GenAI 2.0 is here: Why agentic AI runs on real-time data infrastructure

Artificial intelligence (AI) has evolved from predictive AI, which analyzes data to forecast outcomes, to generative AI (genAI), which creates new content. Now, we have agentic AI, where systems act autonomously in the real world. 

Agentic AI refers to AI systems (often called AI agents) that make decisions and take actions without needing constant human prompts or supervision. In other words, an agentic AI has “agency”: it plans and executes goals on its own, based on its understanding of context and data, rather than only reacting to direct instructions. Agentic systems are designed to collaborate with humans and handle complex tasks autonomously. They understand high-level goals set by people and figure out how to achieve them, adapting to new information along the way.

At the 2024 Real-time Data Summit, Tamar Yehoshua, former president of product and technology at Glean, highlighted how these agents are already automating tasks that span departments and functions.

In her presentation, she breaks down the evolution of the all-encompassing genAI era, starting with the internet in the 1990s, moving on to the advent of always-on mobile around 2007, followed by cloud SaaS in the 2010s, and ending with the genAI 1.0 era (think the ChatGPTs of the world), characterized best by the systems in place that help us automate rote work. “Let’s say you found a doc and you want to summarize it, or there’s data that needs to be analyzed. That’s where we are now, with chatbots and starting to get more assistance.” Yehoshua sees agentic AI as the cornerstone of this new genAI 2.0 era.

Unlike generative AI, such as a chatbot that produces text when asked, agentic AI is proactive. It makes decisions and acts in pursuit of a goal, not merely generating content. Early examples of agentic AI include self-driving vehicles, intelligent virtual assistants, and assistant software that can carry out multi-step tasks with minimal oversight. “We're seeing the beginning of agents that do work for you... that won’t just do one task, but will do multiple tasks... We’ll call those multi-step processes or loops,” says Yehoshua. These systems autonomously handle tasks in changing conditions, demonstrating flexibility and independence beyond traditional rule-based automation. 

In the following sections, we describe the characteristics of agentic AI, how it works under the hood, its benefits and applications, and the challenges and infrastructure considerations that come with this emerging technology.

Characteristics of agentic AI

Agentic AI systems typically exhibit several defining characteristics that distinguish them from more static or tool-like AI. The most important traits include autonomy, adaptability, and goal-directed behavior:

Autonomy

Agentic AI emphasizes autonomy, or the ability for AI to operate independently. Such an AI agent performs tasks on its own without human oversight or step-by-step instructions. For example, once given a high-level objective, an autonomous agent decides its course of action and carries it out, rather than waiting for a human to tell it each step. This independence marks a shift from earlier AI assistants that only act when and as instructed. An agentic system chooses which actions to take and when to take them to meet its goals. This autonomy is what gives agentic AI its name; the system itself has agency in decision-making.

Adaptability

Another core feature of agentic AI is adaptability. AI agents learn from interactions and feedback, so they improve their performance over time and adjust to new situations. Rather than following a fixed set of hard-coded rules, agentic AI uses probabilistic models and continuous learning to respond to changing environments. This means the agent can handle scenarios for which it wasn’t explicitly programmed and modifies its strategy when conditions shift. 

For instance, if an autonomous agent encounters an unexpected obstacle or a change in data patterns, it revises its plans on the fly. This capacity to learn and evolve is supported by techniques such as reinforcement learning, where the agent refines its decisions through trial and error and rewards or penalties from its environment. Overall, adaptability helps agentic AI to remain effective as circumstances change around it, which is important for real-time and open-ended tasks.

Goal orientation

Agentic AI is inherently goal-driven. Humans typically provide a high-level goal or define what success looks like, and the AI agent takes on the responsibility of determining how to achieve it. This goal-oriented behavior means agentic AI organizes complex objectives into smaller tasks, plans sequences of actions, and performs those actions to fulfill the overarching goal. 

For example, if the goal is to schedule and book a multi-leg international trip, an agentic AI independently searches for flights, compares prices, adjusts the itinerary based on preferences, and makes the bookings, all without needing a person to micromanage each step. The key is that the agent reasons about the goal and autonomously decides the intermediate steps to reach it. 

