Streaming specialist Confluent unveils AI development suite

Platform expansion

As enterprises continue to invest heavily in AI development, many data management vendors are trying to meet the demands of their customers by providing environments within their platforms that enable developers to easily build and deploy AI tools trained with proprietary data.

Every software vendor must have an agentic AI story or risk being left behind. Fortunately for Confluent, real-time data can provide important context to agents, many of which operate on a real-time basis.

OpenAI's November 2022 launch of ChatGPT, which marked significant improvement in generative AI (GenAI) technology, sparked the current wave of AI interest. After initially focusing on building chatbots and other applications that enable workers to easily query and analyze data using natural language rather than code, AI development has evolved over the past year-and-a-half to focus on building agents.

Agentic AI tools, unlike chatbots that require human prompts before acting, are trained with contextual awareness and reasoning capabilities that enable them to autonomously carry out tasks to not only help workers be more informed but also make them more efficient.

Vendors such as Databricks and Snowflake introduced capabilities to help customers build agents last spring. More recently, Oracle, Teradata and MongoDB have all followed suit.

Confluent first unveiled Streaming Agents in August. Now, the vendor is adding more capabilities to provide customers with a comprehensive suite for developing agents in a move that was motivated by a mix of customer feedback and ongoing trends, according to Sean Falconer, Confluent's head of AI.

"Some of this was driven by customer engagement, such as our built-in anomaly detection," Falconer said. "Anomaly detection has existed in the batch processing world for a long time, but if you can make that real-time, suddenly that's way more powerful. … Then, if you look at what's going on in the world, we're going through an architectural shift in terms of the way we think about software engineering."

Because of the architectural shift toward AI, and because agents require fresh data to perform properly, Confluent Intelligence is a significant suite of tools for the vendor's users, according to Kevin Petrie, an analyst at BARC U.S.

"Confluent's new release addresses a proven market need for real-time, event-driven data from multiple sources," he said. "Most popular AI use cases depend on real-time user interactions or application processing."

Confluent Intelligence, built on the Confluent Cloud, includes the following:

• Real-Time Context Engine, a fully managed service in early access that features a Model Context Protocol (MCP) integration and streams structured data to provide agents and other applications with needed context.
• Streaming Agents, a development framework in open preview that is based on Apache Flink subproject FLIP-531 and includes Agent Definition to enable users to create agents with just a few lines of code and built-in observability and debugging.
• Native integration with Anthropic's Claude large language model, making it the default LLM in Streaming Agents while still supporting any model customers choose for their needs.
• Built-in machine learning functions in Flink SQL to simplify data science tasks such as forecasting and model inference.

While Confluent Intelligence should enable users to train agents with streaming data, the suite is narrowly focused on feeding agents the type of data in which the vendor specializes rather than the full breadth of data that agents require, according to Menninger.

"The entire agent ecosystem is still fragmented," he said. "No single vendor solves the entire problem yet. These enhancements help Confluent-based agents participate in a larger ecosystem. … To the extent Confluent is used to bring data together from disparate systems, enterprises can begin to coordinate agents, but it's still not going to solve the orchestration issue that arises as agents proliferate."

Petrie, meanwhile, noted that while Confluent Intelligence supports structured streaming data, the vendor did not mention support for unstructured data as well. With unstructured data such as text and images making up a large majority of all data, agents trained only on structured data will be missing context.

"Confluent appears to be focusing on structured event data, which makes sense because that is its heritage," Petrie said. "But a lot of business context comes from unstructured data. It's not clear that Confluent has tackled that requirement yet. If not, they should do so."

Beyond Confluent Intelligence, Confluent Private Cloud enables customers that prefer to keep data operations on premises with the same event streaming and AI development capabilities the vendor offers in the cloud.

Enterprises in highly regulated industries such as financial services and healthcare and other companies that deal with sensitive information often prefer to build and deploy data products and AI applications in secure, private environments rather than public ones.

Confluent Private Cloud, therefore, is a valuable addition for such companies, according to Menningner.

"Many cloud-based software application providers are taking the lessons they have learned in delivering SaaS applications and packaging those for on-premises deployments," he said. "For those enterprises that require on-premises deployments, Confluent Private Cloud will deliver similar productivity improvements."

The new Tableflow integrations, meanwhile, are similarly significant, according to Petrie.

Tableflow, launched in March, is a feature within Confluent Cloud that enables users to automatically convert streaming data to open table formats such as Delta Lake and Apache Iceberg. The data can then  be integrated with other data and accessed to inform real-time AI and analytics tools.

The Tableflow integrations with Delta Lake and the Databricks Unity Catalog eliminate extract, transform and load (ETL) pipelines and remove the need to manually integrate Confluent with the two platforms. A similar integration with Microsoft Azure is in early access, while there is already an integration with AWS.

"It's great to see that Confluent is integrating its events with popular catalogs," Petrie said. "The more enterprises can centralize metadata for all their analytics and AI inputs on a real-time or near real-time basis, the better."

Looking ahead

As Confluent plots out its product development, providing agents with context is a focal point, according to Falconer.

In 2023, vector databases emerged as a means of discovering the data needed to develop GenAI tools. In 2024 and 2025, agents emerged as the dominant trend in AI development.

"Next year is going to be all about context," Falconer said. "You can't make any of these agentic systems go unless you have the context to steer models in ways that are healthy for the task you're trying to perform. Our focus is on how to serve that context to our customers with the Real-Time Context Engine. We'll be paying very close attention to the feedback we get from our design partners."

One way Confluent could better serve data to agents is by adding support for Agent2Agent Protocol, according to Menninger. While MCP is valuable, it only addresses some of what agents need by providing a framework for connecting agents to data sources. A2A extends that with a framework for connecting agents to each other.

"What I'd like to see them do next [is] A2A together with Kafka," Menninger said. "A2A solves a different problem than MCP and implementing point-to-point communications between agents [without A2A] would be difficult."

Eric Avidon is a senior news writer for Informa TechTarget and a journalist with more than 25 years of experience. He covers analytics and data management.