← All implementations

Ramp · Background agent

Inspect

A background coding agent that closes the loop on verifying its own work; runs tests, reviews telemetry, queries feature flags, visually verifies the frontend; now also monitoring production and proposing fixes; also a platform that hosts many internal agents.

1 Supports4 Supports5 Supports

Approach type
Background agent
Work
Coding, Code review, On-call
Human involvement
Drafts reviewed
Invocation
Background, Event driven, Interactive
Interfaces
Slack, Web, Chrome extension, Github
Deployment stage
Scaled
Evidence strength
Mixed
Entry reviewed

How it works

The workflow the sources report for this implementation.

Pre-warmed full dev envs; fast cold start; effectively free to run

1

Server-first agent with a typed SDK + plugin system; code is its own source of truth

1

Any number of people in one session; each change attributed to its author

1

Where people stay involved

  • Inspect coding task → reviewed production mergeWork product review · Level 3

Level 3 for Inspect coding task → reviewed production merge; human attention boundary: work-product-review.

1

Observation date
2026-05

Implementation details

Sandbox

Modal sandboxes; per-repo images rebuilt every 30 min from snapshots; warm-on-keystroke; a pool of warm sandboxes

1

Harness

OpenCode (server-first) as the agent runtime; a plugin blocks writes until sync completes; expanded into production monitoring and self-maintenance; Inspect itself is built with React/Vite, Cloudflare Durable Objects, SQLite, and the Cloudflare Agents SDK

1

Model

All frontier models, MCPs, custom tools, skills

1

Interfaces

slack, web, chrome-extension, github

1

Tool access

Wired into Sentry, Datadog, LaunchDarkly, Braintrust, GitHub, Slack, Buildkite; monitors production, triages issues, proposes fixes; debugging queries a sanitized read-only production DB replica and Snowflake

1

Knowledge

Skills that encode how Ramp ships; repo images with the full dev env (Vite, Postgres, Redis, RabbitMQ, Temporal, Chromium, VS Code Server)

1

Credentials

GitHub auth per user; the sandbox pushes the branch, an API opens the PR with the user's token (no self-approval); production merges retain human review

1

Reported results and limitations

The catalog records what the sources report, with the scope and the denominator of every figure. A qualification below limits the figure it sits under.

Reported metrics

Headline claim

75% of Ramp's merged PRs raised by Inspect sessions (May 2026)

5

Reported by
Ramp
Scope
Ramp merged PRs raised by Inspect sessions by May 2026
Denominator
Merged Ramp pull requests
Observation date
2026-05
Key observation

Around 60% of Ramp PRs authored by Inspect by January 2026

5

Reported by
Ramp
Scope
Ramp PRs authored by Inspect by January 2026
Denominator
Ramp pull requests in the source adoption history
Observation date
2026-01
Key observation

75% of merged PRs raised by Inspect sessions (May 2026)

2 Supports5 Supports

Reported by
Ramp
Scope
Ramp merged PRs raised by Inspect sessions by May 2026
Denominator
Merged Ramp pull requests
Observation date
2026-05
Key observation

Around 30% of merged frontend and backend PRs in the earlier first-party report, after a couple of months of adoption

1

Reported by
Ramp
Scope
Merged PRs in Ramp frontend and backend repositories after the first couple of months of adoption
Denominator
Merged PRs in frontend and backend repositories
Key observation

Around 90% of PRs merged into the Inspect repository come from Inspect sessions in the later interview report

5

Reported by
Ramp
Scope
PRs merged into the Inspect repository
Denominator
Merged PRs in the Inspect repository
Key observation

One million total Inspect sessions crossed in July 2026

5

The source does not report the denominator of this figure.

Reported by
Ramp
Scope
Cumulative Inspect sessions crossing the million mark in July 2026
Observation date
2026-07
Key observation

Under 5 seconds to spin up a fully provisioned remote dev environment

5

The source does not report the denominator of this figure.

Reported by
Ramp
Scope
Provisioning a fully configured remote development environment
Key observation

150+ engineers contributed to the Inspect codebase in the later interview report

5

The source does not report the denominator of this figure.

