--- title: "AI governance between people and systems" url: https://zerofive.ai/en/blog/ai-governance/ai-governance-people-systems canonical: https://zerofive.ai/en/blog/ai-governance/ai-governance-people-systems language: en published: 2026-10-29 updated: 2026-10-28 author: "ZeroFive.AI" tags: AI governance, AI adoption, organisation, skills, AI Rating abstract: "AI spending, productivity, training and ROI often have different owners. Why a single mandate over people and systems matters, and where to start." --- # AI governance between people and systems **In short.** In most companies, AI spending, productivity, training and return on investment each have a different owner, and the value people get from the tools is lost on the way to the organisation. Atlassian chose to put its people and its internal systems under a single leader. Whatever the role is called, the choice shows that one mandate is needed across tools, changing work, skills and risk, and that holds for a mid-sized company too. In April 2026 Atlassian broadened the remit of its head of people, who now also leads internal IT, data science, customer technical support and customer engineering. The group she runs grew from around seven hundred people to more than three thousand, as [Information Age](https://ia.acs.org.au/article/2026/atlassian-s-ceo-on-why-his-it-team-now-reports-to-hr.html) reported in early October. The new title got people talking, and it is easy to see why: a software company moving IT under the function that looks after people makes news. We care little about what the role is called, since in another company it could carry any name, and much more about the reasoning behind the choice, because it describes with precision a problem we see every week in Italian companies adopting artificial intelligence. ## The reasons for the reorganisation Mike Cannon-Brookes, Atlassian's CEO, explained it on [The Verge's Decoder podcast](https://www.theverge.com/podcast/1000914/atlassian-ceo-mike-cannon-brookes-saaspocalypse-ai-enterprise-software-trello-jira). When he listed everything the company had to change to really work with AI, he ended up with two groups of projects, one about people (how to grow them, how to get them using these tools, how to change the way they think) and one about systems, meaning the MCP servers to set up, the operability of internal applications, the control of token spending. In the same interview Cannon-Brookes says he repeatedly found himself refereeing internal discussions about a single issue, to settle whether it was a talent problem or a system problem. The decision to put everything under one leader comes from there, from realising that on AI projects the distinction stops being useful. Anyone working on AI projects inside a company knows that discussion well. An assistant for customer care goes into production, response times drop, and three months later someone asks why the team is still sized as before. IT replies that its job was to release the tool, HR that nobody had involved it, the business that it was waiting for the numbers. Each is right about its own piece, and meanwhile the promised value does not arrive. ## The fragmentation of accountability Avani Prabhakar, the executive who now leads the new group, described the typical situation in an [interview on Workday's Future of Work podcast](https://www.workday.com/en-us/perspectives/hr/2026/09/how-hr-can-lead-ai-transformation.html) in September 2026. In most organisations IT governs model spending, the CTO follows productivity, HR manages training and yet another function measures the return on the transformation. Four owners for the same investment. When results are slow to come, each of them holds part of the explanation and none holds the whole picture, which is exactly the condition in which an AI project stays in pilot for years. In February 2026 the Artificial Intelligence Observatory of Politecnico di Milano counted active AI projects in 71% of large Italian companies and [structured governance in only 9%](/en/blog/approfondimenti/71-percent-ai-projects-9-percent-governance). Little of that gap is technological: it is made of decisions nobody has the authority to take alone, because they touch systems, roles and budget together. AI is not an IT project, it is a project of governing the organisation. We have been saying so for some time to the boards we meet, and Atlassian's choice has the merit of turning that sentence into an org chart. ## The impact of automation on work When a process is automated, the release is the easy part. The hard part comes afterwards and concerns the work that remains: the time freed up has to be reinvested in something, people need skills that were not needed before, responsibilities have to be redesigned because whoever used to execute now checks, corrects and decides on the exceptions. If these decisions are taken by different functions at different times, the time freed up disperses. Atlassian measured it in research with Fortune 500 customers: individual use of AI produces a personal productivity gain of around 33%, while returns at company level remain stuck in the low single digits. Individual productivity is there, and it gets lost on the way from the person to the organisation, which is precisely the space nobody in a traditional org chart is fully accountable for. ## Adoption metrics Anyone setting up a dashboard