Multi-Agent AI lay foundation for AGI-ready enterprises, says Cognizant CAIO – TahawulTech.com
Babak Hodjat, Chief AI Officer at Cognizant, explains how agentic architectures are helping organisations move from isolated AI experiments to scalable, governed systems that deliver measurable ROI and long-term operational resilience.
Enterprises are moving beyond isolated generative AI experiments towards systems that can operate reliably, adaptively, and at scale. Agentic architectures are increasingly shaping this transition, redefining how AI is designed, deployed, and governed within complex organisations. In this interview, Babak Hodjat, Chief AI Officer at Cognizant, shared a practitioner-led perspective on the evolution from early single-agent assistants to today’s modular, multi-agent systems with Tahawultech.com.
Hodjat explains why multi-agent approaches are proving essential for navigating fragmented data, legacy infrastructure, and intricate workflows. Agentic AI will enable incremental ROI, reshape digital operations over the next decade, and brings organisations closer to AGI-like capabilities while embedding governance, trust, and accountability from the outset.
Interview excerpts:
How do you see the evolution from single-agent systems like early voice assistants to today’s modular, multi-agent architectures?
When we developed the Dejima technology, the natural language technology behind Siri, there were no AI assistants, and the state of the art was based on grammatical analysis of language. Our main innovation was to represent various functionality and services as simple AI agents in a network of agents, collaborating in response to user queries and commands. Today, we are able to fit the entire knowledge of humanity into an AI agent in the shape of an LLM, which is able to understand, generate, and communicate using natural language. Today’s multi-agent systems are composed of specialised agents that collaborate—planning, retrieving context, using tools, and coordinating with one another—so they can carry out complex workflows with reliability. This shift enables enterprises to dynamically respond to evolving needs and goals, moving from static Q&A to adaptive orchestration, where agents can reason collectively and self-correct, reliably responding to previously unforeseen situations and context.
Why is a multi-agent framework emerging as the most practical path for enterprises that operate with fragmented data, legacy apps, and complex workflows?
Enterprises are messy by nature: legacy systems, large amounts of often disparate applications, and siloed data. Agentic architectures let us represent disparate data sources and APIs with collaborating agents, working together to provide coherent knowledge retrieval and tool use, making the complexities transparent to the organisation. Modernisation in the form of upgrades and consolidation of API and data systems is still beneficial, making systems more efficient and exposing more functionality, but this can now happen less painfully and more incrementally, with the unified interface and mapping from intent to processes being handled by the agents. This approach also reduces risk because agents can be deployed in controlled domains before scaling across the enterprise. Multi-agent frameworks, like our open-source Neuro AI multi-agent accelerator, are easy to upgrade and wrap legacy systems by providing a bridge between natural language and data retrieval and API calls. This enables enterprises to coordinate and orchestrate AI agents from various sources, whether internal or third-party. This interoperability means businesses can adopt an agile iterative process for AI projects that is guided by measuring efficiency, productivity, and quality gains while guaranteeing that trust and safety are built into all elements of design and implementation. This approach empowers all elements of the organisation to contribute and expand AI use through a process of sandboxing and governance, ensuring safe and rapid scaling.
Many organisations struggle to realise ROI from AI. How do agentic systems enable incremental adoption while still delivering measurable business impact early on?
Agentic systems let you start small. Automate one end-to-end workflow—service desk, claims, supply chain planning—measure cycle time, error rate, and handoff reduction, then scale to adjacent processes. The gains show early because agents attach to existing systems and encode operational checks. This means enterprises can demonstrate tangible business impact within weeks, not years, while building confidence for broader transformation. It’s a pragmatic path that aligns with today’s expectations: low upfront investment, measurable KPIs, and scalability baked in.
“Our own multi-agent system that we launched in mid-2025 at Cognizant for our employee intranet, 1Cognizant, has already resulted in 45% reduction in incident tickets across 11 million inquiries.”
What will this shift mean for how businesses build, manage, and scale digital operations over the next decade?
Digital interactions are still largely human-to-machine or code-to-code. The Agentic Web is a fast-evolving model for how the internet currently operates, in which robust, future-proof, semi-autonomous agents will increasingly handle transactions, negotiations, and exchanges on behalf of individuals and organisations. For operations, this demands systems capable of orchestrating workflows, contextaware decisioning, and selfservice—with governance programmed in end-to-end. Traffic from autonomous agents is already growing; this means organisations will increasingly need to ensure that systems are coherent, reliable, and efficient for software intermediaries. As autonomous AI agentic systems begin to represent users, businesses will need to create digital environments that machines can read and understand clearly, or risk losing visibility with these new digital customers in the marketplace. There is also an immediate need for trust protocols and standards for inter-agent communication.
