GitHub Reworks Copilot Code Review Around CLI Tools
GitHub Changelog said Copilot code review now uses the same file exploration tools available through Copilot CLI and the Copilot SDK. The listed tools include grep, rg, glob, and view, replacing a set of custom exploration tools that previously handled repository inspection.
The change matters for teams that already depend on Copilot reviews as a pull-request gate. GitHub said medium analysis depth preview users also receive pull-request overview attribution and organization-level defaults for review level settings, giving administrators more control over how review effort is applied.
The cost claim is the most concrete number in the announcement. GitHub reported about 20% lower Copilot code review costs while maintaining review quality in offline and online evaluation. That frames the update as an efficiency change, not only a tooling cleanup.
▸ Copilot review deep dive
The switch to grep, rg, glob, and view suggests GitHub is standardizing Copilot review behavior around familiar repository navigation primitives. That is practical for code review because file discovery, targeted search, and local context selection often determine whether an automated reviewer catches a real defect or comments on an isolated line.
The medium analysis depth preview also points to a governance problem inside larger organizations. Review depth is not just a quality setting. It affects cost, latency, and developer trust. An organization-level default gives platform teams a way to choose a baseline before individual repositories tune their own process.
The 20% cost reduction is the key operational figure. GitHub did not present it as a price cut, so buyers should read it as reduced metered consumption or internal review cost under the evaluated workflow. The practical effect is still important: if a team runs automated review on every pull request, small per-review savings compound quickly.
The update also narrows the gap between command-line agent workflows and pull-request review workflows. Copilot CLI and SDK tool use now informs the code review path, which means repository exploration is becoming a shared capability across GitHub’s agent surfaces. For developers, the visible change may be better attribution and more consistent analysis. For administrators, the more important change is that analysis depth becomes easier to budget and standardize.
The main open question is scope. GitHub tied some features to medium analysis depth preview users, so teams outside that preview should not assume every control is available immediately. The update is best read as a staged rollout: common exploration tools are entering the review engine, while the higher-level management controls remain tied to preview availability.
GitHub Adds Marketplace Controls for Copilot Environments
GitHub Changelog said strictKnownMarketplaces entered public preview as an enterprise-managed setting for VS Code and GitHub Copilot CLI. The setting lets organizations restrict which plugin marketplaces users can install from.
The control applies to users licensed through Copilot Business or Copilot Enterprise. That makes the update relevant to managed developer environments, especially where extensions, plugins, and marketplace access fall under security review.
GitHub positioned the setting as a managed policy rather than a developer preference. In practice, it gives administrators a way to limit plugin supply channels without removing Copilot from the editor or command-line workflow.
▸ Marketplace controls deep dive
Plugin marketplace control is a governance feature, and its timing fits the broader shift toward agentic development tools. As assistants gain the ability to inspect code, run commands, and work across local environments, the plugins they can reach become part of the security boundary.
The name strictKnownMarketplaces is technical, but the purpose is direct: reduce uncertainty about where users can obtain installable tooling. In an unmanaged setup, a developer may install extensions from multiple sources. In a regulated or security-sensitive organization, that flexibility can conflict with vendor review, data handling rules, or incident-response requirements.
Public preview status matters. It means administrators can test the setting, but should treat behavior, documentation, and rollout details as subject to change. Teams using Copilot Business or Copilot Enterprise can evaluate how the control affects existing VS Code and GitHub Copilot CLI workflows before making it a hard requirement.
The update also shows how Copilot is becoming less like a single assistant and more like a managed platform surface. Code review depth, plugin marketplace access, Jira integration, and cost centers all point in the same direction: enterprises want central controls that follow developers across editors, command-line sessions, project trackers, and billing systems.
The likely near-term work for platform teams is policy mapping. Security teams will need to decide which marketplaces count as approved. Developer-experience teams will need to watch for friction when legitimate tools are blocked. The setting gives them an enforcement point, but it does not remove the need for a clear internal approval process.
