Agents as a Service

We founded Sierra three years ago, shortly after ChatGPT upended how people use software, convinced that AI would reinvent the customer experience. Fast forward to today, and Sierra is the leading platform for customer experience AI agents, working with 40% of the Fortune 50 and brands as diverse as ADT, Chime, Cigna, Next, Nordstrom, Nubank, Minted, Ramp, Rocket Mortgage, SiriusXM, Singtel, and Wayfair.

Three months ago, we felt a similarly significant technology shift with Codex and Claude Code. Coding agents like these are changing the way companies like Sierra make software—and offer a glimpse of how all software will work in the future.

Code → no-code → no clicks

It’s hard to overstate the impact of software that can build and use software. While no one knows what the future will look like, some things are already clear. First, as clicks turn to prompts, building agents will become as simple as describing what you want. Second, unlike people, computers don’t need simple, clean interfaces—just direct access to the underlying data and actions.

The end of the web app

Put these together, and the web app with all its menus, form fields, and tables starts to feel like a “horseless carriage”—technology built on an assumption that no longer holds. Use Codex or Claude for a day and you feel it immediately: why am I doing all this clicking, scrolling, and typing?

So we asked ourselves: what replaces it? If we were starting from scratch, how would we build Sierra? And that’s what we’re introducing today: a reimagining of Sierra, built around an agent that takes your direction and does the work for you. We call that agent Ghostwriter.

Ghostwriter: the agent-building agent

With Ghostwriter you no longer need to edit journeys, write integrations, create simulations, or triage issues manually. You define what you want, and the agent delivers it.

Upload SOPs, transcripts from support calls, photos of whiteboard sketches, process documentation, and audio recordings. Or explain what you want to achieve in plain English. Ghostwriter takes all of it, identifies the key behaviors and edge cases, and makes it into a production-ready agent across voice, chat, email, and over 30 languages with sophisticated guardrails built-in.

Here’s what it looks like in practice:

But if you’ve ever built an agent before, you know the hard part isn’t going from zero to one, it’s improving it over time. That’s where Explorer comes in. It’s like ChatGPT Deep Research, but instead of doing research on the Internet, it’s using your customer conversations to understand trends and identify areas for improvement.

Under the hood: the agent harness

Agents perform best when they have strong scaffolding—tools, memory, a coherent action space, the ability to plan and reason—and the right context about the task at hand. This scaffolding is the agent harness.

Building Ghostwriter meant rearchitecting Sierra as headless infrastructure so an agent can use it directly. Ghostwriter has access to the platform’s full workspace, as well as a clear way to test and safely validate changes in a sandboxed environment. All this enables Ghostwriter to build and improve the most sophisticated agents reliably and autonomously.

The agent assembly line

Over time, this becomes something of an agent assembly line: Ghostwriter analyzes real interactions, identifies opportunities for improvement, validates them, and prepares them for review. The cycle of analyzing, improving, testing, and shipping happens automatically—and better and better agents emerge.

This is Agents as a Service: prompts, not clicks. No menus, fields, or tables (however beautifully designed), and no co-pilots—just outcomes you define and agents that deliver them. Now we need an agent to fix the acronym.