Autonomous generative AI agents: Under development

Autonomous generative AI agents, or "agentic AI," are software solutions designed to perform complex tasks with minimal human oversight. Unlike current chatbots or co-pilots, which require constant human interaction, agentic AI can sense its environment, break down jobs into steps, make decisions, use tools, collaborate with other systems (multiagent systems), access memory to learn, and execute tasks autonomously to achieve human-set goals.

Deloitte predicts that in 2025, 25% of companies using generative AI will launch agentic AI pilots or proofs of concept, growing to 50% in 2027. Some applications may see adoption into workflows by late 2025.

Potential Benefits:

• Increased Knowledge Worker Productivity: Agentic AI can automate multi-step processes across business functions, potentially boosting the stagnant productivity of knowledge work.
• Enhanced Capabilities: Built on improving Large Language Models (LLMs) and augmented by other technologies (like chain-of-thought reasoning and multimodal data processing), agentic AI is more flexible and capable than traditional automation methods.
• Driving Value from Gen AI Investments: Agentic AI could unlock the quantifiable business value that has often been elusive with earlier generative AI implementations.

Promising Use Cases:

• Customer Support: Handling more complex inquiries and automating resolution.
• Cybersecurity: Detecting attacks and generating reports autonomously, reducing human workload.
• Regulatory Compliance: Analyzing regulations and documents to determine compliance status and provide proactive advice.
• Agent Building and Orchestration: Tools are emerging to help companies build their own custom agents and multiagent systems.

Current State and Challenges:

Agentic AI is still in early development and adoption phases. Current agents can make errors, get stuck in loops, and "hallucinations" can spread in multiagent systems. Reliability for enterprise use is a key challenge.

Recommendations for Companies:

• Prioritize & Redesign Workflows: Identify high-value tasks suited for agentic AI and optimize processes before deployment.
• Focus on Data Governance & Cybersecurity: Ensure strong foundations are in place as agents require access to sensitive data and systems. Many current gen AI users are not highly prepared for this.
• Balance Risk and Reward: Start with low-risk use cases involving non-critical data and human oversight ("human on the loop") to build capabilities before moving to higher-value, more autonomous applications.
• Maintain Healthy Skepticism: Evaluate claims carefully, as challenges may take time to resolve, and real-world performance might lag behind demos.

While the vision of highly autonomous, reliable agents remains under development, the rapid evolution of the technology means companies should prepare now to leverage this potential productivity leap.