Agentic AI in the enterprise: what the early adoption data shows
AI agents that plan and complete multi-step tasks are the newest focus for enterprises and investors. Here is what the early adoption data shows and what founders building agents should consider.
Key points
- McKinsey found 62% of organisations are experimenting with AI agents and 23% are scaling them in at least one function.
- Carta reported more than 60% of venture capital raised on its platform in Q1 2026 went to AI companies.
- Reliability, integration, accountability, measurement and cost slow enterprise deployment of agents.
- Founders should start narrow, design for human oversight and measure outcomes.
The first wave of generative AI products answered questions and drafted content. The next wave aims to do work: plan a sequence of steps, use software tools and complete tasks with limited human supervision. These systems are known as AI agents, and they have become a major focus for enterprise buyers, founders and investors.
What an AI agent is
There is no single definition, but an AI agent generally combines a language model with the ability to plan, call tools such as databases or applications, remember context and take actions toward a goal. Examples include agents that triage customer support tickets and resolve simple cases, reconcile invoices, research and draft reports, or manage software development tasks.
What the adoption data shows
McKinsey’s State of AI 2025 survey, published in November 2025, found that 62% of respondents said their organisations were at least experimenting with AI agents. Twenty-three per cent said they were scaling agentic AI in at least one business function. McKinsey senior fellow Michael Chui noted that scaling was usually limited to one or two functions.
The broader survey found that 88% of organisations use AI in at least one function, but only about a third have begun scaling AI across the enterprise. Agents are at an even earlier stage.
Investment is flowing in
Venture investment in AI has been intense. Carta’s State of Private Markets report for the first quarter of 2026 found that more than 60% of all venture capital raised by companies on its platform went to AI companies. A growing share of new AI startups describe their products as agents or agentic platforms.
Why agents are hard to scale
Several challenges slow enterprise deployment:
- Reliability. Agents can make mistakes, and errors in multi-step tasks can compound. Enterprises need confidence before allowing agents to act without review.
- Integration. Agents need access to many systems, which raises security, permission and data quality issues.
- Accountability. When an agent takes an action, someone must be responsible. Governance frameworks are still developing.
- Measurement. Proving that an agent saves money or improves outcomes requires careful baselines.
- Cost. Running agents that use models extensively can be expensive at scale.
What founders should consider
Founders building agents can improve their chances by:
- Starting narrow. Focus on a well-defined task with clear inputs, outputs and success measures.
- Designing for human oversight. Build review and approval steps that can be relaxed as trust grows.
- Investing in integration and security. Enterprise buyers will scrutinise both.
- Measuring outcomes. Help customers quantify time saved, errors reduced or revenue gained.
- Managing margins. Understand how model costs scale with usage and design pricing accordingly.
Regional differences
The United States leads in agent development and early enterprise adoption. European enterprises face additional considerations under the EU AI Act, which sets obligations depending on the risk level of AI systems. In Asia, adoption varies, with strong interest in markets such as Singapore, Japan and India. Australian enterprises are active experimenters, and local regulators and industry bodies have been developing guidance on safe and responsible AI use.
Governance for agents
As agents take on more tasks, organisations need clear rules for how they are used. Practical governance measures include defining which actions an agent may take without approval, logging every action so it can be reviewed, setting limits on spending or data access and assigning a named person responsible for each agent’s outcomes. Regular testing helps catch errors before they reach customers. These measures may slow early deployment, but they build the trust needed to expand agents’ responsibilities over time and give founders a clearer story to tell enterprise buyers.
Frequently asked questions
How are AI agents different from chatbots?
Chatbots respond to prompts. Agents plan and carry out multi-step tasks, often using other software tools, with some autonomy.
Are AI agents replacing jobs?
Agents automate some tasks, but most current deployments assist workers rather than replace roles entirely. Effects vary widely by task and industry.
What should enterprises check before deploying an agent?
Data access and security, error handling, human review processes, accountability for actions and how success will be measured.
References
- McKinsey: The state of AI in 2025 – Agents, innovation, and transformation
- IT Brief UK: McKinsey report shows AI interest but slow scaling
- Carta: State of Private Markets Q1 2026
Related reading
General information only. This article reflects the author’s views at the time of writing. It does not take into account your objectives, financial situation or needs, and it is not financial, legal or tax advice or a recommendation to invest. Figures are from the sources named and may change. See our terms of use.