Maxima ships Max: an AI agent that preps the close

Maxima, a San Francisco startup that has processed more than $400 billion in accounting transaction volume since 2024, is the latest vendor to put a named AI agent on the chart of accounts. The agent, Max, prepares journal entries, reconciles accounts, and routes every output to an accountant for review. Nothing posts to the general ledger without human approval. That guardrail is the whole pitch.

The launch lands in a market that has spent the last two years flooding enterprise finance with generative AI tools, most of them free of meaningful controls. The pitch from incumbents and startups alike has been the same: post a chatbot in front of the ERP and call it automation. The hard part of accounting is not the question. It is the answer having to survive a regulator, an auditor, and a board. Maxima is leaning into the second problem.

"Accounting is one of the most meticulous and high-risk industries in our entire economy," Yogi Goel, Maxima's co-founder and CEO, said in the announcement. "AI labs and vibe-coded apps give a mirage of accuracy in their race for accessible automation. But lack of controls and 90% accuracy isn't good enough when the 10% that's wrong can lead to adverse audit opinion, regulatory fines, or delisting."

The launch is also a data point on a quieter shift. Enterprise accounting has been the most under-automated corporate function for a decade. The new generation of agents is not trying to replace the controller. It is trying to absorb the preparation work, the journal entries, the reconciliations, the flux analysis, that has historically eaten the calendar of every senior accountant on the team. Max is the cleanest version of that pattern so far.

What the Maxima Max agent actually does

Max ships with six domain-specific skill sets that map to the core accounting lifecycle. Each skill is purpose-built for a discrete slice of the close, and each one is wired to Maxima's broader platform, which sits on top of existing ERPs without requiring a rip-and-replace.

Cash accounting is the entry point. Max classifies bank statements, reconciles cash to the general ledger, and prepares the cash journal entries an accountant would otherwise build by hand at the end of every period. Accruals and accounts payable is the next layer, where Max calculates estimation-based accruals, processes reversing entries, and prepares AP journal entries against invoices and recurring expense streams.

Payroll is the third skill. Max prepares payroll journal entries, reconciles payroll subledgers to the general ledger, and calculates compensation accruals that the controller's team would otherwise reconcile across multiple systems. Intercompany and allocations is the fourth, and it is the part of the close that most enterprise accounting teams still do by hand. Max executes intercompany elimination, builds consolidation support, and applies the allocation methodology a controller has already approved for the entity structure.

Commissions and equity is the fifth. Max capitalizes commissions, tracks equity compensation, and calculates fixed asset depreciation across the asset register. Revenue is the sixth, and it handles revenue recognition, processes true-ups and reversals, and routes the work back into the ERP once an accountant approves it.

Every output Max prepares carries a complete audit trail, step-by-step proof of work, validation checks, and segregated human-in-the-loop approvals. That is the difference between Max and a free chatbot. Every step is auditable. Every posting is gated. The agent does not get to decide what lands in the books.

Customer numbers from the Maxima launch

Maxima is publishing customer-side numbers with the launch. Early customers using Max are reporting up to 70% faster close cycles, as much as 80% less prep time on recurring workflows, and more than 60 hours saved per person per month. The named deployments are Rippling, Miro, Zendesk, Scale AI, and Bilt Rewards. None of these are small finance teams. Scale AI alone processes enough transaction volume to make Maxima's $400 billion figure credible.

Josh Waldron, Chief Accounting Officer and SVP of Finance at Scale AI, gave a written statement for the launch that reads as if it was written by a controller, not a marketing team. "We get Maxima's agentic system of work, and now Max, an always-on teammate that prepares accounting work across our finance stack while operating within our existing SOX requirements, controls, and approval workflows," Waldron said. "The result is more automation, full accuracy, complete auditability, and greater confidence in every output."

The Scale AI quote is the one that matters for the buyer audience, because it lands on the actual point of friction in enterprise finance. The hard part is not getting an AI to write a journal entry. The hard part is getting the AI to write a journal entry that survives a SOX walkthrough. Maxima is putting the control surface in front of the customer, not behind it. Whether that holds at audit is the question that will determine whether Maxima's $400 billion run rate converts into the $4 billion that the company will need to grow into over the next two years.

How the Max agent fits the existing stack

Maxima does not require teams to replace their existing systems. The platform sits on top of the customer's ERP, pulls data from banks, payroll platforms, billing tools, and data warehouses, and pushes audit-ready outputs back into the ERP once an accountant has reviewed and approved them. Max is the agent that operates within that system of work, handling the preparation that has historically consumed the majority of accounting hours.

The architecture is consistent with where enterprise AI is going in 2026. The agent does not own the system of record. It owns the workflow that feeds the system of record, and the system of record retains the approval gate. That is the same pattern that has shown up in the AWS Continuum and Context launch from earlier this month, which is a similar bet that the right place for the agent is in the preparation, not in the posting. The Snyk Evo ADS launch from the same week is also a governance-first story on the developer side, and the through-line is the same. The agent does the work, the human signs off, the system of record stays clean.

The internal framing at Maxima is that Max is "an always-on teammate" rather than a one-off automation script. That is more than marketing language. The skill sets are designed to keep running across periods, with the agent carrying context from close to close, so the third quarter reconciliation does not start from scratch if the second quarter was closed cleanly. Whether the same model behaves well across a year of close cycles is one of the harder open questions, and Maxima is betting the company on the answer being yes.

Maxima was founded in 2024, has raised $41 million in funding, and was recently named by Redpoint as one of the top companies shaping the AI application layer in its inaugural AI64 list. The category Maxima is selling into is not "AI for accounting" in the broad sense. It is a narrower category that enterprise finance teams have been building toward for years: a preparation layer that runs continuously, a control surface that keeps the auditor happy, and a workflow that finally closes the gap between the modern AI stack and the chart of accounts.

The launch lands in a week that has already produced two other major enterprise agent stories, with the AWS Continuum and Context work shipping earlier in June and Snyk shipping Evo ADS for AI coding agents on June 23. The pattern across all three is identical. The vendor is not selling the agent as a replacement for the human. The vendor is selling the agent as preparation work that humans can review faster than they can produce it themselves. The question for the next year of enterprise AI is whether that framing survives contact with a controller who has to sign the year-end package.

For a broader look at how AI agents are reshaping finance and operations, the enterprise AI use cases for finance and operations resource page walks through the current state of the market, including the agent stack, the control surfaces, and where the audit and risk teams are likely to push back as adoption scales.