Copilot for Jira Leaves Public Preview
GitHub Changelog said GitHub Copilot for Jira is now generally available after its public preview period. The integration brings Copilot coding-agent progress into Jira issues in real time.
The GA release also adds post-session steering from the Jira chat panel after a draft pull request is opened. That means a user can continue guiding the coding agent from Jira rather than switching immediately into GitHub or an editor.
GitHub also cited simplified onboarding as part of the release. The change is aimed at teams that use Jira as the operating record for engineering work while using GitHub for source control and pull requests.
▸ Copilot for Jira deep dive
The Jira release is less about a new model capability and more about where agent work becomes visible. Many engineering teams still coordinate through tickets, even when the implementation happens in GitHub. If the coding agent works only inside GitHub, project managers and adjacent teams may see the work late. Streaming progress into Jira moves some of that visibility upstream.
General availability is also a product-status change. Public preview often signals that a feature is still being tested with early users. GA usually means GitHub considers the integration ready for broader production use, though individual organizations still need to validate permissions, audit needs, and workflow fit.
The post-session steering detail is important because agent work rarely ends after the first draft pull request. A draft PR may reveal missing requirements, unclear acceptance criteria, or a mismatch with the ticket. Steering from Jira chat gives the issue thread a role in the iteration loop instead of treating Jira as a static task description.
This also changes who can participate. Developers may still make final code decisions in GitHub, but product managers, QA leads, and engineering managers often live in Jira. If agent progress and follow-up prompts remain visible there, non-coding stakeholders can influence the work earlier.
The risk is duplication. Teams will need to decide whether Jira comments, GitHub PR comments, and agent chat messages carry different authority. Without that discipline, the same instruction can appear in several places. The GA release gives teams a tighter integration, but process design will determine whether it reduces handoff cost or creates another thread to monitor.
GitHub Links Cost Centers to Enterprise Teams
GitHub Changelog said cost centers now support enterprise teams. Organizations can assign enterprise teams to cost centers, and usage by members is attributed to the relevant cost center.
The update supports two membership paths. GitHub said updates can follow manual changes or SCIM identity-provider sync, giving enterprises a way to align usage reporting with existing identity management.
For Copilot and other metered developer tools, the accounting change is operationally significant. It makes usage easier to map to departments, business units, or platform budgets without relying only on individual user records.
▸ Cost center attribution deep dive
Cost allocation becomes more difficult when AI development tools move from small pilots to broad deployment. A flat list of users may be enough for a trial. It becomes less useful when finance teams need to know which department, product group, or regional unit is driving usage.
Enterprise team assignment gives GitHub a cleaner organizational unit for attribution. The SCIM connection is especially relevant because many large companies already manage team membership through identity providers. If membership changes flow from that system, cost-center attribution can follow employment and team changes with less manual reconciliation.
This update should matter to platform engineering groups that administer Copilot Business or Copilot Enterprise at scale. They often need to answer two questions at once: whether usage is growing and who should pay for it. Team-level cost centers help separate those questions from license management.
The change also complements the Copilot code review cost update. GitHub reported lower review costs in evaluation, while cost centers make the remaining usage easier to attribute. One update addresses efficiency; the other addresses accountability. Together, they help administrators treat AI coding assistance as a managed operating expense.
The limitation is that attribution is only as accurate as team membership. If teams are stale, overlapping, or manually maintained outside the identity provider, the cost-center report can inherit that ambiguity. Organizations using SCIM sync will likely get the most consistent result, provided their identity groups already match budget ownership.
OpenAI Frames Agents as Longer-Task Workplace Tools
openai.com published research on how agents are transforming work, saying agentic systems are enabling longer and more complex tasks across roles. The piece does not announce a price change, API deprecation, or versioned product release.
Its relevance for tool users is the direction of travel. OpenAI described agents less as single-turn assistants and more as systems that can carry work across extended task sequences.
