OpenAI released a toolbox this week to help developers orchestrate applications and achieve some of AI’s oft-promised impact.
The toolbox, called AgentKit, allows developers to create self-directed assistants — “agents” in AI vernacular — that plan and execute tasks.
AI Agents
An AI agent is more than a chatbot. It can act on a defined goal rather than responding passively to instructions. A good agent can, in a sense, decide what to do, access external tools, and learn from the results.
For ecommerce, an agent could review Shopify sales data, identify products with slow sell-throughs, and create Google Ads campaigns to move the inventory.
Certainly that process was doable before AI agents, with automations and coordinated prompts. But agents offer a more structured solution. An agent maintains context through each step and likely operates more efficiently, consuming relatively fewer AI tokens and reducing overall compute costs.
Using AgentKit
AgentKit packages several capabilities into a single development environment.
It features Agent Builder, which helps developers define what an agent should do and how it should behave. The Connector Registry manages access to tools and data sources, including analytics software, application programming interfaces, and product databases.
AgentKit includes what it calls ChatKit as the interface layer, making it easier to embed conversational AI into existing apps and websites. AgentKit also helps enforce safety, privacy, and performance standards.
In a sense, AgentKit functions like an operating system for AI assistants. Although it doesn’t manage resources or run processes like an OS, AgentKit transforms the idea of “prompt once and hope for the best” into repeatable, structured workflows. Tasks that required multiple applications and manual coordination are now manageable by a single agent, configured on company guidelines and policies.
Frameworks
AgentKit exemplifies a broader trend toward structured, autonomous AI systems.
The trend started before ChatGPT’s public release. Relatively early frameworks like LangChain and Google’s Vertex AI paved the way for orchestrating multi-step AI processes.
Newer approaches, such as Anthropic’s open-source Model Context Protocol, are establishing standards for how agents securely connect to data and tools.
These frameworks and standards aim to transform general-purpose AI into practical and reliable assistants for businesses.
Building Agents
For an ecommerce executive, AI tools and software can feel both imminent and distant. She can imagine helpful AI tasks without knowing how to implement them.
AgentKit, frameworks, and standards can unlock that second part.
Some ecommerce companies might build relatively complex, automated systems. I’m writing this article, for instance, during my employer’s week-long off-site, where a team of developers and leaders is building an AI system to identify Meta advertising opportunities, generate creatives, and launch campaigns.
The system will act autonomously, optimizing performance and driving sales.
AI in Tools
To be sure, most ecommerce businesses will not build AI agents from scratch using AgentKit or any similar system. Rather, most will rely on tools they already use to include them.
Hence the work for retailers now is preparation.
First, identify automation candidates: repeatable tasks and decisions. Then organize your data. The cleaner the data, the more useful are future AI systems.
Defined rules and guidelines for agent behavior are equally key. Such rules form the structure for AI execution, ensuring that agents act within boundaries, such as reading a product catalog but not altering prices, or sending an email only with human-approved copy.
As more systems integrate these standards, trust and accountability will separate effective automation from risky experiments.
Ecommerce Implication
OpenAI’s AgentKit is an early operational milestone. The technology is nascent, but its direction is clear.
Businesses will move from testing and dabbling with generative AI to deploying autonomous systems that manage repeatable processes under human oversight.
AI agents will handle daily routines, freeing humans to focus on strategy and product development.