Introduction

Many business workflows still rely on navigating complex web applications. Financial analysis is a clear example. Analysts often spend significant time searching for filings, opening documents, extracting tables, and interpreting financial statements across multiple pages. These steps are repetitive and time consuming.

Recent advances in language models and browser automation now make it possible to describe the desired outcome in natural language while an intelligent agent performs the work. UIAgent is built around this idea. It allows users to describe a task in plain language while an autonomous browser agent executes the workflow and, once validated, converts it into a reusable automation.

A key capability of UIAgent is the ability to transform probabilistic agent behavior into deterministic automations. During the initial execution, the agent explores the task space using reasoning and dynamic decision making. Once the workflow is validated, that execution can be captured and converted into a fixed sequence of actions that runs reliably and consistently.

This article demonstrates how UIAgent can automate financial analysis by retrieving and processing SEC filings. The workflow begins with a simple natural language instruction and evolves into an efficient, reusable automation.

Starting the Workflow with Natural Language

The process begins inside the UIAgent interface using the CUGA autonomous computer use agent. Instead of writing scripts or manually defining automation steps, the user simply describes the outcome they want.

In this example, the user asks the agent to analyze the latest financial statements for Tesla directly from the SEC website.

Once the request is submitted, CUGA begins executing the task in the browser. It interprets the request, determines the necessary steps, and starts navigating the relevant websites.

At this stage, the system operates in probabilistic agent mode. The agent reasons about the interface, decides which links to follow, and determines how to extract the required information. This flexibility allows it to solve tasks without requiring predefined scripts or manual automation design.

Navigating the SEC Filing System

The first step is locating the company within the SEC database. The agent navigates to the SEC CIK lookup page and prepares to search for the company.

Next, the agent enters the company name into the search field.

This step retrieves the official company record associated with the Central Index Key. The CIK is the unique identifier used by the SEC to track corporate filings and disclosures.

Locating the Company Filings

After identifying the company, UIAgent opens the EDGAR company page.

This page contains company information along with a list of filings submitted to the SEC.

From this interface, the agent expands the section containing annual and quarterly reports.

Retrieving the Financial Filings

UIAgent then navigates to the complete list of filings and identifies the most recent annual and quarterly reports.

These documents typically include the financial statements required for analysis.

The agent selects the relevant filings and opens the documents in the SEC viewer.

Extracting Financial Data

Once the filing is opened, UIAgent scans the page and extracts the financial tables contained in the report.

These tables usually include the income statement, balance sheet, and statement of cash flows.

The extracted tables are converted into structured data, allowing them to be analyzed programmatically.

Saving the Workflow as a Reusable Automation

After the agent successfully completes the task, UIAgent provides the option to save the workflow as a reusable automation.

This step is central to the system’s design. The initial execution is performed through agent reasoning and exploration. Once the workflow has been validated, the system captures the sequence of actions and converts it into a deterministic automation.

In this example, the workflow is organized into modular components.

Securities search retrieves the company record and filing links.

Get Securities Info extracts financial tables from the filing document.

Analyze Financial Forms processes the extracted financial statements and generates a structured financial analysis.

By converting the agent's exploratory execution into a deterministic workflow, UIAgent ensures that future runs execute reliably without requiring the agent to replan each step.

Why Converting Agent Workflows to Automations Matters

Transforming agent executions into reusable automations provides several important advantages.

Deterministic Execution

During the initial run, the agent operates in a probabilistic mode, using reasoning to determine the correct steps to complete the task. Once validated, the workflow can be stored as a deterministic automation.

This conversion allows organizations to move from exploratory agent execution to predictable automation, ensuring consistent outcomes while preserving the flexibility of natural language interaction during the discovery phase.

Faster Execution

When the workflow is saved, UIAgent no longer needs to dynamically reason about each action. Instead, the automation simply executes the validated sequence of steps.

Because the system skips the reasoning stage, the automation runs significantly faster than the original agent-driven execution.

Cost Efficiency

Agent reasoning typically requires multiple language model calls. By capturing the workflow after the initial run, those reasoning steps are no longer required for subsequent executions.

As a result, token based usage costs are primarily incurred during the initial teaching phase, while repeated executions run with significantly lower AI overhead.

This model allows teams to invest AI reasoning once and then benefit from efficient, low cost execution across many future runs.

Real World Application

In practice, teams can use UIAgent to create automations without writing code. A financial analyst might open a conversation with CUGA and request a financial overview of a company. The agent performs the research by navigating the SEC database, retrieving filings, extracting financial tables, and producing the analysis.

After reviewing the results, the analyst can choose to save the workflow as an automation. UIAgent records the sequence of steps and stores it as a reusable deterministic process.

Future executions no longer require the full reasoning process used during the initial run, allowing the workflow to execute faster and with significantly lower AI cost.

For example, a team might create a reusable automation called Analyze Financial Forms. This automation accepts a company name and a filing link as inputs. When executed, it automatically searches the SEC database, retrieves the latest filings, extracts financial tables, and produces a structured financial analysis.

Because the automation is stored within UIAgent, it can be reused by other workflows or invoked by other agents. A research dashboard could trigger the automation whenever a new filing is released. An internal analytics agent might run the automation across multiple companies to generate comparative financial insights.

Over time, organizations can build a growing library of automations originating from natural language interactions. Each automation begins as an exploratory agent session and evolves into a reliable deterministic workflow that executes quickly and consistently.

Conclusion

UIAgent introduces a new approach to automation where workflows begin as conversations and evolve into reusable systems.

Users interact with the CUGA autonomous agent using natural language. The agent navigates web interfaces, extracts data, and performs complex tasks without requiring predefined scripts.

Once a workflow has been successfully executed, UIAgent allows it to be captured and converted into a deterministic automation. This enables teams to combine the flexibility of agent based discovery with the reliability, speed, and cost efficiency of fixed automation workflows.

Over time, organizations can build a growing library of automations created directly from natural language interaction. In the case of financial analysis, this means that retrieving and analyzing SEC filings can evolve from a manual research process into a fast, repeatable workflow powered by UIAgent.