Automated Accounting Got $175M. Governance Got Nothing.

The money moved fast. Between February 5 and February 24, 2026, two AI accounting startups absorbed $175 million in venture capital. Within the same window, Intuit locked in a multi-year partnership with Anthropic to wire Claude directly into QuickBooks, TurboTax, Credit Karma, and Mailchimp. By March 2, Santander and Mastercard had executed Europe’s first live AI agent payment inside a regulated banking framework. Eight days later, Mastercard unveiled a Virtual C-Suite product designed to hand small business owners an AI-powered CFO they never could have afforded to hire.

That is a staggering amount of real infrastructure built in six weeks. And yet: the governance frameworks these systems need to operate safely in regulated environments remain somewhere between incomplete and nonexistent. EY reports that more than 70 percent of banking firms are already running agentic AI in some capacity. The EU AI Act’s high-risk compliance deadline lands on August 2, 2026. The gap between deployment velocity and regulatory readiness is not narrowing. It is widening.

The firms that survive the next eighteen months will be the ones that treated governance as architecture, not paperwork.

The Q1 Funding Sprint

Start with the capital. Basis, an AI agent platform for accounting workflows, closed a $100 million Series B on February 24 at a $1.15 billion valuation. Accel and GV led the round. The investor list read like a roster of people who understood what was happening: Lloyd Blankfein, Adam D’Angelo, Amjad Masad, Clem Delangue. Roughly 30 percent of the Top 25 accounting firms already deploy Basis agents across tax, audit, and advisory functions. The platform handles document review, reconciliation, tax return preparation, audit testing, and workpaper generation -- these are automated accounting workflows with audit trails baked in, already running at scale.

Nineteen days earlier, Accrual had launched with $75 million from General Catalyst. The pitch was not incremental automation layered on top of legacy tax software. It was a ground-up rebuild of AI bookkeeping and tax preparation. Accrual’s agents function as preparers: they read client inputs, identify missing information, generate follow-up questions, and produce draft returns ready for professional review. The numbers were hard to argue with. Preparation time dropped by more than 85 percent. Review time fell by up to 60 percent. H&R Block and Armanino signed on as early adopters. John Karls at Armanino put it plainly: returns coming out of Accrual were already at manager-review quality. What used to take hours processed in minutes.

Then Intuit and Anthropic announced a multi-year agreement to build custom AI agents using Anthropic’s Model Context Protocol. MCP is the connective layer that lets AI models talk to external tools without losing context -- the same architecture that enables AgentPMT’s dynamic tool relevance, loading only the capabilities an agent needs for a given task rather than drowning it in every available integration. In Intuit’s case, MCP integrations connect Claude to QuickBooks, TurboTax, Credit Karma, and Mailchimp, allowing agents to handle AI financial reporting tasks: analyzing margin variances, connecting project timelines to cash flow forecasts, flagging billing gaps, generating invoices, and surfacing compliance deadline violations. Rollout begins spring 2026. Alex Balazs, Intuit’s CTO, framed the partnership around security, accuracy, and compliance. Paul Smith at Anthropic emphasized that agents would understand specific industry workflows and compliance requirements.

The subtext across all three announcements was identical: the era of AI as a feature bolted onto existing accounting software is over. These are standalone agent systems designed to operate autonomously within defined boundaries. And the accounting profession needs them badly. More than 300,000 accountants have left the field since 2020. Seventy-five percent of CPAs are approaching retirement age. Elizabeth Beastrom at Thomson Reuters predicted that firms will employ at least one virtual agent for every CPA on staff within five years. That is not a forecast about some distant future. That is a workforce planning statement about 2031.

The First Regulated AI Agent Payment

While the accounting world was digesting $175 million in new funding, something arguably more consequential happened on March 2. Santander and Mastercard completed the first live end-to-end payment executed by an AI agent within a regulated banking environment.

The transaction ran through Santander’s operational payments infrastructure using Mastercard’s Agent Pay platform, which had been introduced in 2025 as a framework for agentic commerce. The system allows AI agents to initiate and execute transactions on behalf of customers within predefined spending limits and authorization parameters. Matias Sanchez, Santander’s global head of cards and digital solutions, captured the stakes clearly: building trusted, scalable frameworks will be essential to unlocking the full potential of AI agents in commerce.

