Last Accounting launched this year with a proposition that would have seemed absurd a decade ago: an AI agent named Björn now does the bookkeeping, payroll, and monthly close for dozens of Finnish companies, while certified accountants sign off on whatever Björn produces. The Helsinki startup, operating formally as Bloom Systems Oy and backed by Y Combinator's Summer 2026 batch, claims it has processed more than $80 million in transaction value through a proprietary general ledger as of late July.
The firm's four founders describe themselves as an accounting firm on the outside and an AI research company on the inside. Co-founder Sami Laakkonen wrote in a July 19 blog post that Last Accounting built its own ledger specifically so agents could read and write against live books through a Model Context Protocol server and REST API. The design choice matters. It allows AI tools to operate accounting workflows under scoped permissions with full audit trails, a technical departure from legacy software where bookkeepers click through forms one field at a time. Clients don't log into dashboards. They message Björn through Slack, Teams, or WhatsApp. The company calls this "no UX at all."
Whether that pitch scales beyond one city and a few dozen clients is unproven. What's less debatable is the timing. Last Accounting arrived in a year when professional services firms and enterprise finance departments moved sharply toward agentic AI, with production deployments replacing pilot projects and incumbents racing startups to automate the monthly close.
Accounting's Talent Drain Meets Production-Ready Models
Three forces converged to make 2026 the breakout year for AI-native accounting, though none of them emerged overnight.
First, the profession hemorrhaged people. More than 300,000 accountants have left in recent years, according to Randstad research published in January. The American Institute of CPAs reported some signs of recovery in a September 1 trends update, noting 12 percent year-over-year growth in accounting enrollment during the 2024-25 academic year. But incoming classes remain far below the exit wave, and firms are scrambling.
Second, the technology crossed a threshold. Stanford's AI Index 2026 cited improving agent benchmarks between 2024 and 2025, with OSWorld accuracy scores advancing through July 2026, according to Last Accounting's blog post referencing the index. Third-party audits of AI accuracy remain scarce. Most precision claims come from vendors themselves. Digits, for instance, reported a 97.8 percent transaction categorization rate in 2026, though independent verification is harder to find.
Third, the market opportunity widened fast. ResearchAndMarkets valued AI in accounting at $10.8 billion in 2026 and projected growth to $88.1 billion by 2034, implying a compound annual growth rate near 30 percent. Grand View Research pegged the 2025 market at $6.71 billion with similar growth expectations through 2033. Gartner forecasted worldwide AI spending of $2.59 trillion in 2026, up 47 percent year-over-year, in a May 19 release.
A KPMG US survey released May 11 found that 93 percent of US companies plan to deploy or scale AI in finance within 18 months. Half intend to orchestrate multi-agent systems. Roughly 74 percent reported AI returns on investment that met or exceeded expectations. Weekly AI usage among US tax professionals nearly doubled year-over-year, rising to 60 percent in 2026 from 33 percent in 2025, according to a June 8 survey of more than 1,000 practitioners by Blue J and CPA.com.
Thomson Reuters Institute's February 2026 report on AI in professional services found that enterprise-wide usage nearly doubled in 2025 versus the prior year across legal, tax, accounting, risk, and government sectors. Gartner reported June 8 that 84 percent of finance organizations have implemented or plan to implement AI, though only 7 percent reported high impact. That gap suggests structured roadmaps remain missing at most firms.
The Big Four Move Faster Than Expected
The incumbent response came quicker than many observers anticipated. EY rolled out multi-agent capabilities across its global audit platform EY Canvas in April 2026, according to a September report by the Institute of Chartered Accountants in England and Wales. PwC published a vision document in early September describing an "agentic office of the CFO" as the near-term destination for finance functions. Deloitte's 2026 Finance Trends report called the march toward AI agents "a relatively new frontier," but noted that firms are already piloting agentic systems for reconciliations, variance analysis, and continuous monitoring.
"The ultimate goal is not just automation, it's elevation," Thomas Mackenzie, KPMG US and Global Audit Chief Digital Officer, said in the May KPMG release. "This is the core of our 'human-led, agent-operated' vision."
Enterprise software vendors followed suit. Oracle introduced "Fusion Agentic Applications" on March 24, 2026, with embedded agents that can "make and execute decisions within business processes," according to the company's announcement. Oracle's blog confirmed June 22 that four finance agents covering the general ledger, expenses, payables, and payments reached general availability in the 26B release.
Sage announced a Finance Intelligence Agent rollout April 28, with natural-language Q&A, agentic automation, and full audit trails across finance, HR, and operations workflows. "In finance, 'almost right' isn't good enough," Aaron Harris, Sage CTO, said in the release. "AI must be accurate, auditable and reliable in real workflows, not just impressive in a demo."
