AI Dev Meetup on Coding Agents with OpenAI and LangChain
Last Tuesday, we kicked off our first AI developer meetup of 2026 with a packed room and over 350 signups! This was our first content-focused event since organizing AI Engineer Paris 2025, and it was a great night bringing the AI dev community together to share ideas and learn from some of the most exciting builders in the space.
Want to join next time? Follow our global events calendar to stay in the loop.
Our meetup's theme was coding agents. We heard from speakers at Koyeb, OpenAI, and LangChain. Talks covered everything from agent-driven code generation to sandbox execution and reliability. Below is a recap of the night's talks.
Jen Person – Developer Relations Engineer at Koyeb
Lessons Learned from Building a Lovable Clone App
Jen walked us through how she built a Lovable clone, an app designed to generate website frontends and display previews by running the generated code in ephemeral environments. Why Lovable? Because Lovable leverages sandboxes to generate frontend applications based on user prompts, making it a great subject for testing the Koyeb Sandbox SDKs. Koyeb's sandbox environments run on CPU and GPU Instances, making them a great choice for running CPU workloads like AI app generation, AI agents, reinforcement learning, and model training.
Generating AI code is efficient, but also inherently risky without safeguards. AI models can and will make [dangerous mistakes}(https://www.reddit.com/r/ClaudeAI/comments/1pgxckk/claude_cli_deleted_my_entire_home_directory_wiped/), and this is a core challenge of the AI development today. To mitigate risk, Jen suggested running and testing AI generated code in a sandboxed environment like Koyeb Sandboxes.
Her key lessons from building a Lovable clone app:
- Using the sandbox is the easy part
- Use a custom Docker image to save setup time
- Code everything where possible
- Start basic, and build in explicit instructions where needed
- This level of task complexity requires at least a 30B parameter model. Her Lovable clone leverages Qwen 3 Coder 30B A3B Instruct
Jen’s talk was a reminder that AI coding is fun right up until it deletes something important. Her Lovable clone demo showed why sandboxes are the safest way to let models experiment, fail, and iterate.
Spin up a Koyeb sandbox in seconds. Run AI code in isolated environments.
Marco Perini – Forward Deployed Engineer at LangChain
From ReAct to Deep Agents
Marco discussed the evolution of AI agents, from traditional ReAct approaches to LangChain’s Deep Agents framework. He explained how agents often struggle with reliability when tasks are ambiguous, context is missing, there are too many tools, or they lose track of conversation history.
Marco also traced the evolution of development practices from prompt engineering → context engineering → agent engineering, showing how leveraging a filesystem as part of the agent environment can enhance performance by offloading part of the context into external files or backends.
He then ran a demo using the deepagents-cli package to create a coding agent that generates and runs code both locally and in a sandboxed environment, and showed how to monitor and debug the agent using LangSmith.
Try out the new LangSmith Agent Builder, built on top of the deepagents package.
Build agents that can plan, use subagents, and leverage file systems for complex tasks.
VB Srivastav – Developer Experience at OpenAI
Codex 101
VB introduced the audience to OpenAI’s Codex, diving into how it can accelerate coding workflows and help developers build smarter applications faster. He demonstrated practical examples of using Codex to generate, complete, and refactor code, showing how it integrates seamlessly into real-world developer environments.
The talk also highlighted best practices for prompting Codex effectively, how to avoid common pitfalls, and how developers can leverage the model’s strengths while keeping control over output quality.
Check out how to get started with Codex.
Start using Codex to write features, answer questions about your codebase, fix bugs, and propose pull requests for review.
Paris AI Developer Community
We had an amazing time getting the AI community together! We are thankful to everyone who attended and to all our awesome speakers for getting on stage to share their insights.
If you want to attend any of our future events, subscribe to our global events calendar. You can also follow us on X @gokoyeb and join our Slack community to stay in the loop about our next events.
Want to partner with us for a future event? Reach out to us and join our growing Partner Hub!