From Salesforce to Google: how Big Tech will bring Ai agents into companies

Various experts call them tools that are smarter than chatbots, because they are capable of dialoguing and integrating with each other and interacting with humans, and are therefore destined to spread into corporate IT systems and architectures not only as 'front-end' elements. Welcome to the world of 'AI Agents', a world in which the (at the moment) most advanced expression of artificial intelligence will no longer act as a simple assistant but will take actions on our behalf and suggest what to do to improve specific functions, leaving the human being with the task of direction, supervision and control and the power to devote himself to higher value activities. A recent report by Gartner has ridden these concepts to the limit by stating that the future of AI is probably here, inside these tools. In this (very) near future, according to Accenture's assessments, work teams will be hybrid, part human and part machine, and within the next three years, organisations operating with orchestrated ecosystems of multiple AI agents under human supervision will grow incrementally. Everyone - this is also the vision of Jensen Huang, the charismatic CEO of Nvidia - will be working alongside 'AI employees' with a specific goal: to automate workflows (such as handling service requests or managing customer data) without requiring human intervention and without the need for prompts or specific instructions, as is the case with ChatGPT-like assistants.

This game is being played by specialised companies such as Crew AI, Snowflake, ServiceNow or the very Italian Indigo.AI, which are offering platforms and tools for the creation and orchestration of AI agents in various ways, as well as realities well known to the general public such as Open AI (with its Operator) and nonetheless the big names in the tech galaxy. One of these is Salesforce. No more than a few weeks ago, its CEO, Marc Benioff, in fact sentenced in an interview with the Wall Street Journal how the future of artificial intelligence lies in autonomous agents and not in large-format language models, whose prospects for further development would be limited. The former's abilities to perform tasks independently, in a nutshell, represent the new course of AI and will be used in a wide range of enterprise-class applications (marketing and customer service above all, but not only) to increase productivity and process efficiency. The solution Salesforce has created within its own platform is AgentForce, a suite for generating and distributing AI agents that are fully customisable and capable of operating autonomously in any area of the company, managing the analysis of large volumes of data (structured and unstructured), making decisions and adapting to different contexts and sectors (from retail to healthcare via financial services and manufacturing) with the precise aim of maximising performance. The message that accompanied the announcement in early autumn of Agentforce Service Agent, the first pre-configured agent available on a large scale, is explicit: to go beyond the functionality of traditional chatbots by supporting a wide range of tasks, from the simplest to the most complex.

Microsoft is also moving on this same front and this was also reiterated by CEO Satya Nadella during his recent visit to Italy, where he christened the release of the new Dynamics 365 Copilot 'autonomous agents. Technological evolution (at the level of data and interfaces) and neural networks, according to Redmond's number one, have led us into a world of cross-organisation agents that will operate like AI apps on behalf of an individual, a team or a corporate function, optimising workflows and automatically executing certain business processes, from sales to customer service to supply chain management. In large companies but also in SMEs and PA. These agents are brought to life by a dedicated platform (Copilot Studio) and this model is fed not only with data and information from the Microsoft application ecosystem (the 365 suite) but also with data and information from other platforms (SAP for example) and through a wide range of connectors that enable the agent to run on multiple systems.

And the other BigTechs? They are certainly not standing idly by, moving on different planes and at different speeds. The latest edition of re:Invent, for instance, allowed Amazon Web Services to launch its new family of generative AI models for video and images 'Nova' and to strengthen the weight of the Bedrock platform (where the shopping chatbot Rufus is already resident and where Alexa will be increasingly integrated) with the announcement of cloud-based tools (Bedrock Agents) to create and manage agents that automate customer service or order processing. Another enterprise software giant such as Sap, on the other hand, dropped its first ace back in September 2023 by presenting Joule, its generative AI co-pilot, now evolved with autonomous agents designed to work collaboratively on specific business processes, such as accounting updates or invoice and dispute management. Creating virtual agents based on machine learning and natural language analysis tools for customer service and business automation applications is also the task of Watson Assistant, IBM's artificial intelligence platform. As for Oracle, the latest shot fired by the American multinational dates back to September, on the occasion of the CloudWorld 2024 edition held in Las Vegas, where Oracle launched more than 50 new 'business-to-business' AI agents dedicated to different corporate roles. These are in essence virtual consultants based on generative artificial intelligence and natively integrated into the Fusion Cloud Applications suite.

And finally, 'last but not least', Google. This week, the Mountain View giant took the wraps off the latest version of its AI model, Gemini 2.0, and outlined the horizons in terms of experimental projects in the field of AI agents that will fall within the natural language development platforms (Vertex AI Agent Builder and Dialogflow) of the Google Cloud. These range from agents capable of spatial and multimodal reasoning (Project Astra) to others that can perform complex tasks (Project Mariner), from coding agents designed for developers (Jules) to tools designed to help users operate in the physical world by applying Gemini 2.0 capabilities to robotics.