TL;DR
- Among the software buyers G2 surveyed in July 2026, 66% said their organizations had agents in production. This includes limited deployments.
- Larger companies are further along. In McKinsey’s 2026 survey, 40% of respondents at businesses with annual revenue above $1 billion reported scaling agents in at least one function, compared with 22% at smaller firms.
- Savings from one deployment can make only a small difference to total profit. Across AI use as a whole, McKinsey’s 2026 survey found that 37% of respondents reported some contribution to company-wide earnings before interest and taxes.
- Only 21% of respondents in Deloitte’s 2026 research said their organizations had mature governance for AI agents.
- In its June 2025 forecast, Gartner predicted that more than 40% of agentic AI projects would be canceled by the end of 2027 because of costs, unclear value and inadequate risk controls.
- Test agents on your team’s own work. Count the time spent reviewing and correcting results, along with running and support costs, before funding a wider rollout.
Headlines about widespread AI adoption can make it easy to miss the difference between a short trial and a system people rely on every day. When companies say they use AI agents, they can mean very different things.
In McKinsey’s 2026 survey, 40% of respondents from companies with annual revenue above $1 billion said they were scaling AI agents in at least one business function. Among respondents from smaller companies, the figure was 22%. Deloitte’s 2026 study surveyed leaders at US companies already testing or using agents. Only 15% reported scaling systems where multiple agents work together across departments.
When you compare your own systems with these figures, focus on businesses of a similar size and agents with similar levels of autonomy.
How many companies use AI agents?
In July 2026, G2 surveyed 1,019 people involved in B2B software purchasing. Of those, 66% reported that their organizations had agents in production. This includes deployments within individual teams, where agents work with real business data and systems.
A team can compare the time spent on a task before and after deployment. For a private equity firm, Tensorway built custom agents to find and rank potential investments, extract financial data from documents, and prepare investment committee presentations. The published case study reports an 80% reduction in initial screening time and eightfold faster investment deck preparation.
How widely are companies deploying AI agents?
In McKinsey’s 2026 global survey, about one in five respondents said their organizations were scaling AI agents across the business. Rolling agents out to more teams means agreeing on data access and who will maintain each deployment after launch.
Deloitte’s August 2026 study surveyed 501 US business leaders at companies already testing or using agents. The largest group was expanding deployment across business functions. Coordinated multi-agent systems across multiple workflows were less common in this sample.
Which companies get furthest with AI deployment?
The largest ones by revenue. Stanford publishes the distribution of AI deployment stages by company size, and the relationship is monotonic. Among companies with more than $5 billion in revenue, 39% are at the scaling stage and 10% report being fully scaled. Among companies below $100 million, those figures are 25% and 5%.
The pattern reverses at the experimentation end. In the smallest revenue band, 39% are still experimenting; in the largest, 17%.
Stanford attributes this directly to resourcing: given the resource and investment demands of integration, larger companies were the most likely to report that their AI programs had reached a scaling phase.
Growth is uneven across regions. Reported AI use reached 88% across all geographies in 2025, with the largest year-over-year increases in China and Europe at 13 and 11 percentage points respectively. More than half of respondents reported AI in three or more business functions..
Where are companies scaling AI agents?
In the technology sector, in engineering and IT functions. Almost nowhere else.
Technology companies still lead. The Stanford AI Index 2026, drawing on McKinsey’s 2025 survey, reports that 24% of respondents in the technology sector were scaling agents in software engineering. The figures for IT and service operations were 22% and 21%.
IT, knowledge management and software engineering are also the leading functions for agent scaling in McKinsey’s 2026 survey. The most common uses vary by industry. Consumer goods and retail companies most often use agents in marketing and sales. Automotive, aerospace and semiconductor companies use them in supply chains, inventory management and manufacturing.
At logistics provider C.H. Robinson, one agent reads carriers’ emails about available trucks and enters the details into an internal system used to match trucks with loads. Other agents produce freight quotes, process orders and book pickup and delivery appointments. In April 2025, the company reported that its agents had completed more than 3 million shipping tasks.
For a pilot like this, track how long it takes to turn an email into an order, how accurately the details are transferred and how many orders the team can process each day. Those results help you decide whether to roll the system out more widely.
How fast is agent usage growing within platforms?
Between March 2025 and March 2026, the number of active agents in Microsoft 365 grew 15-fold. Among large enterprises, growth reached 18-fold. Microsoft published these figures in its Work Trend Index on May 5, 2026.
