Guggenheim Survey Finds 81% of Big Companies Already Running AI Agents, But Token Costs Are Starting to Pinch

The adoption numbers are real

Goldman Sachs estimates AI adoption across corporate America currently sits at roughly 20.6%, with a projected climb to 24% by the end of the year. A Guggenheim Securities survey of 150 large-enterprise IT professionals suggests that number understates what's happening on the ground: 81% of respondents said their companies have already deployed AI agents or chatbots.

Among AI-native platforms, Anthropic and OpenAI are pulling ahead of the pack, according to the survey, reinforcing what's increasingly looking like a two-company race for the enterprise chatbot and frontier-model market.

Workers are using it, and it's working

About 42% of employees at surveyed companies actively use AI tools, spending roughly 22% of their workday with them, the survey found. That translates to an estimated 18% productivity gain for the workers using the tools.

AI now accounts for an average of 19% of corporate IT budgets, according to the survey, concentrated mostly in software development, data analytics, and IT operations. Roughly half of respondents expect AI to become its own separate budget line rather than get folded into existing IT spending, and 37% of those companies plan to fund it with money on top of what they already allocate for IT.

The catch: token costs are starting to bite

Respondents on average still expect AI to have a positive impact on operating margins, and they expect that benefit to grow to 3.1% by 2027.

But the survey also found more respondents now anticipate mid-to-high single-digit margin decreases in 2027 compared with 2026, a signal that token costs are starting to outweigh the benefits for a meaningful slice of companies, even though the overall expected benefit still outweighs the cost concern for now.

That tracks with what's already showing up elsewhere. Uber capped its AI coding spend after burning through its entire 2026 agentic AI budget in just four months. One company received a mysterious $500 million Claude bill. UBS checks found token costs have become a live concern for roughly 60% of enterprise customers. And at one company, leadership's reaction to its first AI invoice was blunt: "We don't have the money for this."

The layoffs argument, and its limits

The headline finding from Guggenheim's survey directly undercuts the loudest fear circulating around AI adoption: that agents are about to gut white-collar payrolls. Respondents reported an average headcount reduction of just 2.5% tied to AI, far short of the mass-layoff forecasts that dominated headlines earlier this year. Marc Andreessen has dismissed the narrative of sweeping AI-driven job cuts as "fake," and Guggenheim's findings lend some support to that view: AI is improving productivity and easing labor constraints, but it has yet to trigger mass layoffs in this data set.

With 81% adoption and only 42% of employees actively using the tools for a fifth of their day, the data doesn't show mass displacement. It shows augmentation, at least so far. Whether productivity gains eventually translate into fewer new hires down the road, even without outright cuts to existing headcount, is a separate question this survey doesn't settle — it measures current deployment and current productivity, not hiring plans two or three years out.

What happens next

The test is 2027. Guggenheim's own respondents are split: most still expect AI to boost margins, but a rising share now expect it to hurt them, driven by token costs that have already forced Uber to cap its coding budget and left at least one unnamed company staring down a bill it couldn't pay. Whether enterprise AI spending settles into a predictable budget line, as roughly half of respondents expect, or keeps generating surprise invoices, will decide whether the productivity story or the cost story wins out heading into next year's budget cycles.