What is robotic process automation (RPA)?

Robotic process automation (RPA) is software that mimics human interactions with computer interfaces — clicking buttons, typing into fields, copying data between applications, and navigating menus. The "robot" isn't a physical machine; it's a script that interacts with existing software the same way a human would, through the UI. Gartner defines RPA as "a productivity tool that allows a user to configure one or more scripts to activate specific keystrokes in an automated fashion." The market grew to $2.9 billion in 2025 according to Forrester, but growth is slowing as organizations discover RPA's limitations and shift toward AI-native automation.

Unlike generic AI automation posts, this guide shows real CodeWords workflows — not just theory.

Related reading: what is hyperautomation, what is an AI agent, AI workflow automation, workflow automation tools, automation platform, CodeWords integrations, CodeWords templates.

Why RPA exists

RPA emerged to solve a specific problem: automating tasks in legacy systems that lack APIs. If your accounting system is a 20-year-old desktop application with no API, the only way to automate data entry is to script the mouse clicks and keyboard inputs a human would use.

The core RPA use cases:

• Data entry into legacy systems without APIs
• Copy-paste operations between applications that don't integrate
• Form filling in web or desktop applications
• Report extraction from systems that only output to screen or print
• Rule-based processing of repetitive, high-volume tasks

RPA tools like UiPath, Automation Anywhere, and Blue Prism provide visual recorders that watch you perform a task and generate the automation script. You demonstrate the process; the tool replicates it.

How RPA works

The RPA execution model:

1. Record or script the interaction — mouse coordinates, field identifiers, keyboard inputs, wait conditions
2. Configure triggers — schedule, event, or manual start
3. Execute the bot — the script interacts with the application UI
4. Handle exceptions — when the UI changes or unexpected dialogs appear, the bot follows pre-defined error handling rules
5. Log results — record what was processed, what succeeded, what failed

RPA bots operate at the UI layer. They "see" the screen the same way a user does (or use application object models to identify UI elements). This makes them fragile — any change to the UI layout, button position, or field structure can break the automation.

Where RPA falls short

UI brittleness. A minor application update that moves a button or renames a field breaks the bot. RPA maintenance costs are significant — Forrester found that organizations spend 30-50% of their initial RPA investment on annual maintenance.

No intelligence. RPA follows scripts. It can't interpret ambiguous data, make judgment calls, or handle exceptions it wasn't explicitly programmed for. If an invoice has an unusual format, the bot fails.

Limited scalability. Each bot typically requires a licensed "robot" and a virtual machine to run on. Scaling to hundreds of bots means hundreds of VMs and licenses. Costs compound.

Surface-level automation. RPA automates the UI interaction, not the underlying process. You're automating a workaround (manual data entry) rather than solving the root problem (lack of API integration).

RPA vs API-based automation

API-based automation connects directly to application backends — faster, more reliable, and not affected by UI changes. This is what platforms like CodeWords, Zapier, and Make do.

RPA advantages over API automation: - Works with applications that have no API - Can automate desktop applications (not just web/cloud) - No development effort on the target application side

API automation advantages over RPA: - 10-100x more reliable (no UI fragility) - Faster execution (direct data transfer vs. simulated clicks) - Lower maintenance costs (APIs are versioned and documented) - More scalable (no per-bot licensing or VM requirements)

The practical guideline: if the application has an API, use API-based automation. If it doesn't, RPA is the fallback.

RPA vs AI automation

This is the more important distinction in 2026. RPA automates repetitive, rule-based tasks. AI automation handles tasks requiring judgment, interpretation, and decision-making.

RPA handles: "Copy the amount from cell B5 and paste it into field #invoice-total" AI automation handles: "Read this invoice, extract the total amount regardless of format, validate it against the purchase order, and flag discrepancies"

CodeWords operates in the AI automation space. Workflows run as Python microservices with native LLM access (OpenAI, Anthropic, Google Gemini), 500+ integrations, and web scraping capabilities (Firecrawl, AI Web Agent). Where RPA mimics human clicks, CodeWords workflows perform human-level reasoning — classifying documents, extracting entities from unstructured text, making routing decisions based on content analysis.

The key difference: RPA requires you to predefine every interaction step. AI automation requires you to define the goal and the guardrails — the AI figures out the steps.

When RPA still makes sense

RPA isn't dead, but its scope is shrinking:

• Legacy mainframe systems with terminal interfaces and no API layer
• Desktop applications that process-critical departments refuse to replace
• Regulatory environments where the legacy system is the system of record and must be used directly
• Bridge automation while an organization migrates from legacy to modern systems

For everything else — especially workflows involving data interpretation, content processing, or multi-system orchestration — API-based automation with AI reasoning (the CodeWords approach) is more reliable, more flexible, and more cost-effective.

Deloitte's 2025 RPA survey found that 52% of organizations with mature RPA programs are actively migrating workflows from RPA to API-based or AI-native automation platforms — the highest migration rate since RPA's mainstream adoption.

FAQs

Is RPA the same as AI? No. RPA follows pre-scripted rules. AI makes decisions based on learned patterns. They're different technologies that can be combined — "intelligent automation" uses RPA for UI interaction and AI for decision points.

Should I start with RPA or AI automation? Start with AI-native automation (CodeWords) for any workflow involving modern applications with APIs. Resort to RPA only for legacy systems that genuinely lack API access.

How much does RPA cost? Enterprise RPA licenses (UiPath, Automation Anywhere) range from $5,000-$15,000 per bot per year, plus infrastructure costs. CodeWords' usage-based pricing avoids per-bot licensing entirely.

Build AI-native automation at codewords.agemo.ai.