The assessment is in every review; counted approval stays off
GitHub Copilot code review, in public preview for Copilot Pro, Pro+, Max, Business and Enterprise, now places an approval assessment in the overview comment of every Copilot review. The assessment says whether Copilot considers the pull request ready to approve, and it sits beside the detailed comments Copilot already writes. GitHub writes that this assessment alone does not count toward merge requirements. Copilot does not approve pull requests by default. When an administrator enables the feature, Copilot can submit an approval that counts toward the repository's required-approvals rule. The new product is the counted vote; the comment labor was already in the workflow.[1]
The task that moved is the counted approval, not the written comment. Copilot already performed review-comment work; the new switch decides whether that same tool can occupy a slot a person had to fill before a merge proceeds. That is a redesign of the approval task inside existing work, not a disclosed cut in reviewer employment or pay. An alternative reading is that organisations keep human required-reviewers in place and treat Copilot's vote as extra capacity, so reviewer hours and work intensity do not move. GitHub's changelog, however, states neither use and gives no figure for reviewer hours, hidden labor or headcount. Any time saved sits with whoever currently waits on a required approval; residual error still sits with the organisation that enabled the vote.[1]
The switch sits with the administrator, the file path with people
Administrators configure Copilot approval at 3 levels. An enterprise can leave approvals off for the whole enterprise or let organisations decide. An organisation can turn approvals on across the organisation, enable them for specific repositories, let repository administrators decide, or turn them off across the organisation. A repository can turn approvals on or off and choose which file paths Copilot is allowed to approve. If new commits are pushed after Copilot approves, that approval is dismissed just as a human reviewer's would be, and a new review can be requested from Copilot for a fresh approval. The product copies the human form of the vote, including the rule that a later push drops it.[1]
The party that can put Copilot into a required-approvals slot is the administrator at enterprise, organisation or repository level, not the reviewer whose bargaining power over that slot had been the merge rule. File paths are the remaining written boundary of human-only approval. GitHub does not publish how many repositories have enabled the vote, which workers lost a required-reviewer designation, or whether review time was reserved as paid work. An alternative is that path lists keep Copilot on tests and generated files while people keep production paths; the changelog describes the control, not the pattern of use. Without a published allocation of hours, exception handling stays unpaid hidden labor.[1]
Empirik also lets the model through and leaves the exception with a person
Empirik came out of Sequoia's incubator with 21 million dollars in seed funding from Sequoia, Canapi and Alumni Ventures. The tool tracks how a change to one system ripples through the rest of the infrastructure and uses large language models to let low-risk changes through, fence larger ones and send dangerous updates to a human. Sequoia partner Bogomil Balkansky describes that split as an autonomous traffic cop. Customers named in the launch include S&P Global and Guardant Health. Kartik Chandrayana, the chief executive and formerly chief product officer at Quantum Metric, says Empirik wants to do for infrastructure engineers what agentic coding tools did for software developers; the launch copy also names Cursor and Claude Code. Founders Avon Puri, formerly of Rubrik and VMware, and Sudheer Dhurjati built the product. That comparison is a stated aim and is not yet proven at large volume; the announcement states no movement in infrastructure-engineer employment, hours or wages.[2]
GitHub Copilot's counted approval and Empirik's change permit share a constraint: the model sits on the default gate for the named task, the worker sits on the exception queue, and the switch that puts the model on a counted gate is an administrator or vendor setting rather than a bargained reallocation of hours. Neither announcement states a share of counted approvals Copilot submitted, a share of Empirik-tracked changes sent to a human, or a movement in reviewer or infrastructure-engineer employment. The next useful figure is that share, in either product. If Copilot's counted approval remains a documented GitHub feature, the Copilot code review documentation or changelog will still describe that approval as off by default at the enterprise, organisation and repository levels through the end of 2026; a flip of that default would reallocate bargaining power over the vote without saying who works the exception queue.[1], [2]