Vendors
日本語
search
  2. //
  3. Reports & insights
  4. //
  5. The Future of GenAI is Agentic

The Future of GenAI is Agentic

Create a vendor selection project
Click to express your interest in this report
Indication of coverage against your requirements
A subscription is required to activate this feature. Contact us for more info.
Celent have reviewed this profile and believe it to be accurate.
We are waiting for the vendor to publish their solution profile. Contact us or request the RFX.
Projects allow you to export Registered Vendor details and survey responses for analysis outside of Marsh CND. Please refer to the Marsh CND User Guide for detailed instructions.
Download Registered Vendor Survey responses as PDF
Contact vendor directly with specific questions (ie. pricing, capacity, etc)
8 September 2024
Jaspaul Saini

Currently, there is a great deal of excitement in the artificial intelligence (AI) community about agentic AI, multi-agent systems and GenAI agents. This has nothing to do with spy novels or insurance agents but everything to do with the future of GenAI.

In the words of Andrew Ng, the founder and the lead of Google’s Brain Project, “AI agent workflows will drive massive AI progress this year—perhaps even more than the next generation of foundation models”.

Agentic AI refers to artificial intelligence systems that possess autonomy and decision-making capabilities, allowing them to act independently in complex environments. They can an adapt their behavior based on real-time data, learn from experiences, collaborate with other systems, and make decisions aligned with specific goals. For example, autonomous vehicles demonstrate agentic AI by making real-time decisions based on traffic conditions and environmental factors.

In the GenAI world, agentic concepts are being introduced through GenAI agents. These agents are role-based applications that specialize on a specific task or objective. They act autonomously to determine the best way to complete their tasks. By utilizing multiple agents, a multi-agent system can be created, where agents coordinate and collaborate to solve complex problems and automate processes. This constant learning and improvement can lead to significant process enhancements and innovative approaches.

Another interesting application of multi-agent systems is using multiple LLMs to engage in discussions and arguments, enhancing performance and answer accuracy.

A good example of where Agentic AI can be applied to provide end-to-end process capability is in software development. By employing multiple agents specialized in different roles, such as solution architects, developers, data engineers, and testers, the entire software development process can be automated. While this may sound ambitious, there are already Agentic AI applications claiming to achieve this, such as Devin.ai and ChatDev.

Personal assistants are another obvious category where an agentic AI approach will be crucial. Expectations of what an AI assistant could do, have been shaped by numerous Hollywood movies, and expectations are high. Apple's announcement of Apple Intelligence positions them well to deliver a consumer-focused AI personal assistant, given their large user base and commitment to a multi-agent AI platform. Startups like Mindy have also developed agent-based personal assistants that act as "everybody's personal Chief of Staff." You can only communicate with it using email and it will take actions on your behalf.

An AI agent development ecosystem is already emerging, with various platforms, frameworks, and tools supporting agentic system development, including Crew AI, Zapier Central, Relevance AI, MindStudio, Microsoft AutoGen, and LangChain.

However, there are challenges and considerations to address. When it comes to building GenAI systems, the industry best practice is ‘human in the loop’. This is in direct contrast to agentic AI, with its focus on autonomous AI. It seems a halfway ground will need to be established, with strong AI governance frameworks and guardrails to oversee automated AI processes.

Also, nobody wants an upstart little AI that can you do your job better than you can. The potential impact on employees should be carefully considered to ensure the technology's long-term sociological implications are managed appropriately.

Comments