This stands in contrast to simpler AI tools that might assist with individual tasks but lack the initiative or reasoning to accomplish a multi-step objective. In summary, agentic AI tools are proactive problem-solvers: Once a goal is set, they continuously ask “What should I do next to get closer to that goal?” and act accordingly.

How agentic AI works

Under the hood, agentic AI systems are composed of autonomous software components known as agents. Each agent is designed with certain abilities and a specific purpose, and they operate together, sometimes in coordination with human-provided instructions or other agents, to tackle complex tasks. These agents draw from large amounts of data and use advanced AI techniques to make decisions. 

Yehoshua describes how agentic assistants today go beyond static tools and enter a more dynamic relationship with data and users. "Let’s say you found a doc and you want to summarize it, or there’s data that needs to be analyzed. That’s where we are now, with chatbots and starting to get more assistance."

Key technologies supporting agentic AI include:

• Machine learning models, especially deep learning neural networks

• Natural language processing to understand instructions or communicate

• Knowledge representation systems that allow an agent to model its environment and context.

Agentic AI often uses large language models (LLMs) as a reasoning engine to help specify tasks and propose solutions in a human-like way. In combination with other specialized models or algorithms, the LLM guides the agent’s decision-making process for planning and problem-solving.

Notably, the architecture for many agentic AI implementations is built on distributed systems to improve scalability and real-time performance. Multiple agents, or multiple instances of an agent, run in parallel across different servers or nodes, sharing information and coordinating their actions. This distributed, cloud-like setup means an agentic AI platform handles large workloads and continues operating even as demand grows, all while maintaining low latency. Agents often communicate with each other and with external systems through defined interfaces or APIs, which enables them to take actions such as querying databases, calling web services, or controlling IoT devices.

In operation, an agent typically loops through a cycle of perceive, reason, act, and learn.

• First, the agent perceives by gathering information from its environment or data sources, such as reading sensor inputs, retrieving data from databases, or observing user behavior.

• It interprets this information to build context. Next, it enters a reasoning phase where it formulates a plan or solution. At this stage, a large language model or other AI logic may be used to understand the task and break it into discrete steps or decisions.

• Then comes the act phase: The agent performs actions to advance toward its goal, often by interacting with external systems through APIs or issuing commands. For instance, an agent might invoke a transaction in a finance system or send instructions to a robot arm. At this stage, agentic AI systems include built-in guardrails or constraints for safety, compliance, and ethical behavior. For example, an agent might be prevented from completing certain high-impact actions without human approval, or it may be sandboxed to operate only within certain limits.

• After acting, the agent learns from the outcome: It takes in feedback on whether the action was successful or if adjustments are needed, refining its knowledge and updating its strategy for next time. This could involve learning from explicit feedback, such as rewards/penalties in a reinforcement learning setting, or analyzing results about whether the plan achieved the desired outcome. 

The cycle then repeats, with the agent continuously improving its performance through each iteration. Through this autonomous loop, agentic AI agents handle complex, multi-step tasks with minimal human intervention. They adapt their behavior based on experience and new data, which makes them robust in dynamic environments. 

To illustrate, imagine agentic AI in charge of stock inventory management: It would perceive sales data and sensor data from warehouses, reason about restocking needs or product redistribution, act by placing orders or rerouting shipments, and learn whether those actions kept inventory optimal, adjusting its strategy for the future. All of this could happen continuously and quickly, driven by streams of data.

Agentic AI’s effectiveness relies on advanced algorithms and a strong infrastructure. Algorithms such as deep learning provide pattern recognition and prediction capabilities, while techniques such as reinforcement learning help the agent improve itself through trial and error. 

Tamar Yehoshua explained how this works at Glean: It combines structured and unstructured data, security, and retrieval augmented generation (RAG). "Your enterprise AI data stack will take that data, make sure it understands its permissions and security, then build a RAG engine and a search index, probably use a vector database, use LLMs, and then build copilots, chatbots, and assistants on top of that data."

