Workflow · Production

Replit builds a multi-agent AI coding assistant with LangGraph and LangSmith

The problem

Building a fully functioning software app requires significant upfront setup — coding logic, environments, databases — creating a high activation barrier that leaves developers staring at an empty editor, a phenomenon Replit calls 'blank page syndrome'.

First attempt

A single-agent architecture increased error rates as one agent was responsible for all tools, and early experiments with fine-tuning failed to yield performance breakthroughs.

Workflow diagram · grounded in source
1
User submits plain English prompt
trigger
“allowing users to whip up a project with a simple prompt in plain English”
2
Manager agent coordinates roles
routing
“A manager agent to oversee the workflow.”
3
Editor agents perform coding tasks
ai_action
“Editor agents to handle specific coding tasks.”
4
Verifier agent checks and involves user
human_review
“Their verifier agent, for example, is unique in that it doesn't just check code and try to progress with a decision. It often falls back to talking to the user in order to enforce continuous user feedback in the development process.”
5
Auto-commit enables user reversion
feedback_loop
“At every major step of the agent's workflow, Replit automatically commits changes under the hood. This lets users "travel back in time" to any previous point and make corrections.”
6
LangSmith trace observability
integration
“Replit integrated LangSmith as their observability tool to track and action upon problematic agent interactions in their traces.”
7
One-click application deployment
output
“Replit also lets you deploy your application in a few clicks. The ability to publish and share applications is integrated smoothly in the agent workflow.”
Reported outcome

Replit Agent allows users to create applications from a plain English prompt with multi-step execution and infrastructure management, with significant performance improvements achieved by adopting Claude 3.5 Sonnet.

Reported metrics
performance improvement from Claude 3.5 Sonnetsignificant performance improvements
tools in Replit's library30+
Alpha testers~15
Reported stack
LangSmithLangGraphClaude 3.5 SonnetGit
Source
https://www.langchain.com/breakoutagents/replit
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Replit Agent allows users to create applications from a plain English prompt with multi-step execution and infrastructure management, with significant performance improvements achieved by adopting Claude 3.5 Sonnet.

What tools did this team use?

LangSmith, LangGraph, Claude 3.5 Sonnet, Git.

What results were reported?

performance improvement from Claude 3.5 Sonnet: significant performance improvements; tools in Replit's library: 30+; Alpha testers: ~15 (source-reported, not independently verified).

What failed first in this deployment?

A single-agent architecture increased error rates as one agent was responsible for all tools, and early experiments with fine-tuning failed to yield performance breakthroughs.

How is this workflow AI workflow structured?

User submits plain English prompt → Manager agent coordinates roles → Editor agents perform coding tasks → Verifier agent checks and involves user → Auto-commit enables user reversion → LangSmith trace observability → One-click application deployment.