Reported by
Ramp
Scope
Engineers who contributed to the Inspect codebase in the later interview report
Key observation

5.5-person Inspect team (four engineers, a director, and a part-time PM)

5

The source does not report the denominator of this figure.

Reported by
Ramp
Scope
Inspect team staffing: four engineers, one director, and a part-time PM
Key observation

More than 80% of Inspect is written in Inspect sessions

5

Reported by
Ramp
Scope
Inspect code written in Inspect sessions; distinct from merged PR share
Denominator
Inspect code; the source does not define a code-volume counting method

Reported outcomes and statements

Key observationFact

Inspect underpins internal agents including ReviewBuddy, Oncall Assistant, Testo, Ramp Research, Voice of the Customer, and error automations

5

Scope
Examples of internal agents built on Inspect
Key observationOpinion

Design goal: session speed should be limited only by model-provider time-to-first-token

1

Scope
Target session startup speed, excluding precompleted cloning and installation

Lessons and interpretation

Own the tooling; it only has to work on your code, which lets you build something more powerful than off-the-shelf

1

Work in public spaces to create virality loops; let the product do the talking, don't mandate

1

Ramp argues that a fast background agent can add remote resources and concurrency to the same model

1

Move as much as possible into the image-build step so users never wait on setup

1

The v1 Chrome extension saw little adoption; the pivot to a centrally configured remote dev environment with a coding agent on top drove adoption

5

Sources and research details

Citations link to the original publisher. Each source also keeps a preserved copy in the repository, so a changed or removed page stays checkable.

