2025, the industry is racing to adopt Agentic AI - autonomous systems that are capable of reasoning, planning, and executing tasks with minimal human oversight. Over 72% of organisations report that they are currently using some form of Agentic AI to automate tasks and enhance customer experience with plans to expand the use of Agentic systems into more complex functionalities such as decision-making assistance and strategic planning.
Whilst adoption is growing, so are the concerns around governance. It's reported that 76% of leaders report that AI governance is their top concern. With the sheer acceleration of the rate of adoption, it's easy to lose sight of arguably one of the most important questions when considering the adoption of AI. How do we embrace autonomy without losing control?
From Generative to Agentic
The last wave of AI adoption was heavily driven by generative systems, tools that could create text, images and code at 10x the rate of your average developer. For the sake of this post, we'll avoid the quality and "vibe-coding" conversations here. Yes - the quality of the code generated can be called into question at times, but when placed in the right hands and when used to accelerate an existing skillset as opposed to being a replacement for one, AI is undoubtedly a net positive.
With this said, generative AI is ultimately reactive. It requires human prompts, constant oversight, and rarely delivered business impact without requiring additional layers of process and review.
Agentic AI represents the next stage of this evolution. As opposed to waiting for instructions, agentic systems are designed to pursue a defined set of goals. They can break down complex tasks into smaller steps, making decisions and invoking actions within their environment under minimal supervision. We reviewed the different types of agents in A Practical Guide to AI Agents and if you haven't read it, I would recommend checking it out. In practice, this shift means we are moving away from reactive, one-off outputs to continuous, autonomous workflows.
For organisations, this shift is quite significant. Generative AI has helped accelerate productivity in pockets - summarising documents or generating code snippets. Agentic AI, in contrast to this, is capable of orchestrating entire workflows - ultimately delivering outcomes rather than suggestions.
Why Businesses are Investing
The promise of Agentic AI lies in its ability to unlock value at scale. For many organisations, early experiments with generative tools showed potential but delivered only incremental gains. By contrast, Agentic AI is positioned to be a driver of measurable outcomes across organisations, delivering tangible business value. With 72% of organisations already using AI and a further 21% planning adoption within a year or two, this is no longer just experimentation at the margins - it's a signal organisations see real, measurable value and are putting weight behind agentic AI.
The most immediate driver for adoption is efficiency. Nearly three quarters of business leaders (74%) cited operational efficiency as their top reason for investing. The capability to orchestrate and carry out complex workflows that agentic AI provides makes it an attractive option for cost reduction, faster response times and scaling services.
As with anything pertaining to AI over the past year, the expectations are high and this confidence on ROI is further fuelling adoption. In fact, 62% of organisations anticipate returns of over 100%, some expecting figures of over 171% with the optimism being reflected in the budget. Almost half of all companies surveyed (49%) have carved out new budgets for agentic AI initiatives, whilst another 35% have reallocated funding to support development.
Beyond the common boastings of productivity and cost savings, leaders have reported that they see a strategic potential in the use of Agentic AI. It's already being used to accelerate research and development, optimise decision making and even train the next generation of LLMs. It's not easy to quantify the productivity impact that agentic AI will have across an entire industry, but to put a figure to it - analysts have estimated that agentic AI could contribute between $2.6 and $4.4 trillion to the global GDP by 2030, underscoring its role as not simply just an efficiency tool, but instead as a mass driver of economic growth.
The Moral Crumple Zone
As adoption accelerates, so too does the concerns surrounding how agentic AI is managed. Gravitee's 2025 survey again highlights this clearly with 76% of leaders ranking AI governance as their top concern when deploying agentic systems. The enthusiasm for automation and efficiency is certainly real, but so is the recognition that autonomy introduces risk.
Governance in this context isn't only about compliance - it's about trust, accountability, and control. Autonomous systems are capable of making decisions, interacting with customers, and invoking actions with third party tooling. Without the correct guardrails, this freedom and enhanced autonomy can quickly become a liability. Leaders are right to worry about this, it's not a lack of faith in the technology - it's a concern that it's implemented in the correct, most responsible way.
The risks extend to accountability. Researchers have raised concerns about "moral crumple zones", where responsibility shifts away from humans and falls unfairly on the technology when failures occur, blurring the line of accountability. In industries with high regulatory stakes, such as finance and healthcare or critical infrastructure, this can raise some difficult questions. Who is accountable when an agent makes a bad decision? How do organisations ensure transparency when reasoning chains are complex and adaptive?
The governance dilemma is further complicated by the rapid pace of adoption. With 72% of organisations already using agentic AI, many are building governance frameworks on the fly - patching risks as they emerge rather than planning ahead. Reactive ethics and governance duct tape may hold in the short term, but at scale this can only introduce real vulnerabilities.
The challenge then, is to find the balance. Embracing autonomy without losing control. This is undoubtedly easier said than done but there are things which we should keep in mind as we traverse this new, unproven agentic landscape. Governance should be embedded into the design of agentic systems from the start and not treated as an afterthought. Agentic systems should be aligned with technical safeguards and ethical principles - ensuring that the drive for efficiency doesn't outpace the organisations' ability to manage risk.
Trends and Applications
Whilst governance remains a challenge, organisations are not slowing down in their exploration of what agentic AI can do. The diversity of emerging agentic applications is one of the strongest signals that the technology is moving beyond hype and into applied practice.
Customer Experience and Support
Early adoption of agentic systems has focused heavily on service orientated functions, where agents are being used to deliver faster, more consistent customer support. In contrast to traditional chatbots, these systems can resolve end-to-end issues, escalating to human intervention only when necessary. For many organisations this has already translated into faster resolution times and improved user experience.
Decision Making and Business Operation
Beyond customer engagement, agentic AI is providing valuable, real-time decision-making functionalities. Logistics agents are being deployed to monitor inventory, predict demand and adjust supply chain flows based on dynamic input. In finance, they are assisting with risk analysis, compliance monitoring and fraud detection - areas where speed and competency directly impact the bottom line.
Productivity and Knowledge
Knowledge-intensive industries are experimenting with agentic AI to augment research and accelerate development. From scanning large datasets and summarising findings, to proposing hypotheses and modelling potential outcomes, agentic systems are beginning to accelerate discovery cycles that once took weeks or months. We're even seeing this with agentic systems used to aid the training of the next generation of LLMs.
With this in mind, we're also seeing a number of markers across the industry that demonstrate the trend towards the wide adoption of the agentic system model:
- Multi-agent collaboration, where specialised agents coordinate with one another to complete complex tasks.
- Domain-specific agents, tuned for industries like healthcare, manufacturing, or legal services, offering higher precision and reliability.
- Integration standards, such as the Model Context Protocol (MCP), which are emerging to make it easier for agents to interact with enterprise systems in a secure and consistent way.
- Task-based pricing models, where organisations pay per outcome delivered rather than per hour or per seat, shifting how AI investments are costed and measured.
The organisations leading the way are those willing to experiment across functions, while also investing in the infrastructure and governance needed to sustain adoption at scale.
Conclusion
Agentic AI is no longer a concept on the horizon - it's here, and it's reshaping how the industry operates and competes. Adoption is rising, use cases are diversifying and the potential for value add is increasingly significant. With that said, the question still lingers, how do we embrace this new era and maintain our control?
The answer is balance. The organisations and industries that succeed will be those that pair bold experimentation with strong governance, building systems that are not just intelligent but also trustworthy, accountable, and transparent.