on AI adoption should know one detail of the Atlassian case. The first metric the company adopted counted how many employees used at least one AI feature. Within three months it had reached 90%, and it told the company almost nothing useful. They replaced it with a maturity scale that separates non-users, people who use AI for simple tasks, people who use it strategically and people who reason with it, building hypotheses and testing them. In that last group they saw roughly double the output of simple users, and above all an effect on colleagues, because these are the people who build the context (connected data, instructions, workflows) that makes AI useful to everyone else. It is close to the reasoning behind the [delegation levels](/en/blog/strategia/ai-assisted-ai-first-ai-native-delegation-levels) we use in our assessments, which measure how much work is handed to the system instead of how many people hold a licence. IT can manage a diffusion metric on its own, by looking at licence logs. A depth metric requires knowing how people work, which skills they have, where roles are changing, and that is why in our [AI Rating](/en/services/ai-rating) real adoption and user trust have a dimension of their own, Confidence, assessed separately from technical Delivery. ## The order of decisions on tasks and roles The other choice we find transferable concerns the order of decisions. Atlassian worked first on individual tasks, then on the skills those tasks required, and only at the end on roles. It rewrote job descriptions after observing how they were changing, starting from design, where some designers had begun writing and shipping code and deserved a new job family. Prabhakar sums it up this way: instead of asking whether a role will still be needed, or needed more or less, they looked at how that role was really changing. It sounds like a nuance, and yet it changes the nature of the conversation with people, because it starts from what they do every day and only from there gets, if at all, to forecasts about their future. This sequence only works if whoever decides sees systems and people at the same time. Those who look only at systems tend to jump straight to roles, with headcount reduction estimates built on vendor presentations. Those who look only at people tend to stop at generic training, without linking it to the tasks that are changing, a limit we wrote about in the piece on [change management for AI](/en/blog/approfondimenti/change-management-ai). ## The job cuts Alongside the reorganisation there is a figure that completes the picture. In the same year Atlassian cut around 1,600 jobs, about 10% of its global workforce, to self-fund further investment in AI and enterprise sales. Cannon-Brookes spoke of very difficult choices linked to changing the skill mix. Putting people and technology under the same leader also means that the same function decides on training and on exits. It is consistent, because these really are connected decisions, and for that very reason it requires explicit criteria, written beforehand, and a board that approves them knowing what it is approving. Cannon-Brookes distinguishes between transparency on every detail, which he says the company does not owe its employees, and clarity on the reasons behind each decision, which he considers owed. On that second point, Italian companies are often far behind. ## The implications for Italian companies A mid-sized company does not have three thousand people to reorganise and does not need a new title in its org chart. It needs someone, with a clear mandate from the top, who is accountable for the AI programme as a whole: the tools, the work that changes, the skills, the risk. In many companies it will be the general manager, in others the CFO or whoever runs operations, who it is matters less than the mandate being single, the same principle we ask for every agent in production, which must have a [named owner](/en/blog/ai-governance/ai-agents-named-owner-production). How things start matters too. Atlassian began the work around three years ago with a table bringing together the CISO, CTO, CIO and head of people, tackling security rules first and then workflows. People were at the table from day one, while decisions were still being taken, instead of being brought in once the project was over to manage its consequences. The rest follows from there: an AI budget read in one place, metrics that measure depth of use and value produced, decisions on the work that remains taken together with those on automation. It is also why our work almost always starts from a maturity measurement across four dimensions (Readiness, Delivery, Risk and Confidence) with thresholds that do not offset one another: excellent technical infrastructure does not cover for an organisation that does not yet know what to do with the time AI gives back. The question to bring to the board, at this point, is who is accountable for the work that remains once automation is in production, and how many of the companies we know could put a name to it. On how to set up that conversation with the board, see the article on [how to present the state of AI to the board of directors](/en/blog/ai-governance/presenting-ai-to-the-board). To see where your organisation starts from you can [start an AI Rating](/en/start-ai-rating) or [book an assessment call](https://calendly.com/fabiolalli/zerofive).