Were chatbots a half-baked concept pushed to market too quickly? According to a February article in The Information, chatbots on Facebook Messenger could fulfill only “about 30 percent of requests with human agents” during testing. If that data is accurate, it’s a big problem for Facebook, which spent 2016 playing up the utility of chatbots to businesses of all types. That mediocre in performance, in turn, has reportedly kicked off an internal repositioning of bots. “Messenger is trying to find ways to inject buttons into conversations between friends when it thinks they intend to do things like make restaurant reservations,” the publication added. There are other signs that the environment for bots has roughened. Online retailer Everlane, one of Facebook’s initial launch partners, has discontinued the use of bots as a notification pool. “It was a good couple of years, but we’ve decided to stick with what we do best—email,” the company told its users in an email. Last week, Facebook also tweaked Messenger’s bot functionality with a new, persistent menu structure (giving users easier access to all of a bot’s particular features), additional social-sharing options, longer text limits for messages, and a variety of APIs including attachment uploads. Another tool gives developers better insight into bot engagement, including threads, block, delete, and report rates. All platforms evolve, often after some early stumbles. Speaking to an audience at TechCrunch Disrupt last December, Facebook Messenger head David Marcus even admitted that bots hadn’t yet proven “good enough to basically replace traditional app interfaces and experiences.” (He also suggested that bots might become “really overhyped, very, very quickly.”) But considering the amount of well-funded competition in the A.I. and machine-learning arena, Facebook has no choice but to radically improve its bot platform as fast as possible. Tech firms from Microsoft to Atlassian have all broadcast their desire to dominate the bot market. If developers and other tech pros don’t think Facebook’s platform has enough to offer (despite the early hype), they’ll easily turn to other options—or, as in Everlane’s case, abandon the technology entirely. As for those developers who remain cautious about bots, it may pay to tread carefully for the time being; while the technology offers substantial promise when it comes to tasks such as automating customer service, it’s nowhere near perfected.