Looking ahead, how do multi-agent architectures move us closer to AGI-capable systems — and what guardrails must enterprises put in place to adopt them responsibly?
Before AI is close to what we consider ‘AGI’, several fundamental breakthroughs will still be needed: Large Language Models are still struggling with key AGI markers such as memory retention and online learning. But what multi-agent systems do already very well is combining specialisation (by breaking complex workflows into individual tasks), coordination (by enabling collaborative interaction and communication), and reliability (by facilitating collective reasoning and error mitigation). To deploy them responsibly, organisations need to apply oversight and controls like our TRUST framework—which is itself a multi-agent system that provides real-time auditability, governance, testing and validation, transparency and documentation, balancing innovation with accountability.
Given your experience at Cognizant, where do you see the fastest adoption of enterprise-grade AI agents in the Middle East, and what differentiates this region’s readiness?
Building on a mix of the region’s entrepreneurial spirit, progressive regulation, AI talent, and cost-effective data centers, the Middle East has the ingredients to be ahead on widespread adoption and innovation in multi-agent systems. There have been early investments in core open-source AI models here, which have kept the region at the forefront of research and development in core AI. Building on this momentum, localised agentic systems for consumer and business use are conceivable.
“The region can also pioneer inter-institutional multi-agent systems by facilitating standards and sandboxes for validating agentic systems that would span consumer, financial, and business networks in a trustworthy manner, enabling increased reliable autonomy of agentic systems.”
What misconceptions do business leaders still have about AGI or agentic systems, and how should they be reframing their expectations?
As businesses turn to AI to automate long, complex, business-critical processes, leaders should remain mindful that bigger doesn’t automatically mean better. Here’s why: today’s large language models struggle with long, multi-step reasoning. They can make catastrophic mistakes after only a couple of hundred reasoning steps, and in enterprise workflows, these mistakes aren’t small—they compound. That’s why organisations are hitting a ceiling with AI agents. Recent research by our Cognizant AI Lab suggests a better approach: to use many smaller AI agents working together. Our new system, MAKER, solved a million-step reasoning problem with zero errors—something no single model has ever done. Error correction methods, analogous to what made telecommunications systems viable, will unlock reliable, enterprise-grade multi-agent systems.
As AI agents begin to collaborate, reason, and self-optimise, how should enterprises rethink talent, workflows, and governance?
AI will increasingly move from just augmenting our work to AI agents as team members, with a level of autonomy in their operations and human-on-the-loop oversight. When it comes to talent, the workforce needs to evolve from prompt engineers to agent orchestrators. These roles will focus on designing multi-agent systems, managing, maintaining, and extending agentic systems. Domain experts will become ‘agent product managers,’ pairing business and systems knowledge with technical oversight to ensure agents deliver measurable and accurate outcomes. This requires enterprise-wide skilling programmes and organisations empowering their employees to take ownership of innovating and responsibly building their own systems. Workflow processes will shift from siloed applications to agentified ecosystems where connected agents handle multi-step tasks seamlessly. Instead of employees navigating multiple, siloed systems, agents will negotiate and collaborate behind the scenes, improving speed and consistency. Enterprises should prioritise modular multi-agent architectures and adopt an LLM-agnostic approach for robustness and future-proofing. For governance, controlled autonomy and engineered innovation are key. Let teams build their own agents, but plug them into a discoverable, governed multi-agent ecosystem with observability and auditability built in, including logs of agent decisions and redundancy mechanisms like ‘agents monitoring agents’. Finally, humans must remain involved to take responsibility for consequential decisions, ensuring trust and compliance as AI scales across the enterprise.
“In return for the investment, the U.S. tariff rate applied to Taiwanese goods was capped at 15 %”.
Learn more about these moves to improve U.S. technology manufacturing below.
https://www.tahawultech.com/industry/technology/u-s-and-taiwan-enter-trade-deal-over-chipmaking/
#USA #Taiwan #tahawultech
CPI Media and tahawultech are hosting the inaugural Future Enterprise Awards in Riyadh. These awards are designed to recognise IT leaders that are driving digital transformation across the Kingdom.
Learn more below.
https://tahawultech.com/ksa-futureenterpriseawards/2025/
#KSAFEA2026 #tahawultech
“All customer data, including billing and configurations, remain entirely within the region”.
Learn more about @awscloud’s push into Europe below.
https://www.tahawultech.com/industry/technology/aws-invests-in-european-sovereign-cloud/
#AWS #SovereignCloudServices #tahawultech
GET TAHAWULTECH.COM IN YOUR INBOX
source
This article was autogenerated from a news feed from CDO TIMES selected high quality news and research sources. There was no editorial review conducted beyond that by CDO TIMES staff. Need help with any of the topics in our articles? Schedule your free CDO TIMES Tech Navigator call today to stay ahead of the curve and gain insider advantages to propel your business!
