Bret Taylor strides onstage at HumanX, San Francisco buzzing, and drops the bomb: the era of clicking buttons? Over. Dead. Kaput.
Sierra’s Ghostwriter—launched last month—promises to torch traditional software interfaces, replacing them with natural language commands that birth entire AI agents on demand. Taylor, ex-Salesforce co-CEO, isn’t mincing words. Enterprises waste billions on clunky tools nobody loves, he says, tools you touch twice a year if you’re lucky.
“You sign into Workday when you onboard as a new employee, and maybe for open enrollment,” Taylor told the audience at the HumanX conference taking place this week in San Francisco.
And here’s the kicker: Ghostwriter doesn’t just chat. It builds. Describe your need—say, handle Nordstrom’s returns chaos—and it crafts, deploys, and runs a custom agent. Taylor boasts they did Nordstrom in four weeks flat. Sierra’s already at $100M ARR, valued at $10B. Impressive? Sure. But let’s peel back the hype.
How Does Ghostwriter Actually Build Agents?
Start with the architecture. Ghostwriter sits atop Sierra’s agentic stack—think multimodal LLMs fine-tuned for enterprise drudgery, hooked to APIs, databases, the works. You prompt in English: “Create an agent that flags fraudulent claims in real-time.” Boom. It reasons step-by-step, scaffolds tools (email, Slack, ERP pulls), iterates on failures via self-reflection loops. No code. No low-code drag-and-drop. Pure language.
But wait—under the hood? It’s not magic. Sierra’s models chain reasoning like Devin or Anthropic’s Claude, but optimized for customer service agents. They parse intent, orchestrate sub-agents for subtasks (auth, data fetch, decision), then loop back with human-in-loop for edge cases. Taylor calls it “agent as a service.” Sounds slick. Enterprises crave this; they don’t build software, they solve pains.
Yet skeptics—techies and VCs whispering to TechCrunch—point out the catch. Autonomy? Not quite. “Forward-deployed engineers” tweak these beasts constantly. Harvey.ai does it too. Sierra’s no different. Taylor spins it as speed; critics call it smoke.
One paragraph. That’s all it takes to smell the parallel: remember the GUI revolution? DOS commands died under mouse tyranny. Now Taylor wants NLP to bury the GUI. Bold. But here’s my unique take—it’s not evolution, it’s regression with steroids. Command lines were precise; GUIs dumbed it down for masses. Agents? They’re command lines reborn, but probabilistic, hallucination-prone. What if your “precise” agent books the wrong flight because it misread sarcasm?
Why Do Enterprises Still Need Sierra’s Hand-Holding?
Dig deeper. Sierra hit $100M ARR in 21 months—thunderclap growth. Greenoaks poured $350M at $10B valuation. Why? Because legacy software rots. SAP, Oracle—behemoths gathering dust. Taylor’s right: most firms want outcomes, not UIs.
Ghostwriter’s how: modular agents compose like Lego. Base layer: perception (parse emails, calls). Action layer: API calls, scripts. Guardrails: compliance checks, audit trails. Deploy in days, not quarters. Nordstrom example? They slashed resolution times 40%, per Sierra. But peel the onion—those four weeks included Sierra engineers embedding, training on proprietary data, iterating prompts.
And the why? Architectural shift from monolithic apps to agent swarms. Think microservices, but alive. No more vertical silos; horizontal, language-driven flows. Taylor predicts ubiquity: “I truly think that’s where the world is going.”
But call the PR spin. Unparalleled speeds? Sure, if you count Sierra’s army. True zero-touch autonomy lurks years off—LLMs falter on novel scenarios, long contexts crumble. Prediction: by 2027, 30% of enterprise tasks agent-ified, but 80% with human oversight. Sierra wins either way.
Is Bret Taylor Overhyping the Button Apocalypse?
Look, Taylor’s no newbie. Salesforce CTO, Twitter chair— he’s seen shifts. But this feels like Salesforce’s Einstein 2.0: promise the moon, deliver copilots. Ghostwriter’s edge? Vertical focus on service agents, not generalists.
Critique time. Companies claim agents; reality’s hybrid. Forward-deployed teams mean costs rival custom dev. Scale that to Fortune 500? Sierra’s $10B bet assumes moat via data flywheels—customer convos fine-tune models. Risk: commoditization. OpenAI’s o1, xAI’s Grok loom.
Wander a sec: imagine HR sans Workday logins. “Promote Alice, bump salary 10%.” Agent executes, emails draft, payroll pings. Frictionless? If reliable. Early tests show 85% success rates—good, not godlike. Taylor’s vision scales if agents self-improve via RLHF loops on enterprise data.
Skepticism peaks here. TechCrunch notes: many agents need constant tuning. Sierra concurs privately—it’s their secret sauce. So, buttons die slow. Agents augment first.
Shift gears. Historical echo: 90s client-server wars. Thin clients promised no local apps; clouds ate them. Agents? Cloud brains for software.
What Happens When Agents Go Rogue—or Wrong?
Edge cases haunt. Bias in claims processing? Hallucinated compliance? Sierra bakes observability—dashboards track agent “thoughts,” intervene realtime. Smart.
Bold call: this sparks agent governance wars. Regulators eye in; EU AI Act classifies high-risk. Sierra positions as safe—auditable, explainable chains. Competitors scramble.
Enterprises test: Verizon, Block piloted. Wins: 24/7 coverage, no hiring sprees. Losses: trust gaps. Taylor’s unfazed—iterate fast.
Wrap the deep-dive. Sierra’s not ending clicks tomorrow. But the architecture tilts: language-first stacks redefine UIs as relics. Watch ARR climb; valuation holds if delivery matches talk.
**
🧬 Related Insights
- Read more: LangSmith Fleet Gains 7,500 Arcade Tools: Secure Gateway for Agents or Just Hype?
- Read more: Suno’s Flimsy Filters: AI Clones Beyoncé and Floods Streams with Ripoffs
Frequently Asked Questions**
What is Sierra Ghostwriter?
Ghostwriter’s Sierra’s tool to auto-build AI agents from natural language prompts, targeting enterprise customer service tasks—no coding needed.
Will AI agents like Ghostwriter replace human engineers?
Not yet—they speed deployment but rely on Sierra’s forward-deployed experts for tuning and reliability.
Is Sierra’s $10B valuation justified?
Growth to $100M ARR in under two years suggests yes, but sustained autonomy will decide.
