Six months ago, Sierra co-founders Bret Taylor and Clay Bavor declared “the end of the web app” as they launched Ghostwriter, an AI agent for customers to build other AI agents.
Now, the software unicorn recently valued at $15.8 billion is leveling up that product in what Bavor calls Sierra’s biggest product updates of the year: launching a new, more proactive Ghostwriter that doesn’t just build agents, but identifies and suggests improvements and fixes on its own.
“The key insight we had is that every one of these interactions that our agents are having is an opportunity to improve how the agent performs, behaves, and does the thing it’s doing,” Bavor tells Upstarts.
The upgraded Ghostwriter lives in Slack (or Microsoft Teams and email) and can assign work, ask for feedback on simulated experiments, and make counter proposals around new feature requests. Ghostwriter can also monitor conversations against outcomes and goals set by its human managers, using that information to propose new features or strategies, unprompted.
Sierra has been testing this souped-up version of its agent builder with about 40 customers for the past several months, the company tells Upstarts exclusively. For the AI unicorn, which recently signed on corporations like Hyatt, Santander, Uber and Virgin Media O2 as customers, the launch represents another step toward widening the aperture of how it helps the world’s biggest companies with AI agents in areas expanding beyond its original customer service focus.
And for the wider startup ecosystem watching Sierra – earlier this month, we had a startup CEO tell us that they were taking on “everything customer-facing” except “what Sierra’s doing” – the release is a smoke signal that customer expectations for how fast agents can ship, update, and even self-improve, are going to continue to increase.
And owning that loop – plus working directly with the proprietary data that enables it – will increasingly separate valuable AI startups from the pack, argues Bavor.
“Everyone will rent the models, but we think people really need to own the information they have about their customers, and how they interact with them,” he says. (Given his co-founder Taylor is also chairman of OpenAI, there’s double reason to pay attention.)
More on what Ghostwriter does, and the wider implications, below.
Event alert: Upstarts is headed to SF Tech Week next week, and we’d love to see you at our next event. We’re hosting a tactical founder lunch panel and Q+A with three great founders – Ivan Burazin of Daytona, Sara Du of Ando, and Nick Rudder of Sphere – alongside our co-hosts Airwallex and Vanta on Tuesday, October 6.
We’re gearing this event toward early-stage founders and startup operators, and it’s free to attend. If you or someone you know would benefit, please apply to join us here.
Sierra’s new status quo
Ghostwriter has already had meaningful business impact for test customers, says Bavor, both in its classic customer service use cases and more revenue-focused ones.
“It’s one of the coolest things I’ve seen since starting Sierra,” he says. “There have been several head exploding moments where we saw it do something and were like, ‘This is clearly the future of how software should work.’”
In one instance, one of Sierra’s account leads was at dinner with a customer – Bavor calls them a “large travel company” – when Ghostwriter messaged them on Slack to flag issues with the company’s live service agent. It turned out that the customer had pushed an update that had broken an API. Triaging on their phones from the restaurant, the two were able to unwind the release and get the agent working again.
“It’s pretty neat knowing that there is a software agent looking over your shoulder and reading everything, analyzing what’s happening while you sleep, while you’re at dinner,” Bavor says.
In another scenario, Ghostwriter analyzed calls from one of its customers that works in subscriptions to provide three ideas on how it could improve retention. Ghostwriter simulated each, and was able to rank them in order of importance for the customer to try.
One feature that Sierra spent time trying to get right: how noisy Ghostwriter should be. The startup believes that the agent-builder it’s shipping will be able to learn from social cues, incorporate the feedback when no one is responding to it, and adapt to the level of input a certain team wants.
In other news: Upstarts recently ranked No. 2 on Inc.’s New Media Power Rankings! A big thanks to the founders and PR pros in our community for your support 🙏
Want to align your brand with one of the top trusted destinations for startup news? Drop us a line at partnerships [at] upstartsmedia [dot] com to set up a chat 👋
A mountain’s shadow
Our own human filters at Upstarts flags two takeaways worth bothering you about: one grounded in Sierra’s strategy, one in the wider moment for AI assistants.
From a company standpoint, these releases are pushing Sierra more into “cost savings” and revenue-facing areas of its customers, beyond its original focus on customer service. “An increasing percentage of what we do is directly connected with growing top-line revenue [for customers],” Bavor says. Revenue-aligned agents, then, are increasingly focused not just on being able to respond to service issues effectively and quickly, but on speaking up on their own – handling outbound, setting follow-ups, pursuing leads and even carrying out actions like processing a refinancing or making a subscriber a retention offer.
If you’re a startup building in revenue operations or sales, it’s probably wise to keep a close eye on how Sierra’s product migrates in this direction. Sierra’s enterprise focus might mean you have very little customer overlap today, but that might not be the case a year from now, as you grow, or Sierra moves earlier in the funnel. More broadly: while conversation volume with agents is up 500% year over year, as Bavor tells me, the nature of those conversations could be shifting to more agent-driven than human-prompted.
It’s no accident that in their own blog post about this release, Taylor and Bavor cite new assistant products Instinct and Muse as helping to spark an expectation of proactivity – agents spotting things we’ve missed, or offering solutions to problems we don’t know we have.
Increasingly, agents that customers value will need to be able to self-repair and self-improve. I’m seeing this with my own personal use of Instinct and Muse: they can execute little tasks I think up for them, like checking for tickets or a calendar conflict, but are stickier and more useful if they can regularly filter to me tips or productivity gains I’m not thinking of – where I’d say I currently have more mixed results.
Ghostwriter’s goal is to lower the barrier to building high-quality agents, Bavor says, but an increasingly big part of that is the speed of agent development. Bavor claims that Ghostwriter’s new release accelerates that cycle to be 5x or 10x faster. If he’s right, such gains will need to eventually become table stakes for everyone else.