Fuel Agents

• Running daily bank reconciliation across several accounts so the controller only reviews the unmatched lines and reconciling items
• Scheduling a month-end close checklist that gathers missing documents, checks closed periods for edits, ties out the balance sheet and produces one findings report
• Matching payroll to the payroll register and rolling forward liabilities before the books are closed
• Keeping a 13-week cash forecast and runway current, with a Slack alert when cash falls below a set floor
• Drafting monthly management commentary and a board pack in which every number links back to its source
• Querying Fuel's normalized actuals and forecast from Claude or ChatGPT through the Fuel MCP server

Most AI finance agents aim at FP&A: forecasts, scenarios and answers to questions. Fuel Agents aims at the less glamorous controller work underneath, the reconciliations, recognition checks, tie-outs and closed-period scans that decide whether the numbers can be trusted. Its design choices fit that job: narrow agents instead of one general assistant, matching done in code, testing against periods you have already signed off, a completeness score on every run, and drafts that wait for an approver. The catch is that you are buying an engagement as much as a product. There is no public pricing, documentation or integration list, the data has to flow through Fuel's warehouse, and the first workflows are built by Fuel's own engineers. That suits a mid-market finance team that wants help getting started, and it is less suited to anyone who wants to evaluate the tool alone. Use the four-week pilot on one painful process, such as bank reconciliation, and check that your ERP's write-back path and your auditors' evidence requirements are covered before you widen it.

About Fuel Agents

Fuelfinance, founded in 2019, built its business on FP&A for startups and small businesses that combined software with its own finance professionals, and launched Fuel Agents on September 30, 2026; its site now leads with the agents. The idea is that each agent does one narrow job and workflows do the rest. The agents library says it has 35+ agents (the page currently lists 39) in four groups. Close and control covers document gathering, closed-period and chart-of-accounts change detection, completeness and duplicate checks, revenue and expense recognition, payroll register matching, intercompany matching, balance sheet reconciliation and tie-out, and a consolidation agent that rolls every finding into one report with an accuracy score. Cash and treasury covers bank and vendor reconciliation, multi-bank cash position, a 13-week direct-method cash forecast, runway alerts, AR aging, collections ranking and payment-run proposals. Analysis and reporting covers P&L decomposition, variance drill-downs, plan vs actual, scenarios, unit economics, KPI snapshots, management commentary and board packs. Platform agents keep company context current, sequence runs and halt on critical findings, and check permissions and data exposure before output ships. You pick agents, set the order and the schedule, and the workflow asks for missing files, reports in Slack or Teams, and remembers your corrections so a finding you dismissed does not come back next month. Agents read your systems by default; they can draft invoices, journal entries and payment runs in your ERP, billing or CRM, but nothing posts until someone with approval rights signs off, and an audit log records the data used, changes, assumptions, approver and timestamp. Fuelfinance says deterministic steps such as matching and tie-outs run as code rather than through a model, and that each agent is tested on your past periods before it runs on the current one. Data comes through Fuel's own warehouse, which normalizes sources (Fuel cites 300+ integrations, including QuickBooks, NetSuite, Stripe and HubSpot), and a separate Fuel MCP server exposes the same data to Claude, ChatGPT and other MCP clients. What sets it apart from a general assistant connected to QuickBooks is the governed close-and-control layer: scheduled runs, testing on prior periods, completeness scoring, approval gates and an audit trail. It is also part software and part service: deployment starts with a four-week pilot in which Fuel's finance engineers build and run one or two workflows on your data, then train your team to build its own, and the pricing page still sells dedicated FP&A manager or CFO support alongside the platform. The tradeoffs: pricing is custom and not published, there is no public documentation, API reference or full integration list, the product pages state no compliance certifications (Fuelfinance does run a separate trust center), the AI models are not disclosed, and the product is newly launched.