The timing was not accidental. The same day, the OCC released its proposed rulemaking to implement the GENIUS Act -- the first federal regulatory framework for payment stablecoins, passed by Congress in July 2025. The proposed rules create a new entity type, Permitted Payment Stablecoin Issuers, with 1:1 reserve backing requirements and OCC supervision. The comment period closes May 1, 2026. When autonomous agents can both initiate payments and settle in stablecoins, the infrastructure underneath those transactions needs to be regulated. The GENIUS Act rulemaking is that infrastructure.

Eight days later, Mastercard launched Virtual C-Suite, starting with a Virtual CFO for small and medium-sized businesses. The product does cash flow risk detection, benchmarking, anomaly detection, and supplier payment optimization through conversational interaction rather than dashboards. Mark Barnett, Mastercard’s global head of SME, described the shift: moving from reading a dashboard to having a dialogue with financial data. The tool will be distributed through financial institutions and accounting platforms later in 2026.

The pattern here matters more than any individual product. Agents are moving from analyzing data to moving money. This is the jump from advisory to fiduciary. And the infrastructure supporting autonomous financial transactions -- payment rails, permission engines, policy frameworks -- is where the real race is happening. European AI agent startups raised over one billion euros across 54 deals in 2025 alone, with fintech AI infrastructure companies like GoCardless, Airwallex, and Payrails leading the way. As Roy Asiedu at J.P. Morgan put it: what matters now is programmable rails, policy-driven permissions, and infrastructure capable of operating safely at scale. The parallel to how AgentPMT handles agent financial operations is direct -- its x402Direct protocol and just-in-time funding system were built precisely because autonomous agents that can spend money need permission boundaries and audit trails that exist at the infrastructure level, not as afterthoughts.

The Governance Gap

Here is where the story turns uncomfortable. All of this deployment is happening against a governance backdrop that is, at best, half-built.

EY’s Global Financial Services Regulatory Outlook for 2026, produced with MIT Technology Review Insights, found that more than 70 percent of banking firms are using agentic AI to some degree -- 16 percent with fully deployed solutions, 52 percent running pilots. The report’s central finding was blunt: rapid AI adoption is outpacing regulatory oversight, and there is a general lack of robust governance frameworks across the industry. EY recommends audit trails, data security controls, and policies to control unofficial employee AI use -- the basics of responsible banking automation. Standard governance hygiene. That it still needs to be recommended tells you how early we are.

Oliver Wyman’s analysis of agentic AI in compliance reaches a similar conclusion from a different angle. Their finding that 70 percent of manual compliance work can be automated with AI compliance tools is eye-catching, yet the more important point is what they recommend alongside it: disciplined rollout with quality gates and performance validation, strict boundaries for autonomous operations, robust traceability, and workforce transformation. They describe agentic AI as a semi-autonomous orchestrator that plans, acts, and coordinates compliance workflows with human oversight. The emphasis is on “with human oversight.” Remove that phrase and you have a liability engine.

Basware, which unveiled agentic AI capabilities for its invoice processing platform, landed on perhaps the most quotable articulation of the problem. CEO Jason Kurtz stated the company’s goal as 100 percent automated, 100 percent compliant, and 100 percent protected invoice processing. Then he added the line that should be pinned above every AI deployment dashboard in every finance department: “Autonomy without trust is just risk.”

Basware’s approach embeds what they call autonomy gates -- a central policy engine that applies business rules and compliance requirements before any autonomous action executes. Every decision gets a complete audit trail. A single governed execution path for all AI actions. This is governance-by-design, and it is the exception rather than the rule. Basware’s own FT Longitude survey found that 61 percent of finance leaders have deployed financial automation agents experimentally, but 25 percent admitted they have an unclear understanding of how to implement them in practice.

The EU AI Act high-risk deadline on August 2, 2026, applies directly to credit scoring, fraud detection, and anti-money laundering systems -- exactly the categories where financial AI agents and regtech AI solutions are being deployed fastest. Penalties for non-compliance reach 35 million euros or 7 percent of global turnover, whichever is higher. That is not a wrist slap. That is an existential threat to mid-market firms that automated first and governed later.

AgentPMT’s architecture reflects the same governance-by-design principle that separates the firms getting this right from those accumulating risk. Its audit system logs every agent action with full traceability. Its human-in-the-loop checkpoints ensure that high-stakes decisions -- financial transactions, compliance-sensitive outputs, production deployments -- require explicit human approval before execution. These are not features added to satisfy a checklist. They are load-bearing structural elements without which the system does not function.

The Firms That Will Survive This

The dividing line is becoming clear. On one side: firms deploying AI agents with governance embedded in the architecture from day one. On the other: firms that shipped agents into production and plan to figure out compliance later. The EU AI Act will make this distinction financially consequential in less than five months.