For developers, designers, and planners, the practical implication is workflow design. Tools that can sustain longer tasks require clearer instructions, better checkpoints, and stronger review points than short chat completions.
▸ OpenAI agents deep dive
The OpenAI item sits differently from the GitHub changelog entries. GitHub announced specific product controls and integration changes. OpenAI supplied a research frame for why agent features are becoming central to workplace software.
The useful distinction is task length. Short AI interactions often produce a single answer, draft, or code snippet. Longer agentic work spans planning, tool use, revision, and follow-up. That shift changes how teams should evaluate tools. Accuracy at one prompt is still important, but continuity, state handling, and escalation behavior become just as important.
OpenAI’s framing also helps explain why GitHub’s June 25 updates focus on management surfaces. If agents do more work, organizations need more controls around where they operate, what tools they can reach, how their costs are assigned, and how their progress is reported. The research narrative and GitHub product changes point toward the same operational concern: agent work must be observable and governable.
The workplace impact will vary by role. Developers may see agents enter code review, issue implementation, and repository search. Designers may use them for structured iteration and asset preparation. Planners may use them to turn requirements into tracked tasks. In each case, the value depends less on novelty and more on whether the agent can work inside existing systems without breaking accountability.
The main caution is that research language can run ahead of product guarantees. The OpenAI post supports the broader direction, but teams should still evaluate concrete tool behavior through documented features, pricing terms, and audit controls. For June 25, the actionable product changes came mainly from GitHub’s changelog; OpenAI supplied context for why those changes matter.
Red Panther: Figma's new updates are...kinda great - This is the first AI tool that I'm actually half excited about. Lots of potential, but it's not all perfect. Watch the full Config 2026: ...
Copilot code review: Analysis depth and efficiency updates
Reported by GitHub Changelog. GitHub says Copilot code review now uses file exploration tools from Copilot CLI and SDK, including grep, rg, glob, and view, replacing cus…
Cost centers now support enterprise teams
Reported by GitHub Changelog. GitHub added support for assigning enterprise teams to cost centers.
The AI coding craze gave GitHub its best month ever, executive tells employees
Reported by Business Insider. Business Insider reports that GitHub CTO Vladimir Fedorov told employees June was GitHub's best month ever, with usage increasing after Cop…
Google delays Gemini 3.5 Pro launch to July as it tweaks its new frontier AI model
Reported by Business Insider. Business Insider reports that Google pushed Gemini 3.5 Pro from a planned June rollout to July to gather more feedback and tune the model.
At a glance
Fact
Publisher
Source
Copilot code review now uses grep, rg, glob, and view for repository exploration.
Q1. What changed in Copilot code review on June 25?
A. GitHub Changelog said Copilot code review now uses grep, rg, glob, and view from Copilot CLI and SDK for repository exploration. GitHub also reported about 20% lower review costs in evaluation while maintaining review quality.
Q2. How should administrators use strictKnownMarketplaces?
A. GitHub Changelog described strictKnownMarketplaces as a public-preview enterprise-managed setting for VS Code and GitHub Copilot CLI. Copilot Business and Copilot Enterprise administrators can use it to restrict plugin marketplace sources for licensed users.
Q3. Does Copilot for Jira change developer workflow?
A. Yes. GitHub Changelog said the GA release streams coding-agent progress into Jira issues and allows post-session steering from Jira chat after a draft pull request opens. That keeps ticket context active during agent-driven implementation.
Q4. How do cost centers and review-cost reductions differ?
A. GitHub’s 20% code review cost figure concerns review efficiency, while cost centers concern attribution. GitHub Changelog said enterprise teams can now map usage to cost centers through manual membership changes or SCIM identity-provider sync.
Q5. What should teams watch after these updates?
A. Watch preview status and rollout scope. GitHub marked strictKnownMarketplaces as public preview, and medium analysis depth features apply to preview users. openai.com also framed agents as longer-task systems, which raises the need for stronger review and governance controls.
