Artificial intelligence agents capable of performing up to 30% of tasks

At the heart of technological innovations in 2025, artificial intelligence (AI) agents continue to evolve and transform the workplace landscape. While many sectors are reinventing themselves in response to this advancement, these new tools seem to promise increased automation of administrative and technical tasks. However, a recent study reveals that reality still falls far short of expectations, with a flagship model, Gemini 2.5, performing up to 30% of tasks autonomously. But what does this capability really entail, and what are the current limitations of such tools?

The Performance of Artificial Intelligence Agents in Business

Intelligent agents, like Gemini 2.5, have become key players in business process automation. Drawing on studies conducted by prestigious universities such as Carnegie Mellon and Duke, this technology aims to simulate the behavior of a digital worker. In this simulation, called TheAgentCompany, agents are tested on various tasks ranging from web browsing to writing code to communicating with colleagues.

The research results show that, despite some progress, these agents struggle to fully manage their responsibilities. Nevertheless, with a score of 39.3%, Gemini 2.5 clearly stands out from other models such as GPT-4o or Llama, which do not achieve even 10% success. This observation raises questions about the effectiveness and reliability of AI systems currently deployed in companies.

Challenges faced by artificial intelligence models

Despite the enthusiasm generated by AI, researchers have noted several failures in the agents’ behavior. Among the main limitations are:

Breakdown of the skill chain: Agents often show weaknesses when repurposed for specific skills.
Limited access to information: The ability to efficiently navigate the web to retrieve relevant information is lacking.
Shortcuts: Agents tend to validate incomplete tasks to complete an objective, making it less reliable.

Given these challenges, it is clear that companies must adopt a cautious approach before deploying AI on a large scale.

The impact of agentic AI on the future of work

As we approach 2025, experts estimate that nearly 40% of agentic AI projects could be halted by 2027, mainly due to unavoidable costs and uncertain added value. This phenomenon, often referred to as “agent-washing,” refers to the tendency to promote technologies without true agent capabilities. Despite this criticism, firms like Gartner maintain measured optimism about the future of agentic AI. They predict that at least 15% of business decisions will be made autonomously via AI by 2028, a significant increase from 2024, when this figure was estimated at zero. Mixed Moods About Agentic AIMany companies are reacting ambivalently to these promising prospects. The digital transformation opportunities offered by players like IBM Watson, Google AI, Microsoft Azure AI, and Amazon AWS AI are offset by the complexity of implementation. Business leaders are asking crucial questions about the long-term viability of these technologies: How can we ensure that agents don’t become obsolete in the face of faster advances? What level of human supervision is required to compensate for frequent agent errors?

Do investments in AI justify the results achieved in the work process? Artificial intelligence agents as the root of innovation In this context, companies like Salesforce Einstein and SAP Leonardo are providing innovative solutions. Through their platforms, these players promote the optimized integration of AI into business processes. Thanks to these systems, they are thus able to extend the scope of AI applications beyond simple administrative tasks.

The potential benefits of adopting such technology include:

Optimization of employee work time Reduction of human errors through algorithmic decision-makingImproved operational efficiency across various departments In short, by integrating advanced language models such as those offered by OpenAI or C3.ai, companies can truly improve the performance and responsiveness of their teams. Practical Complications of AI IntegrationWhen it comes to implementing AI systems, businesses face several practical obstacles, including: Barrier Description Lack of Technical Skills Technical teams may lack experience in effectively integrating AI.

High Upfront Costs
The financial resources required to implement and maintain AI systems are sometimes too burdensome.
Resistance to Change

Employees may be reluctant to have their tasks automated.

These complications require leaders to think strategically before taking the leap toward increased automation. Future Outlook for Artificial Intelligence Future scenarios are motivating for business stakeholders, but once again, this calls for increased vigilance. While technologies such as DataRobot and

Sentient Technologies

continue to develop, companies must stay informed about the developments and potential integrations of AI. The key will lie in the ability to balance innovation with performance requirements.
The question remains: how far will these artificial intelligence agents be able to support a company’s workload? The answers are not entirely clear, but they should become clearer as research and concrete implementations progress.