Basis represents one model of getting it right. Their platform combines large language models with rule-based controls, maintaining audit trails across every workflow. The agent operates autonomously within boundaries that are explicit and enforceable. Matt Harpe, the CEO, frames their focus as equipping accountants with the highest-performing, most accurate AI for accounting. Accuracy matters because in regulated environments, an agent that is fast but wrong is worse than no agent at all.

Basware’s autonomy gates represent another model. Rather than trusting agents and auditing after the fact, the system enforces compliance before action. Every autonomous decision passes through a central policy engine. The architecture does not permit unaudited autonomous behavior. This is the difference between monitoring and governance -- monitoring tells you what happened, governance determines what is allowed to happen.

The advisory-first shift is accelerating. Ninety-three percent of accounting firms already offer advisory services, according to Accounting Today. As agents take over preparation, review, and processing tasks, the human role shifts toward judgment, client relationships, and strategic interpretation. New positions are emerging: AI compliance officers, finance technologists, AI operations managers. Oliver Wyman explicitly recommends reskilling compliance professionals as AI oversight specialists focused on escalation and quality control. The firms that treat this workforce transformation as optional will find themselves unable to staff the governance functions that regulators are about to require.

LucaNet’s analysis of AI trends for CFOs in 2026 reinforces the point. Over 70 percent of CFOs now hold direct responsibility for data, analytics, and AI strategy. PwC emphasizes that CFOs must make AI outputs explainable and compliant to build stakeholder trust. The concept of shadow AI -- uncontrolled employee use of AI tools outside sanctioned workflows -- is a growing concern. Governance is not just about the agents you deploy intentionally. It is about the agents your employees deploy without telling you.

AgentPMT’s budget system and workflow builder address this directly. By defining spending limits, permission boundaries, and approval workflows at the platform level, organizations maintain control over what autonomous agents can do and how much they can spend. The mobile biometric approval system adds a physical verification layer for high-value decisions -- a human checkpoint that cannot be bypassed by software. These are the kinds of structural controls that regulators will increasingly require and that firms deploying agents without them will struggle to retrofit.

August 2 Is Not Far Away

The question hanging over all of this is straightforward: will governance catch up to deployment before the EU AI Act’s high-risk deadline forces the issue?

The honest answer is probably not -- at least not uniformly. The firms that began building compliance into their agent architectures in 2025 will be ready. Basis, Basware, and organizations using governance-by-design platforms like AgentPMT have a structural advantage because their audit trails, permission systems, and human checkpoints already exist. The 25 percent of finance leaders who told Basware’s survey they do not clearly understand how to implement AI agents in practice are running out of time to figure it out.

The accounting profession lost 300,000 people and cannot replace them through hiring alone. The capital is deployed. The agents are in production. The first AI-initiated payment in a regulated banking environment has already happened. There is no going back to the pre-agent world. The only question left is whether the rules governing that world will be written deliberately or discovered through enforcement actions after something goes wrong. Five months is not a long time to write a governance playbook. It is especially not a long time when the draft is still mostly blank pages.

---

Sources

Basis Series B announcement, February 24, 2026. Accel and GV led the round at a 1.15 billion dollar valuation.

Accrual launch announcement, February 5, 2026. General Catalyst led the 75 million dollar funding round.

Intuit and Anthropic multi-year partnership announcement, February 2026. Custom AI agents via Model Context Protocol for QuickBooks, TurboTax, Credit Karma, and Mailchimp.

Santander and Mastercard first regulated AI agent payment, March 2, 2026. Executed through Mastercard Agent Pay platform.

Mastercard Virtual C-Suite launch, March 10, 2026. Virtual CFO product for small and medium-sized businesses.

OCC proposed rulemaking implementing the GENIUS Act, March 2, 2026. Framework for Permitted Payment Stablecoin Issuers with 1:1 reserve requirements.

EY Global Financial Services Regulatory Outlook 2026, produced with MIT Technology Review Insights. Survey of banking firm agentic AI adoption.

Oliver Wyman analysis of agentic AI in compliance, 2026. Finding that 70 percent of manual compliance work can be automated.

Basware agentic AI platform capabilities, 2026. FT Longitude survey of finance leader AI deployment.

Accounting Today advisory services survey. 93 percent of accounting firms offering advisory services.

LucaNet AI trends for CFOs, 2026. Over 70 percent of CFOs hold direct responsibility for AI strategy.

Elizabeth Beastrom, Thomson Reuters, workforce prediction on virtual agents per CPA within five years.