Meanwhile, infrastructure elements such as distributed computing, pipelines, and APIs to connect systems give the agent the information and means to act in a timely manner. When designed well, an agentic AI system combines the flexible intelligence of an AI model with the reliability and precision of software engineering, resulting in an AI system that not only thinks, like an LLM generating a plan, but also does, like a software script executing actions, all autonomously. 

Benefits of agentic AI

Giving AI agency offers a range of benefits for businesses and end users. By allowing AI agents to make decisions and take initiative, organizations automate more complex processes and achieve outcomes not possible with traditional tools. Some benefits include:

Increased efficiency and productivity

Agentic AI takes on complex, decision-intensive tasks that previously required human effort, thus streamlining workflows. By automating these processes, companies accomplish more in less time, so employees focus on higher-level strategic work. The autonomy of AI agents means work continues around the clock without waiting for human input, boosting throughput in areas such as customer service, data analysis, and operations. This translates to faster processing, fewer bottlenecks, and lower operational costs.

This aligns with what Glean is seeing in practice, as Tamar Yehoshua noted during her talk.

"These AI agents will be redefining productivity at work."

Personalized and improved customer experiences

Autonomous AI agents deliver responsive and tailored interactions at scale. For example, an agentic AI customer support bot infers a user’s needs and preferences, then provides solutions or recommendations customized to that individual, all in real time. These AI agents operate around the clock, so customers always receive prompt service. The result is a more engaging and satisfying user experience because the AI system dynamically adapts its responses and offers for each person. Consistent, context-aware service powered by agentic AI makes customer more loyal and meets their rising expectations for quick, personalized assistance.

Data-driven decision making

Because agentic AI systems analyze vast streams of data in real time, they discover patterns and insights that lead to smarter decisions. An autonomous agent processes information much faster than humans, scanning market trends, monitoring equipment sensors, or reading through documents, and then acts on that information. This helps organizations respond to opportunities or risks faster and more precisely. Decisions guided by up-to-date data and sometimes by predictive models tend to be more informed, whether it’s an AI agent improving a supply chain route or adjusting financial portfolios based on current market conditions. In essence, agentic AI brings the power of big data analytics into operational decision-making. 

Augmented human capabilities and innovation

Rather than replacing humans, agentic AI is intended to work alongside people, handling routine or complex analytical tasks, freeing humans to focus on creative, strategic, or interpersonal activities. This collaboration leads to higher overall productivity and new opportunities for innovation.

 For instance, in a team setting, AI agents might prepare initial analyses, draft content, or manage low-level processes, while human experts provide guidance, handle exceptions, and make judgment calls. By combining human judgment with AI efficiency, businesses tackle challenges that were previously too resource-intensive. 

Additionally, agentic AI automates or optimizes many processes that were beyond the reach of traditional automation because they required judgment or adaptation. This opens the door to redesigning workflows and exploring new service models that use autonomous agents to better compete in the market.

Applications of agentic AI

Because it autonomously handles decisions and adapts to change, agentic AI has a broad range of applications across industries. Below are some examples:

Customer service and support

One of the most visible use cases is in customer engagement. Agentic AI–powered virtual assistants and chatbots manage customer inquiries, troubleshoot problems, and execute service requests without human intervention. These agents maintain context over long conversations, personalize their responses, and escalate to human agents when necessary. 

For example, a banking virtual agent could guide a customer through securing a loan, or an e-commerce AI assistant might handle returns and refunds autonomously. By operating continuously and learning from each interaction, such AI agents reduce wait times and improve service quality for customers. They effectively act as tireless customer support reps that adapt to each customer’s needs.

Supply chain and logistics

In industries where timing and efficiency are important, agentic AI optimizes operations in real time. Logistics agents analyze large volumes of data, such as tracking shipments, traffic conditions, inventory levels, and demand forecasts, and then decide to route deliveries or adjust inventory allocations on the fly. 

For instance, if an unexpected delay occurs at a shipping port, an autonomous supply chain agent could proactively reroute trucks or find alternate suppliers to prevent running out of stock. These agents continuously re-optimize routes and schedules as new data comes in, improving reliability and reducing costs. As Harvard Business Review noted, an AI supply-chain specialist could dynamically optimize inventories in response to real-time demand fluctuations, something that traditionally required teams of humans to manage. Agentic AI helps make supply chains more responsive and resilient.

