Colombian fintech Simetrik launched an autonomous AI agent in August 2026 that handles end-to-end financial control tasks, from configuring data sources to implementing reconciliations and analyzing exceptions. The twist: an "open box" architecture the company says solves the trust problem that has slowed agentic AI adoption in finance departments.
The product, called Simetrik Agent, became available to customers on August 3, 2026. It operates in three modes: natively inside Simetrik's platform, as a Model Context Protocol server connecting Claude, Copilot, and Gemini to the company's control plane, or through a command-line interface for terminal and CI/CD workflows.
Simetrik's pitch hinges on splitting the workload. The agent handles probabilistic tasks like understanding natural-language requests and interpreting patterns, but when precision matters, it relies on what the company describes as 110 deterministic financial-control functions. Those produce exact, verifiable results with full audit trails, according to the launch materials.
"It is not a black box, it is an open box," Simetrik wrote in its announcement. Every decision gets expressed in financial language. Users can supervise and modify the agent's work, and the system requests human involvement when it reaches the edge of its confidence.
Whether that architecture addresses the core tension finance executives face remains an open question. Teams are racing to automate reconciliations and month-end close processes, but regulators and auditors want explainability. KPMG noted in May that finance leaders need assurance as they scale AI, and that auditors will soon provide sign-off on AI systems alongside financial statements. EY argued in June that banks require an orchestration layer to enforce guardrails and auditability as agents proliferate.
The International Monetary Fund raised similar concerns in an April fintech paper, pointing out that agentic AI in payments must reconcile probabilistic decision-making with deterministic financial infrastructures. J.P. Morgan Payments wrote in June that an agentic treasury system must emphasize audit trails for comfort with high-value payments, including exception context and confidence intervals.
Simetrik has staked out a position in its blog posts that AI agents should not "guess" when financial decisions require reproducibility. The company argues AI should handle interpretation and pattern recognition, while execution runs through deterministic controls and governed approvals.
Plain Language, Strict Permissions
Users can operate Simetrik by talking to the agent in conversational language. No commands to memorize, no IDs to look up, the company's documentation states. If the action falls within Simetrik's normal operation, the agent handles it.
That scope includes loading data, running reconciliations, scheduling the close, creating and editing reconciliations, defining rules and transformations, building joins and consolidations, configuring exports to Amazon S3, setting alarms, requesting approvals, viewing dependency graphs, and monitoring task status.
The system operates under each user's existing role and permissions. It cannot bypass configured approvals, manage users or roles, or permanently delete anything without explicit confirmation.
The MCP server requires one URL and OAuth sign-in. Users can optionally pre-approve read-only tools to reduce prompts, according to Simetrik's connection guide. The server currently runs in the U.S. region only. MCP tools break into three categories: read-only tools, session tools for scope, and one agent tool that builds or changes anything. Write operations require user confirmation, and the server enforces the user's permissions throughout.
The CLI installs with one line on macOS, Linux, and Windows. It ships with a Claude skill. Users paste a "first conversation" prompt into Claude to confirm connectivity.
Crowded Field
Simetrik is hardly alone in the race to automate financial operations with agentic AI. Trintech launched Flux Agent and Variance Analysis Agent on June 25 for the financial close and FP&A. Workiva announced specialized AI agents and an intelligence layer in August to accelerate reporting and compliance with traceability. BlackLine has positioned its Verity and Studio360 products as agentic AI tools for autonomous accounting. Microsoft introduced Copilot for Finance in February 2024 with a roadmap extending through 2026.
Adyen offers reconciliation dashboards and automation. In June, TI Inside reported that Adyen partnered with Simetrik to streamline reconciliation and control workflows through a template integration.
Traction and Scale
Simetrik says it serves 180 clients across more than 50 countries, processes 2.5 billion records daily, and reconciles more than $500 billion in annual transaction value with a greater-than-99.9-percent automatic reconciliation rate. Those figures are company-stated and not independently verified.
The company closed a $30 million Series B extension in June 2025 from Goldman Sachs Alternatives, bringing total Series B capital to $85 million, according to Contxto. Goldman Sachs Asset Management led the initial $55 million Series B round in February 2024, TechCrunch reported at the time. Investors include Cometa, FinTech Collective, Moore Strategic Ventures, Monashees, and Upload Ventures.
Mastercard selected Simetrik on September 3, 2026 for Start Path's inaugural Agentic Commerce and Services cohort. The company also joined AWS Marketplace in May.
"The future of financial control is agentic and autonomous," Santiago Gómez, Simetrik's co-founder and chief operating officer, said in the launch post.
Writing on LinkedIn in mid-August, Gómez framed Simetrik Agent not as a new interface but as a way to deploy the knowledge the company has encoded from 180 enterprises and thousands of implementations. "The difference is what sits on top" of frontier models, Gómez wrote, a reference perhaps more pointed than he let on.
The agent became available to customers on August 3, 2026. The company has not disclosed pricing or expansion plans beyond its current U.S. MCP server region.
