Tax solutions provider
"The most trusted AI breakthroughs are purpose-built," said Scott McFarlane, CEO and co-founder of Avalara, in a statement. "Just as Harvey AI transformed legal workflows and PathAI redefined diagnostics, Avalara is now setting the standard as the domain-specialized leader in agentic tax and compliance."
This is part of a larger concept that Avalara calls "Agentic Compliance," an idea that compliance solutions should not be based on bolted-on point products but AI agents that regularly and repeatedly integrate with business systems; observe data, workflows and user behavior; advise on actions, risk exposure and compliance obligations; and execute calculations, filings, validations and document classification. The idea is to shift further away from manual processes performed at certain points in time and into automated processes that are always on, reliably repeatable, and executed by AI agents. When asked just how independent these agents can get, Jayme Fishman, Avalara's chief strategy and product officer, said it varies based on what the user wants.
"What an agent really represents is this always on, always ready, 24/7, 365 capability that can be as busy as you want it to be, but it can also be inactive when it's not being utilized by whatever process you're trying to automate," he said.
The Returns AI, for example, can ingest transaction data, invoke Avalara's headless Returns APIs, apply the appropriate forms and jurisdictional logic, and, upon approval, file on behalf of the client. Fishman also said an agent can oversee compliance for e-commerce merchants. Overall, he said, there were already dozens of agents in the system and dozens more are currently in development.
Key to their effectiveness is the ability for these agents to talk to each other and collaborate. For example, say the aforementioned e-commerce agent needs a tax return for a specific user. That agent reaches out to what Fishman called their "orchestrator agent," Avi, and communicates that it wants to create a return. Avi then calls on the e-commerce agent to send it the required information, which is then passed on to the aforementioned returns agent, which eventually creates the return based on what the e-commerce agent said.
"These three agents are all working with each other to create the outcome," he said.
It is not just Avalara's agents that can talk to each other. They are also designed to collaborate with other agents within an enterprise, whether they originate from an ERP, POS or e-commerce system. This means that productivity tools, development environments, file storage systems and even user devices can connect directly to Avalara's agents to validate taxes, classify products, initiate registrations, or file jurisdictional forms. This is the foundation of Avalara's new operating model, which was described as "Have your agent call our agent."
Fishman described the agents as a "new sort of front door to all our applications." When familiar information goes through this front door, the agents can understand and execute previous processes related to this information without the need for human intervention, as "those configurations still live in our system and this is just gathering data. They will use not only their own capacities to do so but also access Model Context Protocol servers that support access to APIs and tools, private LLMs and proprietary domain-specific small language models within Avalara's ALFA control framework for AI. Underlying this is an active-active, hyper-scalable (Avalara says it can already scale its existing architecture to support every transaction in the world), multi-cloud deployment spanning multiple geographies and providers to produce an average processing time of about 15 milliseconds.
When encountering novel situations for which the agents do not have established procedures, or for key decisions that will affect the company's compliance footprint, they'll notify the human for review and approval, with Fishman noting that keeping a human in the loop remains a priority.
With this new model of agentic tax and compliance will also come new ways to interface with Avalara solutions. Instead of logging in or navigating screens, customers have the ability to just ask an AI agent to do the work, from tax calculations to certificate validations and more.
Agents have already been deployed throughout the product line. Fishman said that by the end of Q4, Avalara expects all of its products to sport autonomous agents.
"Compliance has always been complex and resource intensive, and Avalara has worked for years to make it easier for businesses of all sizes," said Fishman in a statement. "Now, we're delivering the first-ever digital compliance professionals, AI agents that advise and execute at global scale, making it even easier to handle the most onerous and time-consuming tasks—so businesses can move faster and focus on growth."