Microsoft counts an agent as active if, within a 28-day period:
- a user interacts with it on at least one day
- or it completes at least one autonomous run.
Someone could try an agent once and never open it again, and it would still count as active for that period. The 15× figure alone cannot show how frequently each agent is used.
Salesforce tracks individual agent actions through its own metric, Agentic Work Units. A processed prompt or a tool call can count as one unit. In its August 26, 2026 report, Salesforce reported 3.2 billion AWUs across Agentforce and Slack for the quarter ended July 31. That was a
What ROI do companies report from AI and agents?
If an AI support agent cuts your outsourcing bill, you can calculate the savings and how long it takes to recover the cost of deployment. Those savings may still represent a small share of a large company’s profit. Keep that distinction in mind when reading research on AI returns.
In Google Cloud’s September 2025 study, 74% of executives reported seeing ROI from generative AI within the first year. The survey covered companies that already had the technology in use. Respondents also described changes in revenue and productivity:
- Among those who reported higher revenue, 53% put the increase at 6-10%.
- Productivity at least doubled for 39% of respondents who reported productivity gains.
IDC put a dollar figure on returns: an average of $3.70 for every $1 invested, with an average of $10.30 among leaders. Microsoft published these findings in January 2025, drawing on research conducted in 2024.
For a view of the whole business, McKinsey asks respondents to estimate AI’s contribution to earnings before interest and taxes, or EBIT. In November 2025, 39% reported some contribution, usually below 5% of EBIT. In the August 2026 survey, 37% reported a positive contribution. McKinsey describes the result as essentially unchanged.
About 6% of respondents in both surveys qualified as AI high performers. They attributed at least 5% of EBIT to AI and described the value they received as significant. For a large business, the savings from one successful deployment can be too small to noticeably change overall EBIT.
Company-wide productivity gains can also be modest. In the October 2026 revision of NBER’s Firm Data on AI, nearly 9,000 executives from seven countries estimated that AI had increased productivity by around 0.2% a year over the previous three years. A process that speeds up considerably may still account for a small share of the company’s work.
Two frequently cited figures come from MIT Project NANDA’s preliminary July 2025 report: 95% of organizations seeing no measurable financial impact, and roughly 5% of integrated pilots generating substantial value. These estimates need careful handling. The authors drew on interviews, survey responses and public examples, with varying definitions of success. Those percentages do not establish the failure rate for future AI projects.
Google Cloud’s July 2026 study also asked about agents specifically. Of the respondents, 94% said agents contributed to cost savings and revenue growth. To work out whether a deployment has paid for itself, those gains need to be compared with development, integration and running costs.
For the support agent in our example, that means comparing the savings on outsourcing with the full cost of the system over the same period. If employees spend less time handling queries, look at what they did with the time freed up and what that work was worth to the business. Your own results will give you a firmer basis for the calculation.
How many AI agent projects could be canceled?
In its June 25, 2025 forecast, Gartner predicted that more than 40% of agentic AI projects would be canceled by the end of 2027. It named three reasons: rising costs, unclear business value and inadequate risk controls.
At the time, Gartner described most projects as early experiments. Its January 2025 poll of 3,412 webinar attendees asked about their organizations’ investment in agentic AI:
- 19% reported significant investment.
- 42% said their organizations had invested cautiously.
- 8% reported no investment.
- 31% were taking a wait-and-see approach or were unsure.
Choosing a suitable task was another problem. Gartner pointed out that some use cases marketed as agentic did not need agents. For a process with a fixed sequence of steps, it makes sense to compare an agent with conventional automation before starting development.
Gartner returned to the forecast in its October 2, 2026 article, with a focus on reliability. Its advice is to give agents greater autonomy as they prove they can do the job consistently. Drafting a customer reply and issuing a refund, for example, require different permissions and checks.
In its July 2026 health care outlook, Deloitte shared findings from a survey of 64 CFOs at health systems and health plans. Their responses covered AI initiatives broadly: around a third of those launched over the previous two years had been paused or abandoned.
How do companies oversee AI agents?
If an agent sends a customer a quote with the wrong price, its activity log should show which catalog it opened and where the price came from. The person responsible for the agent can then correct the source data or adjust its settings.
In Deloitte’s January 2026 report, only 21% of respondents said their organizations had a mature governance model for agentic AI. In the same survey, 74% expected at least moderate use of agents by 2027. Deloitte surveyed 3,235 leaders across 24 countries in August and September 2025.
