# How Airbyte uses Devin to keep 700+ connectors running

![Vimeo](https://devin.ai/_next/image?url=%2F_next%2Fstatic%2Fimmutable%2Fmedia%2Fcover.0mg2au9n00bxt.webp\&w=3840\&q=75)

2x

Faster average time to resolve support tickets

990

Engineering issues triaged by Devin in 90 days

700+

Integrations whose docs are kept up-to-date by Devin

## About the company

[Airbyte](https://airbyte.com/) is an open-source data replication company that helps businesses move data between their applications, databases, and data infrastructure, and connect AI agents to their business systems.

Industry: Data Infrastructure**[Visit site](https://airbyte.com/)**

## Maintaining 700+ connectors as third-party APIs change

Airbyte gives companies a common way to move and access data across hundreds of business systems. Instead of building and maintaining their own integrations with Salesforce, Stripe, Snowflake, Slack, and other providers, customers can use Airbyte's connectors as a shared integration layer.

But Airbyte doesn't control the systems on the other side. Providers continually release new API versions, deprecate endpoints, and change how their systems behave. Each change can require Airbyte to update a connector, investigate a customer issue, or revise its documentation.

"We have 700 things that could break at any time, even if they were working yesterday," said AJ Steers, a staff software engineer at Airbyte.

Airbyte uses Devin to handle more of that ongoing work — from investigating and fixing connector issues to resolving support tickets and keeping documentation current.

## Investigating and fixing bugs automatically

Airbyte has built an automated bug-resolution pipeline with Devin.

For incoming engineering issues, Devin first tries to reproduce and triage the problem, then assigns a confidence score from one to five based on how fixable it appears. Issues that score a 4 or 5 move automatically into a new Devin session that attempts the fix and opens a pull request.

A separate Devin session then verifies the proposed fix, gathering evidence that the PR resolves the issue without breaking anything else.

Over the last 90 days, Devin triaged \~990 issues, drafted 413 fixes, and ran 467 verification sessions. About 49% of triage sessions score 4 or 5.

For PRs created through these triage-and-fix workflows, Airbyte ultimately merges roughly 50–60%. Most still receive human review, but for narrow categories where the workflow is already well understood, Devin can merge automatically — for example, dependency version bumps for security issues. Steers estimates that 5–10% of these PRs currently merge without human review.

Airbyte also gives Devin access to internal systems through its own MCP servers. airbyte-ops-mcp alone exposes 81 constrained tools, with human approval required in Slack for certain actions.

## Giving engineers a head start on on-call

Devin can begin investigating issues before an engineer picks them up, reproducing the problem, identifying a likely root cause, and in some cases opening a pull request for review.

"It shortens the amount of time that they have to spend looking at easier or more trivial or more recurring kinds of issues," said Pedro Lopez, a software engineer at Airbyte, "and really focus more of their time on the more complex issues that are harder to reproduce or actually need more engineering time."

## Investigating support tickets around the clock

The same underlying connector problems can first appear as customer-reported support tickets rather than engineering issues.

Airbyte's support team uses Devin directly inside Zendesk. Devin can classify and tag a ticket, identify related or duplicate issues, create a bug or feature request for engineering, and help technical support staff diagnose the problem and determine next steps. Devin handles the investigation and technical analysis; the support team still owns the customer-facing response.

Airbyte's Devin-powered support workflows ran more than 1,600 times in the last 90 days.

Before Airbyte began rolling out this tooling in February 2026, roughly 50% of support tickets received a first response within 24 business hours. By April, that had risen to more than 95%.

Airbyte has also roughly halved its average time to resolution on support tickets.

Because Airbyte supports customers across regions, Devin can investigate issues around the clock rather than waiting for the relevant support team to come online. Steers pointed to one Devin investigation that started at 2:05am.

## Updating documentation as the product changes

Every time Airbyte merges a connector change, Devin checks whether the customer-facing documentation needs to change with it.

For example, when Airbyte updated its Greenhouse connector to use the latest version of Greenhouse's API, Devin compared the changed connector code, the third-party API, and Airbyte's existing documentation. It then assessed the risk and confidence of the proposed documentation update and distinguished claims it could directly verify from anything it had to infer.

High-confidence documentation updates can ship without human review. From June through August, 62% of documentation changes auto-merged.

Lower-confidence changes are routed to Ian Alton, Airbyte's head of technical writing, or the relevant engineer.

Alton is the only person dedicated full-time to this kind of content, while roughly 600–700 connectors can change frequently. He said there would be "essentially no way" to keep the documentation current manually without Devin. Devin helps him ensure the quality of documentation stays high.

Airbyte also runs a separate Devin process that monitors third-party providers for API deprecations, catching external changes that may require connector updates.

With routine documentation maintenance handled by Devin, Alton spends more of his time on product work and research.

## Finding blockers before committing to feature work

Steers also uses Devin to explore new feature work before committing engineering time.

"Devin, here's a hard problem, let's talk about it," Steers said. "And within a day I generally have a PR that either does the thing, or does 50 or 70 percent of the thing and then finds the unblockers that otherwise would have been weeks of investigation to find."

He starts these explorations early and runs several in parallel. The goal isn't necessarily to produce the final implementation. It's to find out what's easy, what's hard, and what the team doesn't yet know.

"The way that I find out what I don't know is just try to have Devin build it and see where it gets stuck," Steers said.

Sometimes a feature that sounds difficult turns out to be relatively easy. Other times, Devin surfaces a refactor or architectural constraint that wasn't obvious from the request.

"Sometimes the cool features we can ship are actually low hanging," Steers said. "We just don't realize they are, or there's some blocker."

As Airbyte expands its use of Devin, the team wants to push further in this direction: using agents not only to automate recurring work, but to explore harder engineering problems earlier and give engineers more information before deciding how to approach them.