Five real product releases from AI companies changed how software teams will run their next sprint. No hype, no guessing about what might ship someday. These all shipped, or were previewed, between June 22 and June 26, 2026.
Two of these stories are about your daily tools getting new agent features. One is about open-source security getting real AI help, with named projects and real CVEs. One is about an AI model that can now click, type, and navigate a screen on its own. And one is a preview of a new frontier model that most teams can't touch yet, gated by the U.S. government, but points at where every AI lab is heading next.
Here's what happened, and what it means for your Scrum team.
Story 1 · June 25, 2026
Jira can now hand a ticket straight to an AI coding agent
GitHub made Copilot for Jira generally available on June 25, after a public preview that started in March. You can now assign a Jira issue to GitHub Copilot's cloud agent the same way you'd assign it to a teammate, either through the assignee field or by mentioning the agent in a comment.
The agent works in the background, then opens a draft pull request. What's new at general availability: the agent's progress now streams live inside the Jira ticket, so you can watch it work without switching to GitHub. And if you leave a follow-up note in the Jira chat panel after the draft pull request shows up, the agent keeps working on that same pull request instead of starting a new one. The integration can also pull extra context from linked Confluence pages.
Sourced facts
- Public preview began March 2026; general availability landed June 25, 2026
- Agent can be assigned via the Jira assignee field or an @mention in comments
- Progress streams live inside the Jira issue as the agent works
- Follow-up instructions in Jira chat continue the same draft pull request
- Setup steps for connecting a GitHub org and repos were reduced at GA
Scrum Team Signal
A Product Owner can now assign a backlog item to an AI agent from the sprint board, and the whole team can watch it work without leaving Jira. Before your next sprint, add one line to your Definition of Done: a human reviewed and approved the agent's draft pull request. Watching an agent work is not the same as reviewing its work.
Source: GitHub Changelog, June 25, 2026
Story 2 · June 22, 2026
JetBrains developers can now pick Claude as their coding agent
GitHub shipped a JetBrains update on June 22 that does two things at once. First, GitHub Copilot's cloud agent reached general availability inside JetBrains IDEs, and organization or enterprise admins can now publish standardized custom agents that every developer on the team can use without extra setup.
Second, and more notable for anyone who likes options, Claude is now available as an agent provider inside GitHub Copilot for JetBrains, in public preview. To use it, a developer installs the Claude Code CLI and points Copilot at it in settings. Anthropic's agent then runs directly inside the JetBrains agent picker alongside GitHub's own models. One catch worth knowing: right now the Claude agent runs in "bypass permissions" mode, meaning file edits and tool calls are automatically approved without a manual check. Configurable permissions are planned for a future release.
Sourced facts
- Cloud agent reached general availability in JetBrains IDEs on June 22, 2026
- Org and enterprise admins can publish custom agents for their whole team
- Claude agent provider entered public preview via the Claude Code CLI
- Claude agent currently runs in bypass permissions mode; all edits auto-approve
- Update also added message queueing during running sessions and a debug logs view
Scrum Team Signal
Your tech lead can now publish one shared agent configuration for the whole team instead of everyone hand-tuning their own prompts. That's a real fix for the "why did your agent write this differently than mine" arguments in code review. But because the Claude preview auto-approves every file edit, don't point it at a shared branch until your team has tested it on a throwaway one first.
Source: GitHub Changelog, June 22, 2026
Story 3 · June 22, 2026
OpenAI starts paying human security engineers to help AI patch open source
OpenAI launched a program called Patch the Planet on June 22, built with the security firm Trail of Bits. The idea: use OpenAI's most cyber-capable models to find vulnerabilities in widely used open-source software, but always route the findings through expert human review before anything reaches the project's maintainers.
OpenAI says AI is speeding up how fast vulnerabilities get discovered, but discovery alone doesn't protect anyone unless someone patches the hole. Security engineers on the program validate what the AI finds, help develop the actual patch and tests, and work through each project's normal disclosure process. The first projects in the program are cURL, NATS Server, pyca/cryptography, Sigstore, aiohttp, the Go project, freenginx, Python, and python.org. OpenAI also said its Codex Security tool had already spotted vulnerable patterns matching four of the six dnsmasq CVEs that were later fixed in version 2.92rel2.
