Safely manage your Zendesk from the AI assistant you already use, via the Deltastring MCP. Beacon configuration platform
← Back to news

Sierra’s Bret Taylor says the era of clicking buttons is over

Bret Taylor's assertion that the era of button-clicking is ending reflects a genuine shift in how enterprise software vendors are positioning themselves, though the gap between rhetoric and reality remains substantial. Sierra's Ghostwriter—an agent that builds other agents through natural language prompts—represents a meaningful departure from traditional UI-driven platforms, and Taylor's track record at Salesforce lends credibility to his vision. The underlying premise is sound: most enterprise tools sit dormant between infrequent use cases (onboarding, open enrollment, quarterly reporting), making them poor candidates for complex interface mastery. If this thesis holds, it fundamentally challenges the design philosophy that has governed CX platforms for two decades. The question for CX teams is whether this shift favours vendors who can retrofit natural language capabilities onto existing systems, or whether it creates an opening for purpose-built agents that bypass traditional ticketing and case management interfaces entirely.

However, the critical caveat embedded in the reporting undermines Taylor's autonomy claims. Sierra and competitors like Harvey employ forward-deployed engineers who continuously tune and maintain customer agents—a labour-intensive model that contradicts the "describe what you need and it deploys itself" narrative. This is not a minor implementation detail; it suggests that agent-as-a-service remains a high-touch, bespoke engagement rather than a plug-and-play alternative to traditional software. For support teams currently managing Zendesk or Freshdesk instances, this distinction matters considerably. The promise of rapid deployment (Nordstrom in four weeks) comes with hidden operational costs that may not scale as Sierra's valuation and growth trajectory suggest. The real question is whether forward-deployed engineering becomes a permanent feature of AI agent implementation, or whether it represents a transitional phase before true autonomy emerges—and which scenario poses greater risk to teams already invested in configurable, self-service platforms.