Robotics, Copilot workflows, and GitHub changes #95

August 1, 2026 — 3 min read

Today's Letter

Google DeepMind introduced Gemini Robotics ER 2 on July 30, 2026 as a robotics model focused on video understanding, task orchestration, and multi-robot collaboration.
Google describes the release as a new model for broader physical-world assistance, extending Gemini-based reasoning into embodied systems.
The model name moves from Gemini Robotics ER 1.6 to ER 2, indicating a new iteration in the Gemini Robotics ER line.
Factcheck materials indicate support for three control modes, alongside coordination across multiple robots in the same workflow.
Google positions the system within its wider Gemini stack, with related references in the materials including the Gemini API and Google AI Studio.

Source: blog.google
More: deepmind.google · news.hada.io · siliconangle.com

GitHub Models was retired on July 30, 2026, according to the GitHub changelog.
The shutdown covers the playground, model catalog, inference API, and bring your own key support.
The services are no longer available to any customer, including existing customers with active usage.
GitHub said the move completes the retirement process first announced on July 1.
For projects that still need model access, GitHub points users to Microsoft Foundry and its broader model catalog.
For AI workflows built directly on GitHub, the company recommends GitHub Copilot as the remaining first-party path.
The update marks the full end of GitHub Models rather than a phased restriction for new customers only.

Source: github.blog

GitHub introduced stacked sessions and stacked pull requests in the GitHub Copilot app, according to a July 30 blog post.
The update is aimed at breaking larger coding tasks into smaller linked work units instead of forcing a single long session or one large pull request.
Stacked sessions group related Copilot interactions across sequential steps, preserving context as work moves from one change set to the next.
Stacked pull requests let that sequence be submitted as smaller reviewable PRs with explicit dependency order.
The model is positioned for multi-step refactors and feature work where reviewability and incremental merge flow matter.
The announcement was published on the official GitHub Blog under the GitHub Copilot product track.
The post appears to describe a Copilot app workflow change rather than a new standalone model or pricing update.

Source: github.blog
More: blockchain.news · itbrief.com.au

Vercel Passport is now generally available as an Enterprise plan feature for protecting deployments behind an external identity provider.
Visitors can authenticate through Okta, Microsoft Entra ID, or another OIDC provider before reaching a protected deployment.
The @vercel/passport package provides getIdentity(), which reads a verified signed identity token injected by Vercel after session validation.
Tokens can include additional claims such as group membership, allowing application code to enforce access rules like role- or team-based authorization.
verifyIdentity(), available in @vercel/passport 1.0.0 and later, can validate forwarded tokens in downstream services against expected team, project, and environment claims.
Successful Passport sign-ins are logged as passport-access-granted events in Activity Log and Audit Logs with visitor, hostname, and project details.
Protection Bypass for Automation and Trusted Sources can keep webhooks, cron jobs, CI runs, and authorized external services working with Passport-enabled deployments.
Passport also now supports custom environments such as staging or QA, extending the same sign-in flow beyond preview and production.

Source: vercel.com

Jocoletter curates AI, software, and product trends for developers and builders.

#GitHub #Google #Vercel