Intuit expanded its "Intuit Assist" platform September 8, embedding generative AI agents across accounting, payments, payroll, sales tax, and business tax products. "We're creating a future where we do the hard work for small businesses and consumers to fuel their financial success," Sasan Goodarzi, Intuit CEO, said in the announcement. QuickBooks help documentation updated through August describes agent features and availability across subscription tiers.
A Crowded Field of Startups and One Notable Collapse
Last Accounting sits in a crowded field of AI-native startups, all racing to automate the monthly close. Y Combinator's Winter and Summer 2026 batches included at least three agent-native firms beyond Last Accounting. Rational markets itself as "the first zero-human accounting firm," building AI employees for existing practices, according to its YC profile. Balance offers "full-stack AI accounting" for small and midsize businesses, with agents automating monthly close and human sign-off, per its YC listing. All three launched within months of each other.
Rillet raised $100 million at a $1 billion valuation in a 48-hour Series C round announced August 21, according to TechCrunch. The company claims 600 customers and an alliance with EY. "Rillet was built for AI agents, not humans, letting humans work alongside the AI agents on corporate bookkeeping," TechCrunch reported. Rillet did not disclose the round's lead investor.
Ramp launched an "Accounting Agent" February 12 to automate bookkeeping and enable real-time close, according to a PR Newswire release. The company introduced "Ramp Stack" for accounting firms June 3. Puzzle went generally available July 9 with "AI Close" agents designed for human-in-the-loop month-end workflows at accounting practices, per a PR Newswire announcement. Vic.ai, focusing on enterprise accounts payable, published private-equity customer case studies through March and August.
A design pattern emerged: startups emphasizing "interfaceable" ledgers built so agents can execute against live books. Zenday, Equated, Paprel, and Finaloop all market agent-native or agent-accessible accounting systems launched or updated in 2026. Money Forward Cloud Accounting in Japan launched a remote MCP server March 26, enabling AI tools to operate accounting workflows through the Model Context Protocol, according to the company's press release and developer documentation updated July 14. Anthropic open-sourced the MCP specification November 25, 2024, with adoption accelerating through 2025 and 2026.
Not every bet paid off. Botkeeper, which raised $100 million to automate bookkeeping, faced significant challenges and retooled its product offering in early 2026, triggering Reddit discussions in February and a product reboot reflected in blog updates from June through September. The failure served as a reminder that automation alone doesn't guarantee a viable business model.
Regulation Will Decide How Far This Goes
Regulatory frameworks will shape how far and how fast agent-native accounting spreads. The European Union's AI Act imposes transparency obligations enforceable from August 2, 2026, with high-risk system requirements phased in by December 2, 2027, following extensions confirmed July 27, according to the European Commission's AI Act service desk. Finance workflows involving automated journal entries, reconciliations, or filings may fall under high-risk categories, requiring explainability, logging, and human oversight.
The US Internal Revenue Service released "Introductory Guidelines for Responsible AI Use in Federal Tax Practice" June 24, emphasizing that preparers must use secure, enterprise-approved AI tools and strictly handle client data under IRC sections 7216 and 6713. The Committee of Sponsoring Organizations of the Treadway Commission published "Achieving Effective Internal Control Over Generative AI" February 23, outlining control practices for companies using AI in financial reporting. The Public Company Accounting Oversight Board issued staff spotlights on generative AI in audits during 2024 and 2025, signaling ongoing scrutiny as Big Four firms scale agentic tools.
The pattern emerging across vendors and firms is "agent plus human sign-off." What Last Accounting, Puzzle, Ramp, and Intuit all emphasize as the default safety model. Agents execute bookkeeping, categorization, reconciliations, and draft filings. Certified accountants review and approve. KPMG's May survey found that 50 percent of finance organizations plan to orchestrate multi-agent systems within 18 months, and third-party AI assurance is becoming a prerequisite for innovation rather than a box-checking exercise, according to the firm.
McKinsey argued July 24 that agentic AI in financial planning and analysis will push organizations toward continuous monitoring and smaller, business-facing finance teams supported by centralized analytics and AI workflows. Gartner advised June 8 that CFOs need structured finance AI roadmaps, noting the gap between intent and impact. ACCA Global's May 2026 Talent Trends report found younger professionals more likely to adopt AI at work, with charts showing rising agent use in finance roles.
Thomson Reuters Institute's 2026 report concluded that professional-grade AI tools are now "non-negotiable" for tax and accounting staff, with firms differentiating not by whether they deploy AI but by how they redeploy freed capacity.
Last Accounting's bet is that the firm of the future looks more like an AI lab with CPAs on the roster than a CPA firm with software licenses. The company remains subscale: four employees, one local market, self-reported metrics. But its timing aligns with a sector crossing from pilot to production. Whether Björn scales beyond Helsinki or becomes a footnote in a consolidation wave is an open question. The direction, though, is no longer in doubt.
