Introducing Trendyol AI Agent Platform

by Huseyin Ulaş

AI Agents orchestrates multiple specialized agents working together in shared workspaces. Instead of one agent doing everything, teams build workflows by combining agents with different expertise like how human teams distribute work. Agents share context, delegate tasks, and integrate with tools to automate complex workflows.

Agents

Agents are specialized AI assistants with specific skills and capabilities. A coding agent uses GitLab to write code, while an analytics agent processes database queries. Each agent can be configured with system prompts, knowledge bases, custom tools and MCP servers.

Workspaces

Workspaces bring multiple agents together for specific automations. A host agent manages conversations and delegates to specialized agents, all sharing the same context. Workspaces transform individual agent capabilities into coordinated team workflows.

What is Trendyol AI Agents Platform?

Trendyol AI Agents is an internal enterprise AI platform designed to empower Trendyol teams with intelligent automation and AI-powered workflows, enabling our employees to focus on high-value strategic work while AI handles routine tasks.

We’re building a platform where any team at Trendyol can leverage AI automation without requiring technical expertise.

Technology teams create coding assistants and automated reviewers integrated with GitLab and Jira to automatically generate complex database queries from natural language, write unit tests to increase code coverage, refactor legacy systems, create test scenarios, and many, many more.

Business teams build agents for growth analysis, customer insights, and content creation, drafting and personalizing push notifications,creating jira tasks from analyzing support requests and much more.

The best part is that teams can share their agents and workflows. When growth creates a lead analysis agent, marketing can clone and adapt it for campaigns. This shared intelligence multiplies productivity across the organization.

How to Use Our Platform to Build & Run AI Agents

Here’s how you can go from idea to a working AI agent in just a few steps:

1. Create your agent — Define its name, description, and system prompt to shape how it thinks and acts.

1. Set up your workspace — Organize your environment and select which agents your team can access and collaborate with.

1. Using your automations — Chat with it directly through our interface or connect it to your existing tools and integrations for real-world use.

1. Create Your Agent

An agent consists of three key parts: name, description, and system prompt that guide its actions.

To make your agents smarter and more effective, you can enhance them with powerful integrations — including MCP Servers, Custom Tools, and a Knowledgebase.

• MCP Servers enable seamless connections with platforms like GitLab, Jira, Figma, BigQuery, Swagger, and our 80+ internal MCP servers.

• Custom Tools let you define HTTP endpoints your agent can call directly by specifying the URL, method, and parameters.

• Integrating a knowledgebase empowers your agent with Retrieval-Augmented Generation (RAG) capabilities. This means your agent can search over relevant information from our internal sources.

2. Creating Workspace

Creating a workspace follows a simple process. Users select the agents required for their specific automation needs, choosing from specialized agents designed for tasks such as coding, data analysis, or documentation generation etc.. The platform then deploys the workspace where the selected agents operate collaboratively.

3. Using your automations

Our platform lets you interact with your AI agents through the built-in chat interface or trigger them externally via GitLab, Jira, Slack, and webhooks. Here’s a closer look at each option.

Get Hüseyin Ulaş’s stories in your inbox

Workspace chat is straightforward — a text input and conversation thread. Responses stream in real-time, so you see text as it’s generated. The workspace maintains full conversation context, meaning agents understand the entire thread, not just your latest message.

Jira Integration: Automate task analysis and coding by triggering workspaces from Jira tasks by dragging the task to READY FOR AI column.

GitLab Integration: Trigger your coding workspaces directly from merge requests. When you need AI assistance with code review or improvements, mention the AI agent in your merge request comments.

Slack Integration: Mention @AI Agents in any Slack channel, followed by your question. The bot will detect your message and run the corresponding workspace automatically.

Webhook Integration: We also have support for calling the AI agents via HTTP requests. Teams can send requests to our agent service by getting a personal access token.

Session History: Your Workspace Activity Log

When you need to dig into a complex task and see exactly what your agent did, Session History is your best debugging tool. It shows every action the workspace took, which agents were involved and any errors that occurred.

Impact by Numbers: Platform Statistics

During this 3 months period, we have had the opportunity to evaluate its adoption by our teams and its impact on our business processes using objective data. Our goal here is to transparently share the initial results of this adaptation process and the tangible outcomes observed.

Platform Usage and Team Adoption

The initial adoption of the platform has provided valuable insight into our teammates engagement.

Performance and Success Metrics

To measure tangible output, we integrated the agents with our Jira workflows and assigned specific tasks to them directly.

A total of 800 story points worth of tasks were assigned to our AI agents by our teams. Those tasks were successfully completed by the agents, which translates to a success rate of 84%.

To provide more detail on this metric: the 84% success rate represents the total volume of work completed. For instance, if an agent successfully handles 4 story points of a 5-story-point task, we count those 4 completed points toward this success metric.

One of the most valuable insights from our initial launch is understanding where our AI agents perform best. The data shows: the lower the story points, the higher the agent’s success rate.

Our agents performs well at detail defined, predictable tasks. By identifying this, we assigned the right kind of work to the AI. This doesn’t just boost the agent’s success rate; it frees up our talented developers from repetitive tasks, allowing them to dedicate their time and energy to solving the more complex challenges.

Tasks initiated and handled by our AI agents were completed 25% faster on average compared to those managed through manual methods. This reduction in cycle time means we are delivering value sooner.

Cycle Time: It’s the total time a task takes from the moment work actively begins on it until the moment it is fully completed. A shorter cycle time is a direct indicator of a faster, more efficient workflow.

Conclusion

As a data-driven team, we believe that what gets measured gets improved. Even in its pilot stage, the platform has demonstrated clear, measurable benefits to our working way. Looking ahead, we are committed to relentlessly enhancing this platform with the latest advancements in LLMs and new technologies. Our goal is to continue pushing the boundaries of productivity and empowering our teams to achieve more.

As agentic-experience team, we’re grateful to the many people across the Trendyol who contributed their expertise, time, and support to make this possible.