When I hear from customers about where AI is actually changing their day-to-day work, the conversation seems to land in the same place: the last mile of compliance that software could never quite close.
For decades, Avalara has helped organizations automate large portions of their tax and compliance workloads. At Avalara CRUSH 2026 in September, we unveiled Aviator, our new AI-powered agentic hub designed to close the remaining gap.
At Avalara, we’ve been developing AI agents for years, but Aviator represents an important step forward. Customers now have direct access to specialized agents that they can incorporate into their everyday work.
The Architecture Behind Aviator
Avalara’s technology infrastructure processes more than 54 billion tax calculations and files more than 7 million tax returns each year. Aviator is our new agent hub for tax and compliance that gives customers direct access to our AI agents. The hub provides access to our orchestrator agent called Avi, which interprets requests and then routes them to dozens of specialized agents that perform tasks related to reporting, tax return filing, and other components of tax compliance.
For example, a customer might ask Avi to generate a report or help complete a tax return. Avi will interpret the request and pass it on to an appropriate lead agent, which will coordinate with additional agents to complete subtasks.
Behind the scenes, Aviator draws on publicly available large language models, as well as our own proprietary models. The hub incorporates both retrieval-augmented generation (RAG) and graph technologies to support its AI capabilities, and it also features a gateway that automatically routes work to the AI model best suited to a given task. Rather than simply using the most advanced (and most expensive) model for every request, this gateway optimizes model selection based on both the work being performed and the associated cost.
Embedding Governance
Tax and compliance systems touch sensitive data and business-critical processes. That means automation without governance isn’t an option. One of our guiding principles is to distinguish between probabilistic tasks that are suited to AI and deterministic tasks that are not.
For instance, we still rely on deterministic systems for tax calculations that require response times under 10 milliseconds. Rather than the AI agents performing those calculations themselves (which could introduce latency or inaccuracies), our agents augment deterministic processes by helping customers with activities such as onboarding, configuration, and data mapping.
We have also implemented governance policies designed to protect information, and we never use customer data to train our own AI models. Agents can only access the data that they should be allowed to access, and they can only perform the tasks that they should be able to perform.
Together, these guardrails ensure that new AI capabilities do not compromise the reliability of our underlying tax and compliance systems and do not create additional risk around sensitive financial information.
Connecting Agents Across Systems
As agentic AI becomes more prevalent, interoperability is becoming increasingly important. Customers shouldn’t have to move between disconnected AI platforms or build separate integrations for every task they want to automate.
That’s why we’re pursuing a strategy that allows our agentic capabilities to operate within and across the third-party tools our customers already use. We’ve developed Model Context Protocol (MCP) servers across our portfolio, enabling our customers, partners, and even third-party developers to access Avalara capabilities. These same MCP servers also power the agents within the Aviator hub itself.
Our agents also support the Agent2Agent (A2A) protocol, allowing outside AI agents to communicate with them directly. For example, a customer might interact with an agent in an e-commerce application, which could then communicate with our agents to initiate a reporting or tax-filing task.
Expanding the Possibilities of Agentic Compliance
We continue to develop specialized agents that support a growing range of tasks for tax and compliance. And we expect that advances in AI models, context windows, and tool selection will change how these specialized agents are organized, potentially allowing individual agents to handle a broader range of responsibilities.
At the same time, we are looking beyond tax automation. Avalara already supports electronic invoicing in more than 60 countries, and we see an opportunity to automate more of this work with agentic AI.
Over time, we will continue to bring intelligence and automation to where tax and compliance work actually happens, letting our customers spend less time managing these complex processes and more time focusing on their business.
To learn more, visit avalara.com/aviator.
