Basis AI and the Making of a $1.15 Billion Unicorn

Basis AI builds autonomous agents that complete accounting workflows end to end. The agents connect to a firm’s ledgers and ERPs, ingest client documents, execute multi-step processes over several hours, and return finished deliverables ready for human review. The New York-based startup, founded in 2023 by Matthew Harpe and Mitchell Troyanovsky, reached a $1.15 billion valuation in February 2026 after closing a $100 million Series B led by Accel. GV (formerly Google Ventures), former Goldman Sachs chief executive Lloyd Blankfein, and existing backer Khosla Ventures also participated.

For context, that valuation places a two-year-old private company within striking distance of BlackLine, the publicly traded accounting automation platform currently worth roughly $1.7 billion on $717 million in trailing revenue. Basis AI has disclosed neither revenue nor ARR.

Total funding now stands at $138 million. The company’s earlier rounds included a $3.6 million seed led by Better Tomorrow Ventures and a $34 million Series A led by Khosla. That trajectory, from seed to unicorn in roughly two years, puts Basis AI among the fastest-growing enterprise AI startups in any vertical. It also puts the company under a level of scrutiny that most accounting technology vendors never face.

Harpe spent his early career at Boston Consulting Group, where he saw first-hand how rarely organisations leveraged the granular accounting data sitting inside their own systems. Troyanovsky brought over a decade of experience in AI and machine learning, spanning NLP, computer vision, and deep learning. The idea for Basis AI came from a shared conviction that accounting was one of the last major professional domains where AI could replace process, not just augment it.

The product spans three practice areas: client accounting services, tax, and audit. In CAS, agents handle transaction processing, reconciliations, journal entries, and month-end close workflows. In tax, the company claims the marquee technical milestone of the first AI agent to autonomously complete an end-to-end 1065 partnership tax return, a filing that involves partner allocations, capital account maintenance, and multi-jurisdiction considerations. That claim, made in the Series B announcement, has not been independently verified. In audit, agents perform testing procedures and workpaper generation, though the depth of autonomy in audit workflows remains unclear. What is clear is that the ambition is total coverage. Basis AI does not want a slice of the workflow. It wants the whole engagement.

The core of the product is what the company calls “long-horizon” agents. Unlike tools that respond to a prompt and return an answer, these agents operate continuously in the background, coordinating tasks across accounting processes for hours before delivering completed work. Think of it less like asking a question and more like handing a file to a junior accountant who works through the night and leaves the finished output on your desk by morning. The architecture uses a multi-agent model. A supervising agent coordinates each workflow by routing tasks to specialised sub-agents based on complexity, latency requirements, and input type. For speed-critical interactions, such as mid-review clarifications, the platform deploys lighter models. For complex scenarios involving unusual transaction patterns or ambiguous classifications, the system leans on deeper reasoning models built on OpenAI’s frontier stack. Each decision is surfaced with its supporting logic, data sources, and confidence level, so the reviewing accountant can see how the agent reached its conclusions.

The company describes the resulting shift as turning accountants from “doers into reviewers.” The agent does the preparation, the human applies judgement. In practice, the reliability of that handoff is what separates a useful product from an expensive liability. If the agent’s work is good enough to review efficiently, the model works. If the accountant has to redo half of it, the tool is just overhead with a venture-backed price tag.

Basis AI currently serves approximately 7 of the top 25 US accounting firms, a figure reported by Reuters. Named clients include Wiss, UHY, Boulay, Clark Nuber, MarksNelson, and Pinion. The company claims efficiency gains of 20% to 50% across practices. Those figures are self-reported and should be treated accordingly. No third-party benchmarks or audited performance data have been published.

“Basis is the most powerful accounting AI we have seen. It has been truly trajectory-defining for Wiss: we use it across the firm, empowering our teams to focus on higher-value work while delivering exceptional experiences for our people and clients.” Wiss & Company

“I expect Basis to deliver the same step-change in accounting that software engineering saw in 2025.” Vinod Khosla, Khosla Ventures

The comparison to AI coding tools is instructive but imperfect. Software tolerates iteration and rollback in ways that tax filings and audit workpapers do not. A code suggestion that is 90% correct can be edited. A tax return that is 90% correct can trigger an IRS notice. The margin for error in accounting is thinner, the consequences more immediate, and the regulatory exposure more personal. Partners sign off on work. Their names are on it.

The platform integrates with QuickBooks, Xero, and major ERP systems. This matters for adoption. Accounting firms are notoriously resistant to replacing existing infrastructure, and the company has positioned its agents to operate within the tools firms already use rather than requiring migration to a new stack. That approach lowers the implementation barrier, though it raises questions about depth of integration and the reliability of agent actions inside third-party systems that were not designed for autonomous operation.

The go-to-market strategy has been deliberately narrow. Rather than chasing volume among small practices, Basis AI has focused on large and mid-tier firms where complex, high-volume engagements offer the clearest return on agent deployment. Access is managed through a waitlist and direct sales. Pricing has not been disclosed publicly, which is typical for enterprise platforms selling to firms where deal sizes vary by scope and practice area. The bet is that once agents prove reliable on one practice area, firms will extend them across others. Internally, the company is reinforcing that bet through a dedicated team called Atlas, tasked with deploying autonomous agents across its own engineering, sales, and talent functions.

Competition is intensifying. Accrual, backed by General Catalyst with $75 million, emerged from stealth in early 2026 targeting the same market. Pilot announced what it calls a fully autonomous “AI Accountant” for small businesses. Adopt AI, FloQast, Docyt, and a growing roster of agent-native platforms are chasing overlapping segments of the workflow. The Big Four have their own internal programmes, with EY alone investing $1.4 billion in AI initiatives. The advantage for the company, for now, is its head start with top-tier firms and its focus on complex engagements rather than basic bookkeeping automation. But head starts erode quickly when capital flows this freely into a category.

There are legitimate questions about where the ceiling sits. Tax preparation and reconciliation are structured enough to be tractable for current AI systems. The inputs are defined, the rules are codified, and the outputs can be validated against known standards. Audit is harder. It requires professional judgement, scepticism, and an understanding of risk that goes beyond pattern matching. It demands the ability to identify what is missing, not just process what is present. No amount of reasoning capability in a foundation model closes that gap entirely.

For firm leaders evaluating Basis AI, the question is not whether autonomous agents belong in accounting. It is whether these agents are accurate enough, transparent enough, and controllable enough to trust with client engagements today.

At $1.15 billion, investors have answered that question for themselves. The profession is still deciding.