Three years (plus) after the arrival of ChatGPT, chatbots are morphing into AI agents. As generative AI models have improved and become able to reason in real time, the major AI labs, starting with Anthropic, have begun to shift their research focus from models that compose and comprehend text to ones that reason, use tools, and work autonomously.
The first kind of agent that matured to the point of having real-world impact was an agent that can write, test, and document computer code. Coding agents, powered by language models, can understand plain language, which has democratized software development and made “vibe coding” possible. Products like Lovable and Bolt allow nontechnical or semi-technical product managers or marketing chiefs to quickly mock up working prototypes of apps or site features. Both products saw significant gains in both users and revenue over the past year. As did Cursor, which works alongside human software engineers in their normal user interface to build within large preexisting codebases. Anthropic’s Claude Code and OpenAI’s Codex also saw surges in usership in the last half of 2025.
Customer service was another early application of AI agents, but many have been limited in their knowledge and reasoning ability, and therefore limited in scope and utility. That too is changing. Sierra, which was founded by ex-Salesforce exec (and OpenAI board member) Bret Taylor and ex-Google hardware chief Clay Bavor, is developing agents that act like all-purpose concierges or “brand representatives.” Sierra agents remember a consumer’s past interactions with the company and are versatile enough to handle a range of tasks including product returns, account updates, subscription issues, and appointment setting. Cognigy (now a part of Nice) formally launched its AI agent platform in 2025; Its customer service AI agents work with human operators to plan and carry out tasks, such as handling complex transactions and coordinating across systems.
ServiceNow, which provides a sort of operating system for customer support, is broadening its platform to enable the deployment and management of AI agents (both its own and those from third parties) across many departments across the enterprise including customer service, human resources, and IT. As AI agents proliferate, governance of the agents will become a big priority for enterprises. That’s why Credo AI’s Agent Registry is timely; the system gives enterprises a way to register all agents used by the company, as well as real-time oversight of the actions agents are taking, what data they’re accessing, and how they’re arriving at their decisions.
In the first half of 2026, coding agents and customer service agents have already begun to reshape how organizations manage—and staff—those key business functions. By this time next year, we may look back at those applications as just the low-hanging fruit, the first of many to come.
1. Sierra
For developing more thoughtful agents for business
Most AI agents in customer service still behave like glorified chatbots—fast, yet they fail to retain conversational context. Sierra, which is building AI agents that function less like ticket responders and more like long-term brand representatives, is focused on a core challenge in enterprise AI: memory. The company was founded in 2023 by Bret Taylor, who currently chairs OpenAI’s board and previously co-created Google Maps before serving as CTO of Facebook and co-CEO of Salesforce, and Clay Bavor, who spent 18 years at Google leading AR and VR initiatives and overseeing product and design for Workspace apps such as Gmail and Docs.
