Unlocking value from data: How AI agents conquered 2024

If 2023 was the year of chatbots and search powered by generative AI, 2024 was all about AI agents. The initiative started by Devin earlier this year has grown into a full-fledged phenomenon, giving businesses and individuals the ability to work at a variety of levels, from programming and development to personal tasks such as planning and booking holiday tickets. provided a way to transform

Among these wide-ranging applications, this year also saw the rise of data agents – AI-powered agents that handle different types of tasks across the data infrastructure stack. Some companies handle basic data integration tasks, while others handle downstream tasks such as analysis and management in the pipeline, making life simpler and easier for enterprise users.

The benefits are increased efficiency and cost savings, and many are wondering, “How will the landscape of data teams change in the next few years?”

Gen AI agent took over data task

Agent capabilities have been around for some time, allowing businesses to automate certain basic tasks, but the rise of generative AI has taken things to the next level entirely.

gen AI’s natural language processing and tooling capabilities allow agents to go beyond simple reasoning and answers to actually plan multi-step actions and independently interact with digital systems to complete the action. You can collaborate with other agents and people at the same time. You will also learn how to improve your performance over time.

Cognition AI’s Devin was the first major agent product to enable engineering operations at scale. Larger companies then began offering more targeted enterprise and individual agents leveraging their models.

In a conversation with VentureBeat earlier this year, Google Cloud’s Gerrit Kazmaier said that data professionals are facing challenges such as automating manual tasks for data teams, reducing cycle times for data pipelines and analytics, and simplifying data management. He said he heard from customers that they are constantly faced with: Essentially, the team was running out of ideas on how to create value from data, but not enough time to execute on those ideas.

To solve this problem, Kazmayer explained that Google has modernized its core data infrastructure product, BigQuery, with Gemini AI. The resulting agent capabilities not only give enterprises the ability to discover, cleanse, and prepare data for downstream applications, but also support pipeline management and analysis, freeing teams to focus on higher-value tasks. You will be able to do it.

Multiple companies, including fintech companies, are currently using Gemini’s agent capabilities in BigQuery juroIt leveraged Gemini’s ability to understand complex data structures to automate the query generation process. Japanese IT company Unelly Also, use BigQuery’s Gemini SQL generation capabilities to help data teams deliver insights faster.

But discovering, preparing, and supporting analytics is just the beginning. As the underlying models evolved, even fine-grained data operations pioneered by domain-specific startups became subject to deeper agent-driven automation.

for example, air bite And Fastn made headlines in the data integration category. The former launched an assistant that creates data connectors from links in API documentation in seconds. The latter, on the other hand, has powered its broader application development services with agents that use only natural language descriptions to generate enterprise-grade APIs for reading or writing information on any topic.

Meanwhile, San Francisco-based Altimate AI targeted a variety of data operations such as documentation, testing, and transformation with its new DataMates technology, which uses agent AI to derive context from the entire data stack. Several other startups, including Redbird and RapidCanvas, are also working in the same direction, claiming to offer AI agents that can handle up to 90% of the data tasks needed in AI and analytics pipelines.

Agents that enhance RAG etc.

Beyond broader data manipulation, agent functionality is also being explored in areas such as search augmentation generation (RAG) and downstream workflow automation. For example, the team behind the vector database Weaviate We recently discussed the idea of ​​Agent RAG. This allows AI agents to access a wide range of tools, including web searches, calculators, and software APIs (Slack/Gmail/CRM, etc.) to retrieve and validate data from multiple sources to improve the accuracy of answers.

Additionally, towards the end of the year, Snowflake Intelligence will arrive with the option to set up a data agent that can leverage not only business intelligence data stored in your Snowflake instance, but also structured and unstructured data across siled third-party tools. was provided to the company. As sales transactions in databases, documents in knowledge bases like SharePoint, and information in productivity tools like Slack, Salesforce, and Google Workspace.

This additional context allows the agent to surface relevant insights in response to natural language questions and take specific actions based on the generated insights. For example, a user can ask a data agent to fill out an editable form with the insights it surfaces and upload the file to Google Drive. You may also be asked to write to Snowflake tables and modify data as needed.

Much more to come

While we haven’t covered every data agent application seen or announced this year, one thing is clear: this technology is here to stay. As Gen AI models continue to evolve, AI agent adoption will take off in earnest, with most organizations, regardless of sector or size, choosing to delegate repetitive tasks to specialized agents. This is directly linked to efficiency.

As evidence, a recent survey of 1,100 technology executives found that cap gemini82% of respondents said they plan to integrate AI-based agents across their stack within the next three years, up from 10% today. More importantly, 70-75% of respondents said they would trust an AI agent to analyze and synthesize data on their behalf and handle tasks like code generation and iterative improvement. .

This agent-driven change also means a significant shift in how data teams function. Currently, agent results are not production-level. This means a human will need to take over at some point to fine-tune the work to the agent’s needs. But with a few more advances over the next few years, this gap will likely close, giving teams AI agents that are faster, more accurate, and less prone to the errors humans typically make.

This means that the role of data scientists and analysts we see today is likely to change, with users moving into the AI ​​monitoring domain (where they can monitor AI behavior) and the higher-value tasks that systems perform. It is possible. It can be difficult to run.