  1. Why We Built Our Own Background Agent (Inspect)https://engineering.ramp.com/post/why-we-built-our-background-agentEngineering blog · First party · Last source verification: 2026-08-31
    Preserved copy ↗
  2. Ramp x Linear (75% of merged PRs via Inspect)https://linear.app/customers/rampCase study · First party · Last source verification: 2026-08-31
    Preserved copy ↗
  3. Hacker News project discussion inspired by Ramp Inspecthttps://news.ycombinator.com/item?id=48042123Hn thread · Community · Last source verification: 2026-08-31
    Preserved copy ↗
  4. Why We Built Our Own Background Agent (Inspect) (Ramp Builders URL)https://builders.ramp.com/post/why-we-built-our-background-agentEngineering blog · First party · Last source verification: 2026-08-31
    Preserved copy ↗
  5. Why Ramp built its own in-house coding agent, Inspecthttps://newsletter.pragmaticengineer.com/p/why-ramp-built-inspectNews · Independent secondary · Last source verification: 2026-08-31
    Preserved copy ↗
Research details for every claim on this page
  1. Summary
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  2. Headline claim
    Statement type
    Metric
    Provenance
    Reported
    Confidence
    Medium
    Confidence reason
    A linked participant or independent source reports the claim.
    Reported by
    Ramp
    Scope
    Ramp merged PRs raised by Inspect sessions by May 2026
    Denominator
    Merged Ramp pull requests
    Method
    Not reported
    Observation date
    2026-05
  3. Sandbox
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  4. Harness
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  5. Model
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  6. Interfaces
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  7. Tool access
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  8. Knowledge
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  9. Credentials
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  10. Supporting component
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  11. Supporting component
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  12. Supporting component
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  13. Key observation
    Statement type
    Metric
    Provenance
    Reported
    Confidence
    Medium
    Confidence reason
    A linked participant or independent source reports the claim.
    Reported by
    Ramp
    Scope
    Ramp PRs authored by Inspect by January 2026
    Denominator
    Ramp pull requests in the source adoption history
    Method
    Not reported
    Observation date
    2026-01
  14. Key observation
    Statement type
    Metric
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
    Reported by
    Ramp
    Scope
    Ramp merged PRs raised by Inspect sessions by May 2026
    Denominator
    Merged Ramp pull requests
    Method
    Not reported
    Observation date
    2026-05
  15. Key observation
    Statement type
    Metric
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
    Reported by
    Ramp
    Scope
    Merged PRs in Ramp frontend and backend repositories after the first couple of months of adoption
    Denominator
    Merged PRs in frontend and backend repositories
    Method
    Not reported
    Observation date
    Not reported
  16. Key observation
    Statement type
    Metric
    Provenance
    Reported
    Confidence
    Medium
    Confidence reason
    A linked participant or independent source reports the claim.
    Reported by
    Ramp
    Scope
    PRs merged into the Inspect repository
    Denominator
    Merged PRs in the Inspect repository
    Method
    Not reported
    Observation date
    Not reported
  17. Key observation
    Statement type
    Metric
    Provenance
    Reported
    Confidence
    Medium
    Confidence reason
    A linked participant or independent source reports the claim.
    Reported by
    Ramp
    Scope
    Cumulative Inspect sessions crossing the million mark in July 2026
    Denominator
    Not reported
    Method
    Not reported
    Observation date
    2026-07
  18. Key observation
    Statement type
    Metric
    Provenance
    Reported
    Confidence
    Medium
    Confidence reason
    A linked participant or independent source reports the claim.
    Reported by
    Ramp
    Scope
    Provisioning a fully configured remote development environment
    Denominator
    Not reported
    Method
    Not reported
    Observation date
    Not reported
  19. Key observation
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    Medium
    Confidence reason
    A linked participant or independent source reports the claim.
    Scope
    Examples of internal agents built on Inspect
  20. Key observation
    Statement type
    Metric
    Provenance
    Reported
    Confidence
    Medium
    Confidence reason
    A linked participant or independent source reports the claim.
    Reported by
    Ramp
    Scope
    Engineers who contributed to the Inspect codebase in the later interview report
    Denominator
    Not reported
    Method
    Not reported
    Observation date
    Not reported
  21. Key observation
    Statement type
    Metric
    Provenance
    Reported
    Confidence
    Medium
    Confidence reason
    A linked participant or independent source reports the claim.
    Reported by
    Ramp
    Scope
    Inspect team staffing: four engineers, one director, and a part-time PM
    Denominator
    Not reported
    Method
    Not reported
    Observation date
    Not reported
  22. Key observation
    Statement type
    Metric
    Provenance
    Reported
    Confidence
    Medium
    Confidence reason
    A linked participant or independent source reports the claim.
    Reported by
    Ramp
    Scope
    Inspect code written in Inspect sessions; distinct from merged PR share
    Denominator
    Inspect code; the source does not define a code-volume counting method
    Method
    Not reported
    Observation date
    Not reported
  23. Key observation
    Statement type
    Opinion
    Provenance
    Reported
    Confidence
    Medium
    Confidence reason
    The source says session speed should only be limited by model time-to-first-token; this is a design goal, not a measured result.
    Scope
    Target session startup speed, excluding precompleted cloning and installation
  24. Lesson
    Statement type
    Inference
    Provenance
    Catalog judgment
    Confidence
    Medium
    Confidence reason
    The catalog derives this observation from the linked sources.
  25. Lesson
    Statement type
    Inference
    Provenance
    Catalog judgment
    Confidence
    Medium
    Confidence reason
    The catalog derives this observation from the linked sources.
  26. Lesson
    Statement type
    Inference
    Provenance
    Catalog judgment
    Confidence
    Medium
    Confidence reason
    The catalog derives this observation from the linked sources.
  27. Lesson
    Statement type
    Inference
    Provenance
    Catalog judgment
    Confidence
    Medium
    Confidence reason
    The catalog derives this observation from the linked sources.
  28. Lesson
    Statement type
    Inference
    Provenance
    Catalog judgment
    Confidence
    Medium
    Confidence reason
    The source reports that v1 saw little adoption and that the November 2025 pivot to a remote dev environment preceded rapid adoption.
  29. Operating model assessment
    Statement type
    Inference
    Provenance
    Catalog judgment
    Confidence
    High
    Confidence reason
    The source explicitly states that Inspect cannot self-approve and production merges retain human review.
    Observation date
    2026-05

The catalog records no related implementation for this entry yet.

Back to the directory →