Enterprises are moving beyond isolated generative AI experiments towards systems that can operate reliably, adaptively, and at scale. Agentic architectures are increasingly shaping this transition, redefining how AI is designed, deployed, and governed within complex organisations. In this interview, Babak Hodjat, Chief AI Officer at Cognizant, shared a practitioner-led perspective on the evolution from early single-agent assistants to today’s modular, multi-agent systems with Tahawultech.com.
Hodjat explains why multi-agent approaches are proving essential for navigating fragmented data, legacy infrastructure, and intricate workflows. Agentic AI will enable incremental ROI, reshape digital operations over the next decade, and brings organisations closer to AGI-like capabilities while embedding governance, trust, and accountability from the outset.
Interview excerpts:
How do you see the evolution from single-agent systems like early voice assistants to today’s modular, multi-agent architectures?
When we developed the Dejima technology, the natural language technology behind Siri, there were no AI assistants, and the state of the art was based on grammatical analysis of language. Our main innovation was to represent various functionality and services as simple AI agents in a network of agents, collaborating in response to user queries and commands. Today, we are able to fit the entire knowledge of humanity into an AI agent in the shape of an LLM, which is able to understand, generate, and communicate using natural language. Today’s multi-agent systems are composed of specialised agents that collaborate—planning, retrieving context, using tools, and coordinating with one another—so they can carry out complex workflows with reliability. This shift enables enterprises to dynamically respond to evolving needs and goals, moving from static Q&A to adaptive orchestration, where agents can reason collectively and self-correct, reliably responding to previously unforeseen situations and context.
Why is a multi-agent framework emerging as the most practical path for enterprises that operate with fragmented data, legacy apps, and complex workflows?
Enterprises are messy by nature: legacy systems, large amounts of often disparate applications, and siloed data. Agentic architectures let us represent disparate data sources and APIs with collaborating agents, working together to provide coherent knowledge retrieval and tool use, making the complexities transparent to the organisation. Modernisation in the form of upgrades and consolidation of API and data systems is still beneficial, making systems more efficient and exposing more functionality, but this can now happen less painfully and more incrementally, with the unified interface and mapping from intent to processes being handled by the agents. This approach also reduces risk because agents can be deployed in controlled domains before scaling across the enterprise. Multi-agent frameworks, like our open-source Neuro AI multi-agent accelerator, are easy to upgrade and wrap legacy systems by providing a bridge between natural language and data retrieval and API calls. This enables enterprises to coordinate and orchestrate AI agents from various sources, whether internal or third-party. This interoperability means businesses can adopt an agile iterative process for AI projects that is guided by measuring efficiency, productivity, and quality gains while guaranteeing that trust and safety are built into all elements of design and implementation. This approach empowers all elements of the organisation to contribute and expand AI use through a process of sandboxing and governance, ensuring safe and rapid scaling.
Many organisations struggle to realise ROI from AI. How do agentic systems enable incremental adoption while still delivering measurable business impact early on?
Agentic systems let you start small. Automate one end-to-end workflow—service desk, claims, supply chain planning—measure cycle time, error rate, and handoff reduction, then scale to adjacent processes. The gains show early because agents attach to existing systems and encode operational checks. This means enterprises can demonstrate tangible business impact within weeks, not years, while building confidence for broader transformation. It’s a pragmatic path that aligns with today’s expectations: low upfront investment, measurable KPIs, and scalability baked in.
“Our own multi-agent system that we launched in mid-2025 at Cognizant for our employee intranet, 1Cognizant, has already resulted in 45% reduction in incident tickets across 11 million inquiries.”
What will this shift mean for how businesses build, manage, and scale digital operations over the next decade?
Digital interactions are still largely human-to-machine or code-to-code. The Agentic Web is a fast-evolving model for how the internet currently operates, in which robust, future-proof, semi-autonomous agents will increasingly handle transactions, negotiations, and exchanges on behalf of individuals and organisations. For operations, this demands systems capable of orchestrating workflows, contextaware decisioning, and selfservice—with governance programmed in end-to-end. Traffic from autonomous agents is already growing; this means organisations will increasingly need to ensure that systems are coherent, reliable, and efficient for software intermediaries. As autonomous AI agentic systems begin to represent users, businesses will need to create digital environments that machines can read and understand clearly, or risk losing visibility with these new digital customers in the marketplace. There is also an immediate need for trust protocols and standards for inter-agent communication.
Looking ahead, how do multi-agent architectures move us closer to AGI-capable systems — and what guardrails must enterprises put in place to adopt them responsibly?