One of IT’s first jobs is to find out which agents different teams are building. Microsoft uses Agent 365 for this. In August 2026, the company reported that the system gave its IT team visibility into more than 500,000 agents. Registry entries include ownership, status and usage information. When an agent causes a problem, the team can find the person responsible for it.
The pricing catalog needs someone to keep it current, too. In its February 26, 2025 publication, Gartner reported that 63% of organizations either lacked the right data management practices for AI or were unsure whether they had them. Gartner also predicted that 60% of AI projects unsupported by AI-ready data would be abandoned by the end of 2026.
For a sales quote, keep a link to the catalog the agent used and a record of the changes it made. That record helps the team trace recurring errors to their source and fix them, rather than manually correcting each document.
How much does AI actually improve productivity?
Stanford aggregates results from several studies, and the pattern holds: gains are largest where work is structured and measurable and where output is easy to monitor.
The strongest measured results:
- Customer support agents using a conversational AI assistant resolved 14-15% more issues per hour.
- Developers using GitHub Copilot completed 26% more pull requests.
- Marketing teams using multimodal AI for ad creation saw a 50% increase in output per worker.
- Accountants using AI-based tools gained 55% in throughput.
The downside is measured too. The most widely cited example is the METR study, in which experienced open-source developers became 19% slower with AI assistance, with a gap between how helpful they believed the tools were and how they actually performed. Stanford adds a qualifier that rarely travels with the figure: METR has not been able to replicate the result in later work, largely because developers grew reluctant to work without AI, and because developers in late 2025 were likely sped up by AI relative to the original study period.
Learning is a separate case. Engineers who relied heavily on AI while picking up new libraries showed no measurable speed improvement and ran into what researchers call learning penalties.
What do analysts forecast for AI agents?
Your company’s next AI agent might arrive in an update to its CRM or customer support software. In its August 26, 2025 forecast, Gartner expected task-specific agents to be integrated into 40% of enterprise applications by the end of 2026. At the time, it put the share at less than 5%.
An earlier Gartner publication from June 25, 2025 set out two other forecasts for 2028:
- 33% of enterprise applications would include agentic AI, up from less than 1% in 2024.
- At least 15% of day-to-day work decisions would be made autonomously through agentic AI, compared with Gartner’s estimate of 0% in 2024.
The updated August forecast also anticipates changes in how people use software. By 2028, agent ecosystems could work across multiple applications and business functions. In this scenario, users would say what they need done, and agents would carry out the steps across several systems.
In an October 22, 2026 article, IDC forecast 2.5 billion agents by 2030. Under Gartner’s best-case scenario, agentic AI could account for roughly 30% of enterprise application software revenue by 2035, compared with 2% in 2025. That share would be worth more than $450 billion.
Companies are also expected to reconsider some of their uses of agents. Gartner forecast that more than 40% of agentic AI projects would be canceled by the end of 2027. In its October 28, 2026 forecast, Forrester expects enterprises to return 20% of agentic workflows to automation based on fixed rules in 2027 to cut costs.
At your next software review, find out which agent features are included, which cost extra and how they’ll connect to your systems. You can then decide what the vendor’s product can handle and what needs additional development.
Where is AI investment actually going?
Here the data is observed rather than projected. Global corporate AI investment reached $581.69 billion in 2025, a 129.9% increase over the previous year. Private investment made up the largest share at $344.66 billion, up 127.5%.
Generative AI companies accounted for $170.9 billion of that, nearly half of all private investment and an increase of over 200% from 2024.
The breakdown by focus area shows where the money concentrates. The largest share went to AI infrastructure, models, research and governance at $143.2 billion. AI agents as a focus area drew $8.02 billion, per the report's focus-area breakdown.
Investment is flowing mainly into the layer agents run on rather than into agent products themselves. That tracks with what the adoption data shows: infrastructure is being built faster than agents are reaching scale.
Conclusion
The 2026 data describes a market where the infrastructure is already in place and agents have reached scale in a narrow set of technology-sector functions. Figures diverge because of differing definitions and samples rather than conflicting sources, so any adoption or ROI number is only meaningful alongside what was measured and who was asked.
Forecasts through 2035 can help you prepare for changes to the software your company uses. Before committing a budget to the next phase, look at what your own pilot has shown about the time saved, the cost of support and whether your team is ready to work with the agent every day.
If you are working out where agents fit into your own processes and want a clear-eyed assessment rather than market averages, talk to the Tensorway team.
