agents are being scaled by 31% of large enterprises specifically. Customer-facing deployments in sales, marketing, and support are growing but face higher governance scrutiny in regulated industries.

Is the build-vs-buy shift real for enterprise software?

Yes, with caveats. McKinsey found 32% of enterprises skipped buying at least one software product because they could build it with agentic coding tools. Gartner quantified the exposure: $234 billion of enterprise application software spending faces agentic arbitrage risk by 2030. But this is about gap-filling and point solutions, not replacing core platforms like Salesforce or SAP yet.

Why aren't productivity gains showing up in company earnings?

McKinsey's data shows 80% of individual workers report AI productivity improvements, but only 37% of organizations see any EBIT impact. The reason: most organizations use AI to speed up existing processes rather than redesigning workflows. High performers -- the 6% attributing 5%+ of EBIT to AI -- pursue growth and redesign work, not just cost reduction.

What does Gartner forecast for agentic AI spending in 2026?

Gartner forecasts $201.9 billion in agentic AI spending for 2026, a 141% increase from 2025. By 2027, agentic AI will surpass chatbot and assistant spending for the first time. However, Gartner also predicts 40% of agentic AI projects will be canceled by 2027 due to production complexity, governance challenges, and the gap between pilot and real-world performance.

Want this built for your business?

Venti Scale builds AI automation systems for businesses that want results without the learning curve. One operator, AI-powered, full marketing stack.

See What We Build
AI Agents First

The daily signal from the frontier of AI agents.

Join builders, founders, and researchers getting the sharpest one-email read on what's actually shipping in AI — every morning.

No spam — unsubscribe anytime