Bret Taylor thinks you’ll stop clicking buttons soon. The Sierra co-founder and CEO told attendees at the HumanX conference in San Francisco that natural language will replace traditional software interfaces, according to TechCrunch AI.
Taylor’s argument is simple: most enterprise software sits unused. “You sign into Workday when you onboard as a new employee, and maybe for open enrollment,” he said. His startup’s answer is Ghostwriter, an agent that builds other agents. Users describe what they need in plain language, and Ghostwriter creates and deploys a specialized agent to handle it.
This is the “agent as a service” pitch, and it’s worth paying attention to for a few reasons.
Why This Matters Now
Taylor isn’t some random founder making bold claims. He was co-CEO of Salesforce and chairman of Twitter’s board during the Elon Musk acquisition. Sierra hit $100 million in annual revenue run rate less than 21 months after founding. The company carries a $10 billion valuation after raising $350 million from Greenoaks Capital last September.
That gives his predictions weight. When he says Sierra deployed an agent for Nordstrom in just four weeks, it signals real enterprise traction.
“Most companies don’t want to make software,” Taylor said. “They want solutions to their problems.” That framing captures what’s shifting. Enterprises have spent decades buying software licenses, training employees on complex UIs, and hiring admins to manage it all. If agents can absorb that complexity, the ROI argument practically makes itself.
The Reality Check
Here’s where it gets interesting. TechCrunch AI reports that several technologists and investors paint a different picture behind the scenes. Many companies claiming to offer AI agents, including Sierra and legal AI startup Harvey, rely on “forward-deployed” engineers who constantly update and fine-tune customer agents to keep them running properly.
That’s a meaningful gap between the vision and the current state. “Autonomous” agent deployment still requires significant human oversight. The four-week Nordstrom deployment likely involved Sierra engineers on-site or deeply embedded, not a self-service Ghostwriter session.
This doesn’t invalidate the direction. It does suggest we’re in the “managed services” phase of AI agents, not the fully autonomous phase Taylor describes.
What to Watch
Three things will determine whether Taylor’s prediction plays out:
- Reliability at scale. Agents that work 95% of the time aren’t good enough for customer-facing deployments. The last 5% is where forward-deployed engineers earn their keep.
- The middleware question. If every enterprise needs custom agent fine-tuning, Sierra is really selling consulting with an AI wrapper. If Ghostwriter can genuinely reduce that customization, the economics change dramatically.
- Incumbent response. Salesforce, ServiceNow, and other enterprise giants are building their own agent layers. Taylor knows Salesforce from the inside. His bet is that a startup can move faster than the platforms he helped build.
What This Means for You
If you’re running an enterprise software team, start identifying workflows where users interact with your product infrequently. Those are the first candidates for agent replacement. If your users need training docs to complete basic tasks, that’s a sign the interface is ripe for disruption.
For AI practitioners, the takeaway is pragmatic: agent-building tools are proliferating, but production reliability remains the bottleneck. The companies that solve monitoring, error handling, and graceful fallbacks for agents will capture more value than those focused purely on generation.
Taylor’s vision of a post-click world is directionally right. The timeline is the open question. Right now, AI agents still need human hands behind the curtain. The full story is available at TechCrunch AI.