April 22, 2026
PolyAI Launches Agent Development Kit to Give Developers Full Control Over CX AI
PolyAI has launched its Agent Development Kit (ADK), a new way for teams to build, deploy, and continuously improve AI agents for customer service. The company says the ADK removes the constraints that have kept developers locked inside vendor-controlled interfaces, giving them the freedom to work with their own tools, their own coding environments, and the rising generation of AI coding assistants.
The ADK hands developers full access to the codebase and lets them use AI-powered coding tools such as Cursor and Claude Code to generate and refine agent logic. Teams can feed in diagrams, spreadsheets, or API documentation and produce working agent flows in minutes rather than days.
Enterprise CX vendors are racing to deliver agentic AI, with Salesforce unveiling its native Agentforce Contact Centre and an increasing number of companies investing in voice-led AI agent capabilities across the industry. PolyAI’s pitch is that building AI agents should follow the same principles as building any other critical software, with code reviews, version control, and collaborative workflows baked in from the start.
Shawn Wen, co-founder and CTO at PolyAI, explained the thinking behind the ADK. He said: “Most AI platforms for CX force developers to work inside a UI, cut off from the way real software gets built. With the ADK, we’re changing that. Developers can now build AI agents with the same tools, workflows, and flexibility they use to build any other critical system.”
Speed as the Selling Point
Early ADK customers have already seen dramatic improvements in how quickly they can get agents up and running. One of the world’s largest utility companies built complex conversation flows in hours, replacing a process that previously took multiple weeks. Meanwhile, one of the UK’s largest banks processed hundreds of FAQs in minutes, a task that had consumed hours of manual effort.
The speed comes partly from the ability to reuse proven patterns that PolyAI has refined over more than 500,000 hours of real-world deployment across industries, including hospitality, utilities, banking, and healthcare. Rather than starting from scratch, developers can draw on tested approaches for integrations, conversation logic, and agent behaviour that already work at enterprise scale.
The company is also using the ADK internally to automate more than 60% of its own engineering work, with developers reviewing and guiding the output.
Agents that Keep Getting Better on Their Own
The ADK also introduces what PolyAI calls self-learning loops. Once an agent is live, it can connect to monitoring tools and automated quality checks, spot areas where it is falling short, and generate its own improvements without waiting for a developer to step in.
Traditionally, teams build an agent, deploy it, and then return periodically to tweak its performance. The ADK makes that cycle continuous and largely automatic, so agents learn from every customer interaction and evolve over time.
Most CX AI vendors have built their platforms around no-code and low-code tools for non-technical users, and PolyAI’s existing Agent Studio platform still serves that audience. The ADK adds a parallel track for engineering teams that want full code access, peer review workflows, and the ability to manage AI agents under the same standards they apply to any other enterprise system.
The distinction is particularly relevant in large organisations where compliance, security reviews, and internal governance demand complete visibility into how automated systems make decisions. In those environments, the ability to inspect and version every piece of agent logic is not optional.
