The tax and accounting profession got a jolt of AI this week as two companies unveiled tools that promise to fundamentally change how financial data is analyzed and acted upon. Accordance, a startup backed by $13 million in seed/pre-seed funding led by Khosla Ventures and General Catalyst, publicly launched its “AI brain” for tax, audit, and accounting. At the same time, Sourcetable announced Superagents—an agentic AI layer that connects its spreadsheet interface to any database, API, or Model Context Protocol (MCP) server on the internet. Together, these releases mark a pivotal moment for agentic AI in professional services, especially for firms running Windows-based workflows.
Accordance positions itself as a frontier model company purpose-built for regulated financial workflows. Its multi-agent system has been trained on an extensive corpus of federal, state, and international tax codes, accounting standards, court precedents, and treaties. The result is what the company calls an “intelligent copilot” that can surface relevant statutes, draft research memos, and provide cited answers—all within a professional’s existing toolset. The $13 million funding round, first reported by Accounting Today, signals strong investor confidence in verticalized AI applications that marry deep domain expertise with agentic capabilities.
Sourcetable, meanwhile, is tackling a different but complementary pain point: the friction of connecting live business data to AI. Its Superagents allow users to reach any data source from a spreadsheet using natural language. If an application has an API or MCP server, Sourcetable’s AI chat acts as a connection wizard, guiding users through authentication and mapping without code. This means an accountant could pull real-time figures from ERPs, CRMs, and bank feeds directly into a familiar grid, then have an AI agent analyze them—all on a Windows laptop running Excel or a browser-based sheet.
Why these announcements matter now boils down to two converging trends: the maturation of agentic AI and the accounting industry’s pressing need for productivity gains amid a staffing shortage. Accordance’s brain, with its cited outputs, aims to slash research time from hours to minutes. Sourcetable’s connectors eliminate the drudgery of data wrangling. Both promise to amplify junior staff’s capabilities while letting senior professionals focus on judgment calls. For a Windows-centric firm, these tools integrate with existing Microsoft 365 workflows, making adoption a natural extension rather than a rip-and-replace.
Under the hood, both products lean heavily on the Model Context Protocol (MCP). Originally popularized in 2024, MCP standardizes how AI agents call external tools, read files, and share context. Accordance likely uses MCP to invoke its proprietary research tools and tax engines, while Sourcetable’s Superagents explicitly leverage MCP for universal connectivity. This protocol also underpins Microsoft’s own Copilot extensions, meaning these third-party agents can plug into the same ecosystem that powers Windows and Office.
But the rise of agentic AI in accounting isn’t without peril. Hallucinations—where an AI cites a non-existent case or miscalculates a deduction—carry real legal risk. Accordance claims performance improvements on its proprietary TaxBench, but independent validation is lacking. Sourcetable’s Superagents, for all their convenience, open new attack vectors: compromised connectors could exfiltrate sensitive client data, or prompt injection could trick an agent into unauthorized actions. Firms must demand immutable audit trails, least-privilege access, and contractual no-train guarantees. On a technical level, Windows shops should verify that connector configurations use ephemeral credentials and enforce role-based access consistent with Active Directory policies.
For firms ready to pilot, the advice is clear: start small. Run Accordance through a battery of known tax scenarios from prior filings and compare its citations. Test Sourcetable’s connectors with non-sensitive data first. Update engagement letters to disclose AI usage, and never let an AI output become the final word without a licensed human’s review. Training is equally critical—staff must learn how to prompt safely and spot AI errors, just as they once learned to use pivot tables.
The competitive landscape is heating up. Incumbents like Thomson Reuters and Wolters Kluwer are already embedding agentic features into their platforms, leveraging decades of curated professional content. Startups like Accordance must therefore differentiate through agility and deeper specialization. Sourcetable’s spreadsheet-centric approach, meanwhile, appeals to the millions of Windows users who live in Excel but crave live data connections—a direct challenge to Microsoft’s own Copilot in Excel, which currently lacks such open-ended agentic connectivity.
Looking ahead, standardization efforts around MCP will be crucial. Microsoft and major cloud providers have signaled support, which could make agentic AI as ubiquitous as ODBC once was for databases. Independent benchmarks like TaxEval or BizFinBench will separate marketing from reality, while professional bodies like the AICPA will likely issue AI guidelines. Windows-based firms that align early with these standards will gain a competitive edge.
In sum, the double launch of Accordance and Sourcetable Superagents demonstrates that agentic AI for tax and accounting is no longer a research project—it’s a deployable reality. For Windows professionals, the path forward is one of cautious optimism: leverage these tools to reclaim time and deepen insight, but do so with governance frameworks that turn “trust me” AI into defensible, auditable assistance. When the numbers have to be right, the best agentic brain still needs a human mind to sign off.