LoginSign Up
← Back to all sparks
S

Sourcegraph

AI-ASSISTANTS
Velocity6.3

Code intelligence platform with the Cody AI coding assistant for navigating and editing large codebases.

sourcegraph.com ↗

Sourcegraph is repositioning code search as agent infrastructure, and benchmarking to prove it.

agent-infrastructurecode-searchmigrationsprovenanceenterprisebenchmarks
◆Current state
The feed is mostly positioning essays, but two real launches sit inside it: Code Finder in July and Agentic Batch Changes entering public beta in June. Both are sold to coding agents rather than to engineers reading results themselves. The essays around them argue the same case from three angles: retrieval quality, migration scale, and security posture measured across a whole codebase rather than one repository.
◆Where it's heading
Sourcegraph is moving up the stack from index-and-search toward running the loop itself. Code Finder executes its own search loop and hands an agent exact files and line ranges; Agentic Batch Changes scopes, executes and ships migrations across hundreds of repositories until each pull request is mergeable. The compliance post shows where the enterprise objection-handling is going, framing scoped retrieval as an audit trail of which files an agent read before it shipped a change.
◆Prediction
The evaluation post asks buyers to measure retrieval, agent completion and cost as three separate lines, which suggests the next push is proof rather than product: more benchmark publishing to defend the retrieval layer's value against agents that just search on their own.

◆Recent moves

  1. 19d ago

    How to evaluate Sourcegraph on your own codebase

    A methodology post arguing that retrieval quality, agent task completion and cost should be measured separately rather than collapsed into one number. No product change, but it sets the scoring rules Sourcegraph would like evaluations to use.

    View source ↗
  2. 23d ago

    Compliance-first AI: proving agent provenance for regulated engineering teams

    A positioning piece reframing scoped retrieval as an audit trail: proving which files an agent read before it shipped a change. It names the compliance gap that the agentic products will have to close, without shipping anything against it yet.

    View source ↗
  3. 27d ago

    Code Finder: fast, efficient code search for coding agents

    ⚡ SPARK

    Code Finder is the clearest expression of the shift this trajectory describes: Sourcegraph stops returning results for a human to read and starts running the search loop on the agent's behalf, returning exact files and line ranges.

    View source ↗
  4. 1mo ago

    Three places enterprise security breaks down at codebase scale (and why your current tools don't cover them)

    An essay on why vulnerability remediation stalls at codebase scale, citing AI-written code lacking context, detection coverage gaps, and multi-week fix cycles across large repository counts. Framing for the batch-changes product rather than a release.

    View source ↗
  5. 1mo ago

    Detection in one repo isn't a security posture

    A short argument that a finding in one repository is not a security posture, and that the useful unit is knowing whether the same problem exists everywhere else. It restates the cross-repository premise the migration and remediation products depend on.

    View source ↗
  6. 1mo ago

    Agentic Batch Changes is now in public beta

    ⚡ SPARK

    Agentic Batch Changes moves Sourcegraph from describing a codebase to changing it, taking the long-standing Batch Changes primitive and putting an agent in charge of scoping and completing the work. It is the execution half of the retrieval story Code Finder later fills in.

    View source ↗