Finance and trading

Financial services benefit from agentic AI through intelligent agents that monitor market conditions, execute trades, and manage portfolios. An autonomous trading agent makes split-second decisions based on streaming market data, taking advantage of opportunities or hedging risks much faster than a human trader could. Similarly, agents in personal finance apps might analyze a user’s income and spending patterns to adjust budgets or recommend investments autonomously. In banking, agentic AI could underwrite loans by evaluating risk factors from vast data sources, all while adapting to new regulatory rules or economic changes. By taking advantage of real-time data analysis and predefined risk constraints, these AI agents aim to improve returns and reduce human error in financial decision-making. They effectively act as diligent financial analysts working around the clock.

Software development and IT automation

Agentic AI is also helping in IT operations and coding. Consider them as AI software engineers who go beyond code completion suggestions. Given a description of a desired program, an agentic AI could plan out the modules needed, write the code for each component, test it, and iterate until it meets the requirements. Early versions of this are seen in experimental tools that build simple applications or solve coding challenges autonomously. Likewise, in IT support, an AI agent could monitor systems, diagnose issues, and apply fixes such as restarting services or patching configurations on its own. 

While today’s generative code assistants still require a human developer to prompt and approve their outputs, the agentic approach points to AI handling more of the software development lifecycle itself. Companies are piloting such agents to automate routine programming tasks or manage cloud infrastructure, which could speed development cycles and reduce the burden on human engineers.

Agentic AI is also being explored in other fields:

• Healthcare, such as virtual health coaches or autonomous diagnostic systems

• Manufacturing, such as adaptive control systems on factory floors

• Cybersecurity with AI agents that detect and respond to threats

• Marketing, such as automating campaign management.

The unifying theme is that an AI system takes on a role traditionally done by a person or a rigid program, and performs it with a degree of independent decision-making and continuous learning.

Challenges and considerations

While agentic AI offers advantages, it also introduces challenges and risks. Designing and deploying autonomous AI requires consideration of the following:

Balancing autonomy with oversight

Granting decision-making power to machines can lead to unintended consequences if not properly supervised. A fully autonomous agent might make a decision that optimizes its goal but violates ethical norms or business rules. It’s important to determine the right level of human oversight or “human in the loop” for each use case. 

For instance, an agentic AI managing financial transactions may operate on its own most of the time but should flag high-value or unusual transactions for human approval. Organizations need to establish clear policies about when an AI agent should defer to human judgment. Achieving the right balance helps reap efficiency gains without ceding too much control or risking rogue behavior.

Transparency and trust

Agentic AI systems often use complex neural network models, which means they can arrive at decisions in non-intuitive ways. This opacity, known as the “black box” problem, sometimes makes it hard for users and stakeholders to understand why the AI system did something. A lack of transparency can erode trust, especially in high-stakes domains such as healthcare or finance, where people need to trust the system’s judgment. The AI system could inadvertently incorporate biases in its decisions or treat cases inconsistently, and if humans cannot audit its reasoning, they may be reluctant to rely on it. 

Addressing this requires adding explainability to AI agents, which involves techniques to make an agent’s internal reasoning more interpretable. It also requires clear communication about how the AI system works. Building trust will also involve setting expectations of what the agent can and cannot do and demonstrating reliability over time.

Security and privacy

By their nature, agentic AI agents often interact with numerous systems and handle sensitive data, from personal information to confidential business data. This interconnectedness gives malicious actors more places to attack security. If compromised, an autonomous agent could do real damage. Imagine a malicious takeover of an AI agent that controls industrial equipment or financial accounts. Ensuring strong security measures through authentication, encryption, and strict access controls is important. 

Additionally, these agents must be designed to protect privacy; they should follow data protection regulations and not expose or misuse personal data they handle. Another aspect is safeguarding against “data poisoning” or feeding incorrect information to the agent, which could skew its decisions. Robust cybersecurity and regular audits are essential to maintain the integrity of agentic AI systems in production.

