Elastic unveils tools to boost reliability of AI agents

Elastic has launched new tools for building and running AI agents, with a focus on getting those agents to retrieve and use information held across fragmented corporate data systems.

The company said its Agent Builder product is now generally available, while Workflows has entered technical preview. Elastic positions the releases as a response to persistent problems organisations report when they try to move beyond basic conversational AI and deploy agents that complete tasks inside business processes.

Elastic sells search and data management software used by enterprises for observability, security, and site and application search. The company has also branded itself as "the Search AI company". It has increasingly tied its product messaging to generative AI over the past year, including the use of retrieval methods that pull relevant internal data into AI prompts.

Agent reliability

Interest in AI agents has grown in Australia as executives look for systems that can perform actions rather than answer questions. The pitch often includes automating customer service, finance and back-office workflows. In practice, companies report that agents can fail when they cannot access the right information, or when they generate incorrect outputs and actions.

One common issue is the spread of relevant data across multiple systems. Customer data may sit in a CRM. Order and inventory information may live in a separate logistics or ERP database. Policy documents may sit in shared drives or PDFs. Staff context may sit in email or collaboration tools. These systems may not connect cleanly, and they may use different identifiers, formats and access controls.

Elastic's announcement centres on that gap between AI models and enterprise data. The company argues that the limiting factor for agents is often search and data access rather than model performance.

New products

Agent Builder focuses on creating AI agents that can search across different data sources. Elastic said it can work with both structured sources such as databases and unstructured sources such as PDFs and emails. The company also highlighted security controls in its description of the product, reflecting the access challenges that emerge when an agent needs to retrieve and act on sensitive information.

Workflows sits alongside Agent Builder as an automation layer. Elastic describes it as rules-based. It maps steps for an agent to follow when completing tasks. The company said this approach can include staged actions such as requiring human review before an update, or running triage steps before an operational or security response proceeds.

These products place Elastic in a crowded market where vendors promise agent orchestration, connectors, and guardrails. Large cloud providers, enterprise software platforms and specialist startups have all introduced similar concepts in recent months. Elastic's approach puts its search technology at the centre of agent design, particularly where organisations have large volumes of text and semi-structured information that remains hard to use in AI systems.

Enterprise integration

Elastic also framed the tools as model-agnostic. The company said it built the products to work with the AI models organisations already use. That stance reflects a broader trend in enterprise procurement, where many companies try to avoid locking their AI programmes to a single model provider. It also reflects operational reality, since model choices often vary across teams and use cases.

The releases also address a deployment challenge. Many enterprises report that AI pilots stall when they meet production requirements such as audit trails, repeatable behaviour, and defined approval steps. Workflows puts a process structure around agent actions. That can increase the amount of work required to implement an agent, but it can also make its behaviour easier to test and monitor.

Elastic has also pointed to research and executive sentiment that suggests a widening gap between AI ambition and operational results. It cited an MIT study in its materials, including a claim that nine out of ten agent projects fail. It also referred to a KPMG survey that found AI ranked as a top concern for Australian executives.

The company's messaging targets boards and senior executives who want AI systems that complete tasks such as booking meetings, processing refunds, and generating reports. Those use cases usually require access to multiple internal systems and consistent business rules. They also create risks if an agent makes an incorrect change or misinterprets a policy.

Elastic did not provide local pricing for the new products in its materials. It also did not outline which third-party applications it supports out of the box, or the breadth of connectors available in ANZ. Those details often determine how quickly enterprises can use an agent product across real workflows, rather than in controlled demonstrations.

As Australian organisations move from AI experimentation into operational roll-outs, vendors are increasingly competing on data access, governance and repeatable execution. Elastic said Agent Builder is now generally available, while Workflows remains in technical preview.