From 28 October 2026, GitHub Copilot is no longer only a coding assistant in your editor. On paid plans, cloud agent sessions persist and follow you across the web, mobile app, command line, VS Code and the GitHub Copilot app — work started on a phone keeps running in the cloud while you steer it from VS Code.

At the same time, Copilot Automations turn routine repository work into shared cloud workflows that propose improvements based on what they learn — improvements a human still reviews and approves. For a business, that means faster triage and review, less lost context, and a new management job: approval gates, permissions and cost control.

GitHub Copilot Cloud Automations: Agents That Keep Working

What GitHub actually announced on 28 October 2026

GitHub announced the connected cloud experience and Copilot Automations at GitHub Universe 2026, rolling out from 28 October 2026 on paid Copilot plans. GitHub's changelog post on the connected Copilot cloud experience describes two changes: where agent work lives, and who can reuse it.

Previously, AI-assisted coding work was tied to one session on one machine. Now the session lives in the cloud and moves with you, while Automations move routine work into the repository itself, where the whole team can see it. That matters to founders as much as to engineers — including teams building with AI agents or commissioning custom software development.

Persistent cloud agents: work that follows you, not your laptop

Start on a phone, steer from VS Code, check on github.com, while the agent keeps working in the cloud. A session is ongoing work — task, context and progress — available across surfaces.

Work no longer stalls because its owner closed a laptop or switched devices: a developer can kick off a task, check progress between meetings, and pick it up in the editor when it needs judgement. Progress does not depend on one person being at one desk.

Scope and permissions still apply

Sessions are scoped to one organisation or to personal repositories by default. Organisation members need single sign-on for organisation-repository session data, and where a user lacks write access, the session is read-only. The cloud agent inherits your permission model rather than bypassing it — the same discipline that matters in SaaS development and Flutter app development.

Copilot Automations: workflows that learn, but do not go rogue

Copilot Automations are AI-powered workflows inside a repository that run in the cloud, triggered manually, by an event, or on a schedule.

Issue triage is a strong first candidate: new issues arrive messy, duplicated or mislabelled. An automation can sort, label and route them consistently, so humans decide only the issues needing judgement. In code review, it can do a first pass and flag risks, leaving the human to judge trade-offs.

Learning with a human approval gate

Automations can learn from outcomes and propose improvements — but users review and approve them; they are not applied automatically. A system that quietly rewrites its own workflow is a governance risk; one that asks “shall we change the rule?” is a productivity tool. See GitHub's post on Copilot Automations.

Shared automations belong to the repository

Shared automations belong to the repository, not to their creator, and survive that person leaving, with a change history. It is the same argument for documented AI automation elsewhere: automate repeatable work, keep reasoning visible, let people handle exceptions. Organisation repositories need the admin policy “Allow use of Copilot cloud agent billed to the organization” enabled.

The wider Universe announcements that change the maths

Copilot Fusion is now generally available, routing work across models. GitHub claims Fusion Max sits within two points of Opus 5.5 Max on the Artificial Analysis Coding Agent Index at half the cost and 40% less time per task — GitHub's claim, not independently verified.

Copilot code review Max targets the hardest reviews: GitHub reports the highest F1 score on ReviewBench for high-severity issues, while Lite and Balanced remain for everyday reviews. Copilot projects extend the idea across multiple repositories in managed parallel sessions. Our guide to coding agents offers a useful comparison.

Enterprise customers and AI credits

From 28 October 2026, GitHub Enterprise customers can use all Copilot features without a separate subscription, paying for AI credits used — via GitHub Sales now, with self-serve coming soon. Existing Business and Enterprise customers can keep their plans or move over. Cloud sandbox usage can draw from the included Actions compute entitlement, with billing beginning in November 2026.

What this means for software teams and non-technical founders

For an engineering team, the wins are triage speed, review coverage and continuity: backlogs sorted faster, first-pass reviews even when senior developers are busy, work continuing while its owner is elsewhere. No public figure quantifies that saving — treat any vendor time-saving claim as an estimate to test in your pilot, not a promise.

For a non-technical founder, you are buying fewer dead hours: less time lost to handoffs and work restarting because context was lost. Teams doing ecommerce development often lose days to triage queues, as do teams doing CRM development. Remember the difference between a chatbot and an AI agent: a chatbot answers, an agent acts — and Automations act only with approval.

The real management job: permissions, approval gates and cost control

The management job is deciding who can do what, what may run without a human, and what it costs. Three controls cover most of the risk.

• Permissions first. Confirm single sign-on, repository scope and read-only rules before wide rollout. A contractor who should not change production code should not do so via the agent.
• Approval gates for every learning proposal. A named reviewer approves each proposal; log who approved what.
• Cost visibility from day one. Track usage per repository and team.

None of this needs a large governance programme: one owner, a short written policy and a monthly check. Cost planning belongs beside build cost — see our guide to custom software development costs in the UK.

A practical adoption checklist

1. Pick one pilot repository. Active work with a real triage burden, but not your most sensitive codebase.
2. Set the admin policy deliberately. Enable the organisation cloud-agent policy only for the pilot, confirm single sign-on, and name who may create automations.
3. Start with two automations. Issue triage first, then first-pass code review. Measure time to first response, time in review, and how often humans overrule them.
4. Write the approval rule down. No learning proposal applies without a named approver. Review approved changes monthly.
5. Track credits weekly. Compare against the included Actions entitlement and forecast November billing.

What not to automate yet

Do not automate decisions you cannot explain or reverse easily: production data changes, security-sensitive releases, customer refunds, or anything hard to undo if wrong. Never let it approve its own proposals or exceed a careful human's access; when in doubt, let it prepare work, not finish it.

That is where an experienced partner helps: choosing the right first workflows, setting approval rules, and keeping costs visible. Coheser builds automation for teams in the United Kingdom, the UAE, Saudi Arabia, the USA and Pakistan — talk to us about what your team should automate first.

Frequently asked questions

What is the GitHub Copilot connected cloud experience?

A persistent cloud agent session, rolling out from 28 October 2026 on paid Copilot plans, that follows you across web, mobile app, command line, VS Code and the GitHub Copilot app.

What are Copilot Automations?

AI-powered workflows inside a repository that run in the cloud, triggered manually, by events or on a schedule — for example, issue triage and code review. They propose improvements learned from outcomes, which users review and approve; nothing applies automatically.

Do shared automations disappear if their creator leaves?

No. Shared automations belong to the repository, so they survive the creator leaving, and they carry a change history. Organisation repositories need the admin policy “Allow use of Copilot cloud agent billed to the organization” enabled.

How does billing work for Enterprise customers?

From 28 October 2026, GitHub Enterprise customers can use all Copilot features without a separate subscription, paying for AI credits used — via GitHub Sales now, self-serve coming soon. Existing Business and Enterprise customers can keep their plans or move over; cloud sandbox billing begins in November 2026.