Beware! The incoming wave of AI agents in 2025

What are AI agents?

Like ChatGPT, AI agents are powered by LLM’s but differ in that they can run independently controlling their own workflows. They operate in systems, adapting to new information and situations.

Key capabilities of AI agents include:

• Autonomous decision-making

• Tool integration for enhanced functionality

• Real-time task execution

• Adaptive learning for improved performance

• Multi-agent capability

• Human-in-the-loop (HITL) functionality

The current state of AI agents

According to LangChain’s report, AI agents are no longer a fringe technology; they are rapidly becoming integral to modern operations. Nearly 50% of surveyed organisations already have AI agents in production, with mid-sized companies leading the way. More so, 78% of organisations plan to implement AI agents soon.

The most popular use cases in 2024 were to use AI agents for research and summarisation, boosting personal productivity, and delivering customer service, where 45.8% of respondents use agents to manage interactions

The impact of AI agents in 2025

In 2025, AI agents are expected to redefine workforce dynamics. We expect this transformation to start within professional services scaling to use-cases involving more advanced hardware such as drones tech and in that; cross-functional workflows, operating across 3rd party platforms. What we also expect is for the integration of multi-agent systems. These will tackle complex challenges across a range of professions, such as project simulations, marketing campaigns, product testing and more.

In project delivery

Projects will be delivered with greater automation by hand of these AI agents. Take resourcing for instance, an agent can analyse project data, and proactively reallocate resource depending on various criteria such as the critical path. When integrated in a major project where risk is elevated, it can suggest resourcing allocation maintaining the expertise of the HITL. Moving into more advanced agentic systems such as safety monitoring, AI-enabled sensors and computer vision systems can then monitor worker movements and site conditions in real-time. These systems can then detect potential hazards, ensure compliance with safety protocols, and alert supervisors to risks, preventing accidents.

Expected business benefits

The adoption of AI agents promises several tangible benefits:

• 24/7 operational capabilities

• Real-time data analysis and decision-making

• Enhanced scalability and reduced labour costs

• Automation of repetitive tasks, freeing employees for value-added work

• Improved customer experience through personalisation

• Scalability for small and medium-sized businesses

The biggest concerns for using AI agents

With benefits come challenges. The biggest business concerns for deploying agents are:

Performance quality (Biggest concern): LLM’s in general are still inconsistent in performance across a variety of tasks. Relying on agents with increased autonomy and inconsistent performance can have a detrimental impact. For this reason AI-human hybrid systems are essential

Cost: As systems scale in complexity so will the costs. Agentic workflows are new, unfamiliar and natively complex, so businesses are likely to be reserved in how they navigate spending as costs are uncertain.

Safety concerns: Feeding off of performance quality, giving agents more agency reduces governance and human control. Couple this with sensitive data creates a risk. Compound this with poor visibility into the underpinnings of how these agent workflows interact is a recipe for disaster. Safety protocols are key.

Latency: Like humans, as systems grow, processes establish and lines of communication increase, output can take longer, particularly with more complex and high-value tasks. The same applies for AI - outputs take longer.

How companies anticipate implementing AI agents

So how do you maximise the upside and minimise the downside? In reading the LangChain report, our takeaways are to always start with phased integrations, using pilots in non-critical business functions. This needs to happen in parallel with upskilling to bridge the skill gap as highlighted by the report, and also careful, accountable governance to deploy agents safely. As with all tech deployments, continuous observation is essential to understand the decisions agents make, how they make it, and why they make it. This practice, known as tracing, emerged as the most popular control measure cited in the report.

To sum up

AI agents represent a seismic shift in how businesses operate. With half of organisations already leveraging these technologies and many more planning to adopt them, the landscape is rapidly evolving. While challenges remain, the potential for enhanced efficiency, scalability, and innovation makes AI agents a cornerstone of the future enterprise. By embracing robust strategies and addressing workforce concerns, companies can harness the full potential of AI agents as they transition from an emerging technology to a business necessity.

Listen to a short podcast of the LangChain State of AI Agents report

The rabbit hole 🐰

The full LangChain State of AI Agents report 🔗

How AI agents are transforming AEC 🔗

Top predictions for AI agents in 2025 (Saleforce) 🔗