The End of RPA As We Know It
I used to sell RPA. Here's why I'd hesitate today.
As an RPA consultant, I learned that a task had to pass three strict tests before it was worth automating:
- Meticulously rule-based - one ambiguous step and the bot breaks.
- High volume - you need enough repetitions to justify the project ROI.
- Small scope - only a fraction of any end-to-end workflow could actually be automated due to the above reasons.
Even after passing all three, a single unexpected popup or a typo in a field name would halt the entire process. Maintenance was constant.
Last week, I used Claude CoWork. It opened Chrome, navigated between tabs, collected data, refined it - and handled every unexpected UI element along the way. No scripted selectors. No brittle workflows.
All I did was install the Claude Chrome extension and Claude Desktop.
What took weeks of RPA development - process mapping, selector tuning, exception handling - was done in minutes with natural language instructions and computer use.
This isn't an incremental improvement. It's a category shift:
RPA:
- Input: Scripted selectors & rules
- Flexibility: Breaks on any UI change
- Scope: Fragment of workflow
- Setup: Weeks of development
- Maintenance: Constant
AI Agent (Computer Use):
- Input: Natural language intent
- Flexibility: Adapts in real-time
- Scope: End-to-end process
- Setup: Minutes of instruction
- Maintenance: Near-zero
RPA solved the "too repetitive for humans" problem. AI agents solve the "too complex for scripts" problem - while also handling everything RPA could.
If you're in the RPA business, this isn't a threat to ignore. It's time to redesign the business model - from selling bots to orchestrating AI agents.
The question isn't whether RPA will be replaced. It's how fast.
← All posts