---
id: "claim-screen-clicking-is-flawed"
type: "claim"
source_timestamps: ["§ Information Systems"]
tags: ["rpa", "ui-automation", "system-architecture"]
related: ["concept-programmatic-agent-interfaces"]
confidence: "high"
testable: true
speakers: ["Harang Ju"]
sources: ["agentic"]
sourceVaultSlug: "hbr-seg-agentic"
originDay: 6
articleStem: "hbr-ext-17-workplace-set-up-for-agents"
sourceUrl: "https://hbr.org/2026/01/is-your-workplace-set-up-for-ai-agents"
sourceTitle: "Is Your Workplace Set Up for AI Agents?"
---
# Screen-clicking AI agents are a fundamentally flawed workaround.

Ju asserts that workarounds where AI tools 'look' at screens and click buttons reveal a deep mismatch between agents and human-designed systems — 'asking a computer to pretend to be a human using a computer' is inefficient and fragile (see [[quote-pretending-to-be-human]]). The true solution is exposing system capabilities through [[concept-programmatic-agent-interfaces|programmatic (API) interfaces]]. Stated as a contrarian position, this becomes [[contrarian-rpa-is-bad|screen-clicking AI (RPA) is a dead end]]. The implementation response is [[action-build-programmatic-interfaces|requiring API-first architecture]].

**Confidence:** high · **Testable:** yes.

**Enrichment / validation:** the architectural critique aligns with software-engineering consensus and agent-framework design (API-first, MCP). Counter-perspective: RPA proponents note UI automation is often the only feasible option in regulated or legacy environments lacking APIs, delivering real ROI; modern 'agentic RPA' blends screen understanding with limited APIs. Best framed as a transitional/bridge technology rather than a strict dead end.