Accuracy and reliability

Even advanced AI agents are not infallible. They can make mistakes or “hallucinate” incorrect information when reasoning, especially if they rely on probabilistic models. An agent might take an action based on faulty understanding or incomplete data and cause an error, such as erroneously ordering an unnecessary supply shipment or misdiagnosing a patient in a medical context. 

Therefore, it’s important to validate the decisions and outputs of agentic AI. There needs to be a framework for catching and correcting errors, whether through human review of critical decisions or automated fail-safes. Accountability is also a question: if an autonomous agent causes harm or a loss, who is responsible? Clear accountability and governance structures must be in place as organizations deploy these systems. 

Ensuring reliability may involve extensive testing of AI agents under various scenarios and continuous monitoring of their performance in the real world, so that any drifts or anomalies in behavior are detected early. While agentic AI can operate without constant human control, it does not absolve humans from oversight. Mechanisms for validation, error handling, and improvement need to accompany any deployment.

In addressing these challenges, experts stress the importance of ethical AI frameworks and robust governance. Researchers and developers are working on guidelines to help agentic AI systems remain fair, accountable, and aligned with human values. Best practices include setting clear boundaries for agent behavior, implementing transparency tools such as audit logs and explainable AI techniques, and maintaining the ability for humans to intervene or override when needed. By considering these issues ahead of time, organizations mitigate risks and deploy agentic AI more confidently.

Importance of real-time data and distributed systems

Agentic AI needs fast, large-scale data and the infrastructure to process it in a timely fashion. Real-time data is often the fuel for intelligent agents: To make informed decisions, an AI agent needs up-to-date information about its environment, users, or system state. Many agentic AI use cases are time-sensitive. For example, an AI system managing inventory must react to changes in demand as they happen, or an autonomous vehicle must respond to sensor readings in milliseconds. If the data feeding the agent is delayed or siloed, the agent’s actions might be ineffective or outdated. 

Yehoshua emphasized that structured data is only a small piece of the puzzle. The challenge is “really about unlocking the right level of enterprise data."

That’s why agentic AI solutions emphasize streaming data and low-latency processing. Autonomous agents analyze data streams in real time, detecting patterns or anomalies and adjusting their behavior based on those insights. This could mean monitoring thousands of metrics from IoT sensors in a smart factory and dynamically tweaking machine settings for optimal performance, or scanning social media trends to let a marketing agent adjust an ad campaign on the fly.

To handle such data-driven operations, agentic AI systems typically rely on distributed computing architectures for scalability and performance. Instead of one monolithic program, multiple agent processes run concurrently across different machines or cloud servers. This parallelism helps the system manage large workloads, such as many customer interactions or data inputs, simultaneously.

A distributed platform also provides redundancy. If one node fails, others can pick up the tasks, which is important for critical autonomous operations. The underlying distributed system means that as the volume of data or number of tasks grows, the agentic AI scales out and continues to meet real-time requirements. In essence, it provides the high-throughput, always-on capabilities that match the agents’ always-learning, always-acting nature.

Another important aspect is integration with data platforms. Agentic AI doesn’t operate in isolation; it often sits on an ecosystem of data sources, from databases and data lakes to live feeds and APIs. Recently, there’s been a focus on unifying diverse data sources into one platform or cloud that agentic AI can tap into. By bringing together structured and unstructured data and making it accessible through a common interface, businesses give their AI agents a holistic view of information. 

Salesforce, for instance, highlights how combining agentic AI with a unified data cloud allows agents to take advantage of enterprise data such as customer records, operational data, and web analytics for more informed automation and better decisions. This is particularly important for agents in areas such as customer service or personalization, which need to draw on many data points from purchase history, support tickets, and inventory levels in real time to do their job effectively.

