AI agents in production: 79% of companies already use them, but only 20% make full use of them
79% of large companies already have AI agents in production, but only 20% use them broadly and consistently, and, on average, only half of their AI projects deliver the expected value. That is the conclusion of “Autonomy in Practice”, a Genpact study published on November 12, 2026. It surveyed 2,005 executives across 17 industries and 14 business functions, at companies with annual revenues of $1 billion or more (Genpact, 2026).
What the data says
| Indicator | Figure | Source |
|---|---|---|
| Companies with AI agents in production | 79% | Genpact, 2026 |
| Of those, with agents across multiple functions, broadly and consistently | 20% | Genpact, 2026 |
| Of those, with agents in some functions | 32% | Genpact, 2026 |
| Of those, with agents only in isolated use cases | 27% | Genpact, 2026 |
| Companies running pilots or proofs of concept | 17% | Genpact, 2026 |
| Companies still at the planning stage | 4% | Genpact, 2026 |
| Companies with agents in production that increased their number of agents in the past year | 69% | Genpact, 2026 |
| AI initiatives funded in the past 24 months that delivered the expected value (average) | 49% | Genpact, 2026 |
| Companies that have fully tied AI investment to their income statement | 15% | Genpact, 2026 |
| Companies that have completed a significant operating model transformation | 9% | Genpact, 2026 |
Sources: Genpact, 2026 and Genpact press release, 2026. The data are self-reported and percentages are rounded.
In other words, eight in ten companies already have agents, but, on average, only half of their AI projects deliver what they promised.
Do AI agents replace people?
The study looked at who makes decisions across eight stages of business processes. On average, AI decides (with a person auditing it) or acts on its own in 40% of cases, up from 19% in 2025. However, fully autonomous execution is just 9%, and people still decide or audit in 87% to 98% of cases, depending on the stage (Genpact, 2026).
Another finding contradicts the most common fear: among companies with agents in production, 92% kept their headcount the same or grew it, and only 8% reduced it. Leading companies also grew on both fronts over the past 12 months: 78% added agents (compared with 66% of the rest) and 45% also hired people (compared with 27%) (Genpact, 2026).
The problem lies elsewhere: making use of the time agents free up. 77% say they have a plan to redeploy that capacity, but 46% say that failing to do so is their main barrier to measurable value (Genpact, 2026).
What leading companies in AI agents do differently
Genpact classifies as “leaders” the 17% of companies with the highest AI maturity (a year earlier they were 12%). In 44% of them, the CEO leads the change, compared with 29% of the rest (Genpact, 2026). What sets them apart is not having more agents but how they organize work around them:
Interactive chart
| Practice | Leaders (17%) | Other companies |
|---|---|---|
| Redesign their operating model across the enterprise | ||
| Create roles for people and agents working together | ||
| Give agents read and write access to the systems they need | ||
| Govern their agents across the full lifecycle | ||
| Tie AI value to specific lines of the income statement |
The biggest difference is in the last practice: leaders measure the value of AI in their income statement seven times as frequently as the rest. It is the same issue we saw in BCG’s study of technology companies in Latin America, where only 22% validate the impact of their AI with Finance (Wasivia, 2026).
What it means for companies in Ecuador
The study does not include Latin America (Genpact, 2026): almost half of the respondents are in the United States (47%), followed by the United Kingdom and Ireland (16%); the rest are in Canada, Australia, Japan, other European countries and India. They are also companies with revenues of $1 billion or more. Even so, its lessons apply to companies of any size. An agent without access to the systems where the work happens, that nobody supervises and whose result is not measured in money ends up as an expensive pilot. By contrast, starting with a specific process, a person in charge and a clear metric is within reach of any mid-sized company.
How to scale AI agents in your company
- Choose a complete process, not an isolated task. Value appears when the agent covers a workflow from start to finish, such as handling a request or reconciling invoices.
- Give it access to the systems it needs, with the minimum permissions and a log of everything it does (Wasivia, 2026).
- Name a person responsible for the agent. Only 20% of the companies in the study have a designated owner for each agent (Genpact, 2026). That person reviews its results, corrects errors and decides when someone on the team should step in.
- Decide before you start how you will measure value: hours saved, errors avoided, sales closed or shorter response times, always translated into money, and compare it with what it costs to run the agent (Wasivia, 2026).
- Prepare your team. If agents change the work, people need to learn to work with them (Wasivia, 2026).
At Wasivia, we design and implement AI agents and multi-agent systems connected to your company’s systems, with human oversight and a value metric from day one.
References
- Genpact — Autonomy in Practice: How agentic operations can turn AI progress into business value (Nov 12, 2026)
- Genpact — Autonomy in Practice, full report (PDF)
- Genpact via PR Newswire — 79% of Enterprises Report AI Agents in Production, but Only 15% Have Fully Tied AI Investment to P&L Outcomes (Nov 12, 2026)
- Wasivia — AI agents that cost more than they deliver: nearly half of companies have already reined them in
- Wasivia — From pilot to production: how to measure the value of AI so that Finance accepts it
- Wasivia — Gemini agent for business: the controls you should require from any AI agent
- Wasivia — AI training: why company culture matters more than the tool
