How Moda moved its entire engineering team from local IDEs to cloud agents
Moda (YC W26) builds a continual learning layer for AI agents. It reads production traces, finds where an agent went wrong across prompts, tools, workflows, memory, and model behavior, then tests fixes against previous runs. The team uses Replicas to run coding agents in the cloud, review pull requests, and automate recurring engineering work.
Visit Moda“Going from local to cloud coding agents was arguably as big of a jump as using coding agents in the first place.”
4.7k
cloud-agent jobs in July
500k+
minutes running coding agents in July
~2k
jobs launched by automations in July
100%
of the engineering team using Replicas over other IDEs
Local coding agents hit a ceiling
Moda’s engineers started with local coding agents. Claude Code, Codex, and Cursor were useful for one-off work, but every run still depended on an engineer opening a laptop and starting the job. That limited how much work the team could hand off, especially recurring maintenance.
The team also moved between interfaces to use different models and harnesses. Claude Code, Codex, and Cursor each tied Moda to its own interface, even as engineers switched between them for different strengths, such as Cursor’s speed and Codex’s performance on harder tasks.
Moda initially tried Cursor for cloud agents and a single interface across models, eventually running nearly all of its coding work through it. But the bill reached $120,000 a month, almost entirely from frontier-model API calls. Moda already had credits with OpenAI, Anthropic, Bedrock, and Azure, plus paid Claude Code and Codex subscriptions. Replicas let the team use those existing credits and subscriptions while running more important work in parallel.
Recurring work that runs itself
Replicas moved that work into scheduled and event-driven cloud agents. Moda now uses them to keep AGENTS.md files in sync with code changes, review new pull requests for ClickHouse query issues, audit ClickHouse usage each day, remove stale feature flags and commented-out code, and write weekly Linear updates.
Each automation runs in its own virtual machine, so it can start on a schedule or from an event without an engineer opening a laptop. Engineers also use the same cloud environments interactively for feature work and code review.
The outcome
In July, Moda ran 4.7k cloud-agent jobs through Replicas, totaling more than 500,000 minutes running coding agents. Nearly 2,000 jobs were launched by automations, making scheduled and event-driven agents Moda’s largest source of Replicas usage. The team used Replicas across thousands of pull requests.
Those agents keep AGENTS.md files current, review ClickHouse queries, watch for database cost anomalies, remove stale feature flags and commented-out code, and write weekly project updates. Recurring maintenance now starts on a schedule or when a pull request opens, without waiting for an engineer.
Moda’s entire engineering team has left other IDEs and now uses Replicas exclusively. Replicas has become the team’s default engineering environment, bringing interactive feature work, code review, and recurring maintenance into one place.