Real-time, distributed data processing is the backbone of agentic AI. Organizations looking to implement agentic AI must often invest in robust data infrastructure such as high-performance databases, streaming data pipelines, and cloud computing resources to support these autonomous agents. The payoff is that with the right infrastructure, agentic AI can truly shine: making split-second decisions based on the latest information and coordinating across a distributed environment to achieve objectives. This tight integration of AI and distributed systems architecture is what helps applications such as an AI ops agent monitoring a global network by analyzing logs from servers worldwide and fixing issues quickly, or an autonomous drone fleet that collectively responds to events. For companies, such as Aerospike, that specialize in real-time databases, this underscores how important low latency, scalable data access is to the future of AI-driven automation.

What’s next for agentic AI

Looking ahead, Yehoshua envisions a future where intelligent assistants are as common as email clients or calendars. "Everyone will have an expert assistant at work."

Agentic is an emerging AI technology, and we are likely only seeing the beginning of its effects. Industry analysts predict rapid experimentation and AI adoption in the coming years as organizations seek to capitalize on autonomous AI capabilities. According to Deloitte, in 2025, about a quarter of companies using AI will have launched pilot projects with agentic AI agents, and that could rise to half of companies by 2027. 

In the enterprise software realm, major tech players are developing platforms and tools for agentic AI. Tech insiders consider autonomous AI agents to be among the top strategic technology trends, emphasizing AI’s shift toward greater autonomy and adaptability. In fact, more than $2 billion has been invested in agentic AI startups in just the last two years. In addition, a flurry of new frameworks, such as open-source agent toolkits and agent-oriented architectures, have been released to support developers in building these systems.

In the workplace, agentic AI has the potential to redefine roles and processes. We may soon interact with AI “colleagues” who manage portions of our workload. For example, sales teams might have AI agents who autonomously research leads and schedule meetings, or an operations department might rely on an AI logistics planner who continuously fine-tunes supply routes. 

This concept of a “virtual workforce” of AI agents could increase organizational capacity and change how work gets divided between humans and machines. In such a future, human workers would focus more on supervising AI, handling exceptions, and providing the creativity and empathy that AI lacks, while delegating routine decision-making tasks to their agentic AI counterparts. Companies that use this well could see huge gains in efficiency and agility by responding to market changes faster, operating around the clock, and innovating in ways that weren’t possible before.

That said, the journey toward widespread agentic AI comes with caution. The “autonomous” aspect of these agents may take time for broad acceptance, as organizations iron out issues of trust, ethics, and control. Early adopters are learning that human oversight and robust governance need to evolve hand-in-hand with AI autonomy. We can expect new regulations and standards to emerge to address questions such as accountability for AI decisions and safety guarantees for autonomous systems. In addition, there will likely be advances in AI techniques to make agentic systems more transparent and align them better with human intentions, such as improvements in explainable AI and value alignment algorithms.

Next steps with Aerospike

As agentic AI moves from theory into practice, an important next step is ensuring the right data infrastructure is in place to support these autonomous systems. AI agents thrive on fast, fresh data streams – in fact, “the need for accurate, comprehensive, real-time data that can scale as needed cannot be overstated.” They depend on extensive knowledge bases plus up-to-the-moment information to make decisions on the fly. Organizations, therefore, must invest in a data backbone that delivers both ultra-low latency and massive scalability, so agentic AI has access to information even under heavy workloads.

This is where Aerospike’s real-time data platform provides a foundation for agentic AI. Aerospike’s high-performance database is engineered for sub-millisecond latency and extreme throughput, so AI agents ingest and retrieve large amounts of data with virtually no delay. In practice, Aerospike powers millisecond-level decisions on billions of events without compromising on performance or predictability, meaning your data layer won’t become a bottleneck as AI agents scale up. Equally important, the Aerospike platform offers integrated connectors for streaming data and compatibility with AI/ML frameworks. This lets your autonomous agents tap into live event streams and rich historical data through one unified, always-on repository, so they have the context they need in real time.

For organizations ready to take advantage of agentic AI, using Aerospike as the real-time data backbone is a natural next step. Aerospike’s proven technology helps your AI agents operate at full potential, with reliable access to the latest data at lightning speed. 

Ready to accelerate your agentic AI projects? Contact Aerospike or start a free trial to discover how the Aerospike real-time data platform can power your next-generation AI applications.