Everyone has a copilot. Almost nobody has a loop that can survive backlog to pull request without a hero. You've probably watched a team demo a coding assistant on a clean file and then watch the same assistant fail the moment it cannot see the ticket. That is not a talent shortage. That is a control plane shortage.

Taroon Mandhana put the gap on the record. Ninety-four per cent of engineering leaders are using AI. Six per cent have systems to scale it across the whole software development lifecycle. Those two numbers are the piece. The industry poetry around them is not.

What they are actually shipping

Code Context sits on the Teamwork Graph so the agent can see the work, not just the file. Agent Context Controls are the governance layer – what the agent is allowed to know. DX analysis in the piece: about 64% more per developer when they have the most Graph context. Treat that as their study, with the hedge that "most context" is doing a lot of work.

Agent loops in Jira run from backlog to pull request, plus Standards, plus AI Review. There is DX for Agentic Development, and a Jira Agent Usage Dashboard so someone can see whether the loop is a toy. Code Context is open beta for paid customers. Agent loops, Standards and AI Review are private early access. Controls and the dashboard are due generally available in the coming months. Atlassian Summit on 22 September 2026 includes a State of AI SDLC session. That session is on the agenda. It is not a shipped system.

Atlassian is an Australian company with a United States gravity well. The buyers sitting in this gap are in Sydney, Austin and every shop that bought Jira and then bought six coding tools that cannot see the ticket. If you are a CTO, Mandhana's 94/6 is the only slide you need this month. If you are in L&D, do not turn it into a prompt course. The 6% have a system. The 94% have a tab.

Governed loops are an agile story wearing an AI hat. The ceremony is not stand-up. It is what context this agent gets, who reviews the patch and what standard fails the build. A sprint that leaves a decision is closer to that than a lunch-and-learn on ChatGPT.

If you already live in Jira, Mandhana is describing the control you should have demanded when the first coding copilot landed. Backlog to pull request without a human who can see the ticket is how you ship a confident wrong patch. Standards and AI Review are the boring products. Buy boring. The dashboard is how you catch teams that turned the loop on for a demo and never looked again.

Availability matters because a steering pack will call something generally available while the page still says early access. Code Context is a paid beta. Loops, Standards and AI Review are private. Controls and the usage dashboard are due in the coming months. The 22 September session is a title on an agenda. If the steering pack says Atlassian has solved agentic software delivery, it is ahead of what has shipped.

The 64% DX figure only exists next to "most Graph context." Teams with a messy Graph will not get it. Clean the work objects before you brief a productivity miracle. That is the enablement job hiding in an engineering post.

What this means for L&D

Australia and the United States both get this piece because Atlassian is the rare company that is both. A Sydney engineering lead and a Bay Area engineering lead are looking at the same 94/6. Neither of them needs another copilot licence. They need a loop with a standard that can fail the build.

Print 94 and 6. Name Mandhana. Do not general-avail a dashboard you have not shipped. Put Graph context in the same sentence as the productivity claim so nobody hears 64% for free.

Sources: Atlassian blog, Taroon Mandhana, 10 September 2026.