Before AI is close to what we consider ‘AGI’, several fundamental breakthroughs will still be needed: Large Language Models are still struggling with key AGI markers such as memory retention and online learning. But what multi-agent systems do already very well is combining specialisation (by breaking complex workflows into individual tasks), coordination (by enabling collaborative interaction and communication), and reliability (by facilitating collective reasoning and error mitigation). To deploy them responsibly, organisations need to apply oversight and controls like our TRUST framework—which is itself a multi-agent system that provides real-time auditability, governance, testing and validation, transparency and documentation, balancing innovation with accountability.
Given your experience at Cognizant, where do you see the fastest adoption of enterprise-grade AI agents in the Middle East, and what differentiates this region’s readiness?
Building on a mix of the region’s entrepreneurial spirit, progressive regulation, AI talent, and cost-effective data centers, the Middle East has the ingredients to be ahead on widespread adoption and innovation in multi-agent systems. There have been early investments in core open-source AI models here, which have kept the region at the forefront of research and development in core AI. Building on this momentum, localised agentic systems for consumer and business use are conceivable.
“The region can also pioneer inter-institutional multi-agent systems by facilitating standards and sandboxes for validating agentic systems that would span consumer, financial, and business networks in a trustworthy manner, enabling increased reliable autonomy of agentic systems.”
What misconceptions do business leaders still have about AGI or agentic systems, and how should they be reframing their expectations?
As businesses turn to AI to automate long, complex, business-critical processes, leaders should remain mindful that bigger doesn’t automatically mean better. Here’s why: today’s large language models struggle with long, multi-step reasoning. They can make catastrophic mistakes after only a couple of hundred reasoning steps, and in enterprise workflows, these mistakes aren’t small—they compound. That’s why organisations are hitting a ceiling with AI agents. Recent research by our Cognizant AI Lab suggests a better approach: to use many smaller AI agents working together. Our new system, MAKER, solved a million-step reasoning problem with zero errors—something no single model has ever done. Error correction methods, analogous to what made telecommunications systems viable, will unlock reliable, enterprise-grade multi-agent systems.
As AI agents begin to collaborate, reason, and self-optimise, how should enterprises rethink talent, workflows, and governance?
AI will increasingly move from just augmenting our work to AI agents as team members, with a level of autonomy in their operations and human-on-the-loop oversight. When it comes to talent, the workforce needs to evolve from prompt engineers to agent orchestrators. These roles will focus on designing multi-agent systems, managing, maintaining, and extending agentic systems. Domain experts will become ‘agent product managers,’ pairing business and systems knowledge with technical oversight to ensure agents deliver measurable and accurate outcomes. This requires enterprise-wide skilling programmes and organisations empowering their employees to take ownership of innovating and responsibly building their own systems. Workflow processes will shift from siloed applications to agentified ecosystems where connected agents handle multi-step tasks seamlessly. Instead of employees navigating multiple, siloed systems, agents will negotiate and collaborate behind the scenes, improving speed and consistency. Enterprises should prioritise modular multi-agent architectures and adopt an LLM-agnostic approach for robustness and future-proofing. For governance, controlled autonomy and engineered innovation are key. Let teams build their own agents, but plug them into a discoverable, governed multi-agent ecosystem with observability and auditability built in, including logs of agent decisions and redundancy mechanisms like ‘agents monitoring agents’. Finally, humans must remain involved to take responsibility for consequential decisions, ensuring trust and compliance as AI scales across the enterprise.
“In return for the investment, the U.S. tariff rate applied to Taiwanese goods was capped at 15 %”.
Learn more about these moves to improve U.S. technology manufacturing below.
https://www.tahawultech.com/industry/technology/u-s-and-taiwan-enter-trade-deal-over-chipmaking/
#USA #Taiwan #tahawultech
CPI Media and tahawultech are hosting the inaugural Future Enterprise Awards in Riyadh. These awards are designed to recognise IT leaders that are driving digital transformation across the Kingdom.
Learn more below.
https://tahawultech.com/ksa-futureenterpriseawards/2025/
#KSAFEA2026 #tahawultech
“All customer data, including billing and configurations, remain entirely within the region”.
Learn more about @awscloud’s push into Europe below.
https://www.tahawultech.com/industry/technology/aws-invests-in-european-sovereign-cloud/
#AWS #SovereignCloudServices #tahawultech
GET TAHAWULTECH.COM IN YOUR INBOX
source
This article was autogenerated from a news feed from CDO TIMES selected high quality news and research sources. There was no editorial review conducted beyond that by CDO TIMES staff. Need help with any of the topics in our articles? Schedule your free CDO TIMES Tech Navigator call today to stay ahead of the curve and gain insider advantages to propel your business!
