Financial Automation and AI Agents: The Future of Finance Processes with Python

At AI Day 2025, held at Instituto Caldeira, more than 4,000 people took part in a full day dedicated to artificial intelligence, with over 50 sessions and 116 experts discussing real-world applications, challenges, and concrete paths for the future of AI in business.

It was in this context that Catarina Pinheiro delivered a provocation that immediately captured the audience’s attention: automation is no longer about speeding up tasks. It’s about transforming how financial decisions are made.

The significance of the moment was reflected in the audience itself: CFOs, controllers, heads of operations, and specialists who feel daily pressure around efficiency, compliance, and speed. For this group, what Catarina presented was not theory — it was a real map of how high-volume financial operations are already being transformed with AI and Python.

She highlighted that the most advanced companies are already operating with leaner teams, smarter processes, and workflows that continuously evolve as AI learns from data.

To frame this shift, Catarina shared data from Gartner, one of the world’s leading technology research and advisory firms. In the report “Top Strategic Technology Trends for 2024–2028”, Gartner projects that by 2028:

• 33% of enterprise software will incorporate agent-based AI,
• 15% of daily operational decisions will be made autonomously.

These are not distant, speculative predictions. They reflect a transformation already underway. And that was the core message of the talk — the future of finance is not far away. It is already being built by companies modernizing their processes through advanced financial automation.

From Automation to Autonomy: How Finance Is Changing

There is a fundamental distinction that many finance teams feel every day but rarely articulate clearly: there is a massive difference between automating tasks and automating decisions.

Most companies stopped at the first stage — digitizing clicks, building workflows, and replicating spreadsheets.

What truly unlocks the strategic potential of finance teams is advancing toward models capable of interpreting information, making decisions, and learning from the process itself.

To explain this shift, Catarina outlined the evolution of financial automation in three major phases. Each one solves a different problem — and each one has a clear limit.

1. RPA: Efficiency Through Structured Repetition

RPA marked the first major turning point.

It allowed companies to replace routine tasks with automated flows: copying data between systems, filling out screens, generating reports. It was revolutionary because it removed finance teams from manual, repetitive work.

But RPA has a defining characteristic:
it only works when the world is predictable.

• How it works: deterministic rules (If/Then), with no room for interpretation.
• Where it works best: stable processes with fixed screens, standardized fields, and zero exceptions.
• Structural challenges:
  • any UI change breaks the automation;
  • reliance on interface automation creates fragility;
  • constant maintenance is required;
  • often runs on local machines — a risk for compliance and business continuity.

RPA delivered speed, yes.
But as Catarina emphasized, it was never designed to handle ambiguity, context, or judgment — elements that define roughly 80% of real financial operations.

That’s why, despite its usefulness, RPA alone cannot keep up with the growing complexity of areas like accounts payable, reconciliations, billing, and tax compliance.

2. AI Agents — Intelligence for the Unpredictable

AI Agents represent the real inflection point in financial automation.
While RPA depends on a perfect environment — stable screens, fixed fields, immutable rules — AI agents can interpret information, make decisions, and adapt workflows based on context.

They use large language models (LLMs) to understand documents, text, and nuances that previously required human intervention. This is automation that goes beyond filled fields: it understands intent, identifies patterns, compares history, and reacts based on what the process actually represents.

During the talk, Catarina shared a simple but powerful example.
If a supplier changes the layout of an invoice from one month to the next, a traditional automation flow would fail immediately. An AI Agent, however, recognizes the new format, interprets the relevant data, and continues the process without interruption.

This ability to handle exceptions, variation, and unstructured information is what makes AI Agents essential for financial processes that have never been 100% standardized in the real world.

Python in Finance: The Engine Behind Modern Automation

At AI Day, Catarina was very direct: the autonomy of AI Agents only exists because there is a solid technical layer executing every action — and that layer is Python.

Agents interpret.
Python ensures execution with precision, scale, and governance.

In finance, where integrations are critical and rules change constantly, Python has become the standard language because it combines technical robustness with operational flexibility.

Why Python became the standard in modern finance

Python addresses three core challenges of financial automation:

• Real integration with the company ecosystem: connects ERPs, banks, and APIs without requiring system migrations.
• High-volume processing: reconciles millions of records, applies tax rules, and validates documents in seconds using libraries like Pandas and NumPy.
• Governance and compliance: provides versioning, complete logs, and full traceability — essential requirements for audits, SOX, and internal controls.

Beyond that, Python adapts quickly to the specifics of each financial operation, from tax changes to day-to-day business exceptions.

The outcome is straightforward:
AI Agents help interpret complex scenarios. Python guarantees reliable and consistent execution of financial processes.

This combination is what enables truly scalable financial automation.

How This Plays Out in Accounts Payable

Catarina also shared practical examples from accounts payable — an area that concentrates much of finance’s operational complexity and where the combination of AI and Python is already eliminating errors, rework, and delays.

Scenario 1 — Tax discrepancies

• Problem: ISS tax rate differs from the purchase order.
• Before: the process would stall.
• Now: the Agent interprets the invoice, checks tax rules, and decides:
  • small adjustments → automatic approval;
  • material discrepancies → drafts an email to the supplier and waits for human validation.

Scenario 2 — Unstructured information in emails

• Problem: vague descriptions sent in the email body.
• Agent: interprets the text, categorizes the expense, identifies the cost center, and prepares the ERP entry.

The team moves from “typing” to supervising decisions.
Work quality increases — and operational risk drops.

Strategic ROI: Far Beyond Efficiency

Catarina reinforced a key point:
financial automation is not about cutting hours — it’s about unlocking financial intelligence.

Impact	Description
Loss reduction	Prevents duplicates and late-payment penalties.
Cash visibility	Real-time updates as invoices enter the system.
Scalability	Growth without inflating headcount.
Active compliance	Certificates, data, and documents validated before every payment.

As she put it on stage:

“Automation doesn’t replace analysts — it promotes them. Finance stops operating in the past and starts operating in real time.”

Security and Governance: Humans Remain in Control

Automation does not mean loss of control.
On the contrary, it delivers predictability.

Abstra operates with a Human-in-the-Loop model:

1. Low risk: utility bills → automatic approval.
2. Medium risk: new suppliers → agent prepares, human validates.
3. High risk: unusual amounts → direct alert to management.

Structured Output: What Ensures Financial Integrity

Even when an Agent “thinks,” its output must be deterministic:

• valid JSON,
• consistent tables,
• properly formatted XML,
• ERP-compatible data.

AI interprets.
Python executes with precision.

The Future of Finance: Intelligent, Adaptive, and Programmable

The conclusion of the talk was clear:
financial processes are living systems — and their automation must be living as well.

Companies adopting AI Agents and Python are moving from task management to decision management.
They are replacing rework with predictability.
They are turning finance into a strategic engine, not just an operational function.

If you want to understand how this architecture fits into your ERP, banking stack, or transaction volume:

👉 Talk to our specialists at Abstra

Read also

• The New Financial Automation Landscape
• How to Write Better Prompts for AI in Finance