Inside Avalara Agentic Tax and Compliance™ engineering

Artificial intelligence demos are easy. Building reliable systems with AI is much harder — especially in regulated industries like tax compliance.

For businesses managing tax filings, payments, and reporting across jurisdictions, there’s little room for ambiguity. Every action must be traceable. Every outcome must be auditable. And every workflow must produce consistent results.

That challenge is exactly what shaped the architecture behind Agentic Tax and Compliance™ systems at Avalara. 

Rather than replacing compliance systems with generative AI, Avalara engineers designed an approach where AI assists decision-making while deterministic systems remain the system of record. This balance helps organizations benefit from intelligent automation without sacrificing control, auditability, or trust.

In this article, we’ll take a look under the hood of that architecture and explore how it helps partners and developers embed tax compliance directly into their workflows.

By the end of this article, you’ll understand:

• Why tax compliance is fundamentally a systems and orchestration challenge

• 
  How agentic AI can support workflows without replacing deterministic platforms

• 
  What model context protocol (MCP) tools are and why they simplify integrations

• 
  How agent-to-agent automation is shaping the next generation of compliance systems

• 
  Why auditability and deterministic state management are essential in regulated environments

Avalara is introducing a new engineering pattern that blends AI-driven assistance with deterministic compliance platforms.

At the center of this approach is an orchestration layer powered by intelligent agents that help guide workflows like:

• Data ingestion
• Validation and normalization
• Variance detection
• Return preparation and filing
• Payment and notice management

In simple terms, an agent is a software system capable of performing tasks on behalf of a user or another system. These agents can interpret context, make recommendations, and interact with APIs or tools to complete tasks.

But unlike fully autonomous AI systems, the Avalara model is designed around a key principle:

AI assists. Humans approve. The platform records.

This separation helps ensure that AI can improve efficiency without introducing guesswork into regulated processes.

When businesses think about tax compliance, they often imagine a single workflow: prepare a return and file it.

In reality, compliance at scale involves coordinating many moving parts.

For example, tax data often comes from multiple sources:

• Enterprise Resource Planning (ERP) exports
• Ecommerce marketplace reports
• Avalara AvaTax transactions
• Accounting adjustments
• Manual uploads or spreadsheets 

At the same time, every jurisdiction has its own rules, deadlines, forms, and payment processes.

The result is a complex operational challenge involving data coordination, workflow orchestration, and regulatory oversight.

Even when companies automate tax calculation, filing often becomes a bottleneck because the integration layer cannot easily handle the complexity of multiple systems.

The engineering approach at Avalara treats compliance as an intelligent execution layer that connects systems together while maintaining strict governance and traceability.

A key architectural idea behind Agentic Tax and Compliance™ systems is separating intelligent assistance from deterministic execution.

Deterministic systems (the core)

Deterministic systems produce the same result every time given the same inputs. This predictability is essential in compliance systems because regulators require predictable outcomes and clear audit trails.

Avalara’s core compliance platforms handle:

• Filing workflows
• Payment processing
• Form generation
• Submission tracking
• Immutable transaction logs

These systems act as the official system of record.

Intelligent systems (the edge)

AI systems operate at the “edges” of the workflow — where judgment and interaction help.

Examples include:

• Helping partners onboard new customers
• Detecting anomalies in tax data
• Explaining compliance workflows in plain language
• Assisting developers integrating tax capabilities

This hybrid approach allows businesses to gain productivity benefits from AI while maintaining strict compliance controls.

One of the most important components of this architecture is the model context protocol (MCP).

MCP provides a standardized way for AI systems and applications to interact with services through structured tools rather than raw API endpoints.

For partners embedding Avalara capabilities, using structured tools matters because it simplifies integration.

Instead of managing complex workflows themselves, partners can call tools such as:

• Discover onboarding requirements
• Trigger data ingestion
• Run variance analysis
• Initiate filings
• Retrieve status updates
• Manage tax notices

Each tool maps directly to deterministic platform workflows behind the scenes.

This approach reduces the need for developers to recreate compliance logic while still giving them flexibility to build native experiences.

Avalara designed its architecture to support multiple integration approaches depending on a partner’s needs.

1. GraphQL APIs — maximum control
 

Partners who want full control over user interfaces and workflows can integrate directly with APIs.

This option provides flexibility but requires partners to manage more orchestration logic themselves. 

2. MCP tools — faster development

The MCP server exposes ready-to-use tools aligned with compliance workflows.

Developers can quickly embed compliance functionality without designing their own orchestration logic.

3. Agent-to-agent workflows — automation at scale

In the most automated model, a partner’s AI agent communicates directly with the Avalara agent through a secure protocol,  enabling systems to coordinate compliance workflows automatically while maintaining platform governance.

Interestingly, the same agentic architecture used in the product is also used inside Avalara’s engineering organization.

Internal engineering agents help developers:

• Analyze engineering tickets
• Generate technical explanations 
• Build test scenarios
• Draft code changes for review

These agents work with deterministic systems like ticket trackers, version control platforms, and CI pipelines.

Just like customer-facing workflows, human review remains a required step before changes are finalized. 

This approach helps teams accelerate development while maintaining software quality and governance.

Although the example here focuses on tax compliance, the same architecture can apply to many regulated domains.

Organizations building agentic systems should consider several principles:

1. Separate intelligence from determinism 
AI can assist decisions, but regulated platforms must remain deterministic.

2. Standardize integration interfaces 
Typed tools or agent-based interfaces simplify development.

3. Maintain complete auditability 
Every transformation and action should be logged.

4. Keep humans in critical decision loops 
Approval steps preserve governance and trust.

5. Design platforms for partners 
Thin integration layers and shared infrastructure improve scalability.

For partners, developers, and technology providers, the next step is exploring how intelligent orchestration can simplify tax compliance workflows.

Organizations embedding Avalara solutions can:

• Use prebuilt integrations or APIs to connect existing systems

   

• Explore MCP-based tool integrations for faster development

   

• Build agent-enabled workflows that integrate compliance into everyday operations

Avalara currently supports more than 1,400 signed partner integrations across ERP, ecommerce, and financial systems, helping businesses automate compliance directly within their existing platforms. 

Ready to connect? Contact us to learn more about Agentic Tax and Compliance™.