Sourced facts
- Program launched June 22, 2026, built in partnership with Trail of Bits
- Security engineers review AI findings before they reach project maintainers
- Initial projects: cURL, NATS Server, pyca/cryptography, Sigstore, aiohttp, Go, freenginx, Python, python.org
- Codex Security independently flagged patterns matching 4 of 6 dnsmasq CVEs later fixed in 2.92rel2
- OpenAI researchers also reported 5 exploitable Chrome V8 bugs and over 10 Safari/WebKit bugs in recent work
Scrum Team Signal
If your codebase depends on curl, Python, aiohttp, NATS, or any library in the pyca/cryptography or Sigstore family, expect more security patches to show up from this program over the coming sprints. Put a standing backlog item on your board now: "review and apply upstream security patches," reviewed every sprint, so these don't pile up into one scary dependency-update sprint later.
Source: OpenAI, June 22, 2026
Story 4 · June 24, 2026
Google's fast, cheap model can now see your screen and click things
Google announced on June 24 that "computer use" is now a built-in tool inside Gemini 3.5 Flash, the same fast, low-cost model developers already use for search and function calling. Before this, computer use only existed as a separate, standalone model. Now a single Gemini 3.5 Flash agent can look at a screenshot, reason about it, and take an action, like clicking a button or typing into a field, across a browser, a mobile app, or a desktop.
Google is pointing this at "long-horizon and enterprise automation tasks like continuous software testing." The company also added targeted training to resist prompt-injection attacks, plus two optional safeguards: requiring a person to confirm sensitive or irreversible actions, and automatically stopping a task if the system detects a hidden prompt-injection attempt. Google's own post is direct about the tradeoff: both of those safeguards are opt-in, not turned on by default.
Sourced facts
- Announced June 24, 2026; available via the Gemini API and Gemini Enterprise Agent Platform
- Computer use is now native in Gemini 3.5 Flash instead of a separate standalone model
- Works across browser, mobile, and desktop environments
- Google added adversarial training against prompt injection during live sessions
- Confirmation-before-action and injection auto-stop safeguards are opt-in, not default
Scrum Team Signal
This is a cheap way to build an agent that exercises your actual UI, useful for exploratory testing inside a sprint instead of relying only on your scripted test suite. But treat it as a privileged tool, not a novelty: since the safety confirmations are opt-in, someone on your team needs to explicitly turn them on before any computer-use agent touches a real environment.
Source: Google, June 24, 2026
Story 5 · June 26, 2026
OpenAI previews its next model, and the government decides who gets it first
OpenAI began a limited preview of the GPT-5.6 family on June 26: Sol, its flagship model, Terra, a cheaper balanced option, and Luna, its fastest and lowest-cost model. GPT-5.6 adds a new "max" reasoning effort setting for harder problems, and a new "ultra mode" that splits complex work across multiple subagents instead of relying on one agent to do everything.
The unusual part is how it's rolling out. OpenAI says it briefed the U.S. government on the model's capabilities ahead of launch, and at the government's request, the preview is starting with a small group of trusted partners whose participation was shared with officials, before wider release. OpenAI states plainly that it does not want this kind of government-gated access to become the normal way new models launch. On cybersecurity benchmarks, GPT-5.6 Sol matches OpenAI's Mythos Preview model using roughly a third of the output tokens, but OpenAI says it did not autonomously produce a working full exploit chain in testing against Chromium or Firefox.
Sourced facts
- Limited preview began June 26, 2026, through the API and Codex only, not ChatGPT
- Access limited to trusted partners and organizations with government visibility into participation
- No general-availability date has been announced; OpenAI plans wider release "in the coming weeks"
- New "ultra mode" splits work across subagents instead of a single agent
- On ExploitBench, GPT-5.6 Sol matches Mythos Preview using about one-third the output tokens
Scrum Team Signal
Most teams can't get their hands on this one yet, so there's no action item for this sprint. But the "ultra mode" pattern, splitting one big task across several agents working in parallel, is now showing up at every frontier lab. Use your next retro to start the conversation: when your team's agent tooling gets this capability, who decides how a story gets split across agents, and who owns reviewing the combined result?
Source: OpenAI, June 26, 2026
Next week
Anthropic just replaced its default model — and priced it to compete
On June 30, Anthropic launched Claude Sonnet 5 as the new default model for Free and Pro plans, claiming performance close to its flagship Opus 4.8 model at a fraction of the cost, plus a new Claude Science desktop app for researchers. Next week's issue breaks down what changed, what it costs, and what it means for teams already running Claude Code in their sprints.
RC
ROD CLAAR
Rod has spent over two decades teaching Scrum, Agile, Test Driven Development, and software design patterns to development teams. He now focuses on how AI is changing the day-to-day work of Scrum Masters, Product Owners, and developers, and writes AgileAIDev.com to keep practitioners current without the hype.