From Collaboration by Exception to Collaboration by Design: New South Wales Lessons for Government 3.0

1. Executive Summary

The Sydney Government 3.0 discussion examined why collaboration remains difficult across New South Wales government, despite broad agreement that connected services, shared infrastructure and cross-agency problem-solving are essential to public sector transformation.

The central insight was that government collaboration does not usually fail because the value is unclear. It fails because the system rewards vertical control more consistently than shared outcomes. Participants highlighted the role of ministerial priorities, central agency sponsorship, governance design, funding models, risk appetite and personal relationships in determining whether cross-agency work succeeds or stalls.

Several practical examples emerged. Service NSW was referenced as a case where a simple narrative — making government simpler — helped align mandate, funding and delivery. COVID-era taskforces showed that government can collaborate rapidly when the burning platform is clear. Digital Restart Fund-style models showed the value of centralised assurance and stopping work that is not delivering. Emerging AI use cases, including AI-enabled legacy system analysis, ministerial correspondence automation and shared agent libraries, illustrated both the opportunity and the risk of disconnected experimentation.

The discussion concluded that Government 3.0 will require more than better technology. It needs a leadership model that makes collaboration legitimate, funded, governed and accountable.

2. Key Themes and Insights

Collaboration Breaks When Control is Rewarded More Than Shared Outcomes

A defining theme was that collaboration often breaks down because senior leaders are structurally incentivised to retain control. Government executives are accountable to ministers, budgets, audit processes and public scrutiny. In that context, shared accountability can feel like shared downside with limited personal or organisational upside.

One participant captured this through the example of the “square wheel” problem: people may recognise that a better way exists, but still claim they are too busy to stop and replace the wheel. This “myth of busyness” is not just a time-management issue. It reflects a system where each agency or division is already carrying urgent obligations, and where cross-agency work can appear additional, risky or politically unrewarded.

This challenge is intensified by the kinds of people who often rise through government systems. Participants observed that those who make it to senior levels frequently value control, certainty and command of their resources. These traits are useful in a Westminster system that emphasises accountability, but they can also discourage shared ownership.

The result is a recurring pattern: leaders may support collaboration in principle, but retreat to vertical delivery when risk, funding or ministerial pressure increases.

The Authorising Environment Matters More Than the Collaboration Forum

The conversation repeatedly returned to the importance of the authorising environment. Cross-agency collaboration works best when there is strong political, departmental or central agency sponsorship. Without that sponsorship, forums may exist, but they often lack the authority to make decisions.

Service NSW was cited as a useful historical example. Its success was not simply the result of digital delivery capability. It had a clear narrative — making government simple — and enough political and executive mandate to bring agencies into alignment. Similarly, Victor Dominello’s role in driving digital reform in New South Wales was raised as an example of how ministerial leadership can create permission for ambitious cross-agency change.

Participants also noted that current government priorities, such as housing, can force agencies to the table. When the Premier’s office, ministerial offices, Cabinet Office, Treasury or Premier’s Department are actively involved, collaboration becomes harder to ignore. Without that centre-led mandate, agencies may collaborate at working level but struggle to scale decisions when funding, risk or policy trade-offs arise.

This points to a central Government 3.0 leadership lesson: collaboration cannot be delegated entirely to working groups. It needs visible sponsorship from the centre and decision-making authority embedded in the governance model.

Machinery of Government Changes Create Both Disruption and Excuses

Machinery-of-government changes were viewed as a major barrier to collaboration. Structural changes can shift mandates, leadership, priorities, funding and operating models. They can also interrupt transformation programs just as they are beginning to show value.

One participant described a transformation function that had been building innovative products across a department, only to have its priorities redirected after machinery-of-government changes. Work that had previously been aligned to broader transformation was suddenly refocused on a newly prioritised business area. This kind of shift may be necessary, but it creates friction for teams attempting to scale solutions across multiple areas.

Another example involved a team developing a useful automation that business-level users in another agency wanted to adopt. The moment the work became formal, enterprise architecture, legal and governance barriers emerged. The solution was ultimately blocked because there was no formal clause permitting one department to process another agency’s information, despite the work involving repetitive, mundane, public-facing information processing.

These examples reveal an important distinction. Machinery-of-government structures may define lawful accountability, but they can also become a convenient reason not to collaborate. Government 3.0 requires mechanisms that allow safe cross-agency reuse without requiring each initiative to fight the same structural battle from scratch.

Personal Relationships Still Drive Too Much of the System

A striking insight was the role of personal and strategic relationships in getting things done. Participants referred to the importance of “friends in high places” and “air cover” — senior relationships that can accelerate responses, unblock decisions or create informal permission.

This is both practical and problematic. In complex systems, relationships matter. Trust between senior leaders can accelerate cross-agency work and reduce the need for formal escalation. However, when collaboration depends too heavily on personal relationships, it becomes inconsistent, opaque and difficult to institutionalise.

The discussion also touched on the political layer. Informal groupings of ministers or senior decision-makers may shape priorities outside formal Cabinet or committee structures. If an initiative is not on the agenda of that informal group, it may struggle for attention regardless of its public value.

The challenge for Government 3.0 is therefore to retain the speed and trust of relationships while building more reliable institutional pathways. Collaboration should not require a personal favour, a veiled escalation or the threat of “my CEO will call your CEO” to move forward.

Governance Can Enable Collaboration or Perform It

Governance was discussed as both a necessary enabler and a potential obstacle. Participants were clear that governance is not inherently the problem. Poorly designed governance is the problem.

Large forums may look collaborative because many agencies are represented, but they can become slow, diffuse and performative. When too many people sit around the table without clear roles, decision rights or accountability, the result is often a “nice chat” followed by limited change. Sponsors may not understand their role. Steering committees may defer decisions. Project managers may be left carrying accountability that should sit with executives.

By contrast, effective governance is proportionate, purposeful and decision-oriented. It clarifies who owns the outcome, who carries the risk, who has authority to decide, and when issues need escalation.

COVID was raised as an example where governance shifted rapidly because the burning platform was undeniable. Agencies that might normally hesitate found ways to collaborate, share information and act. The lesson is not that crisis governance should become normal practice. It is that government can move quickly when purpose, authority and urgency align.

The task now is to design governance that enables pace without relying on crisis conditions.

3. Innovative Ideas and Case Studies

Service NSW and the Power of a Simple Public Narrative

Service NSW was referenced as a strong example of reform enabled by a clear, compelling narrative. The proposition was easy to understand: make government simple. That simplicity mattered because it translated complex back-end work into a citizen-facing outcome that ministers, executives and the public could understand.

Behind the scenes, the work required significant collaboration, mandate and funding. Agencies had to give up control over parts of the service experience. Processes had to be centralised. Delivery models had to be redesigned. Yet the narrative created permission for that complexity because it was anchored in a clear public value proposition.

This is a valuable lesson for Government 3.0. Technical language such as “interoperability”, “technical debt”, “data exchange” and “enterprise architecture” often fails to land with political and public audiences. Reform needs a translation layer. The public value must be legible.

For connected government, the equivalent narrative may be less about digital systems and more about simpler, faster, safer and more trusted services.

Funding Models That Force Collective Ownership

Participants discussed several funding models that support shared outcomes. One example involved sector-wide contributions based on proportional interest or usage. For electric vehicle fleet planning, agencies could contribute according to the size of their fleet. For infrastructure assurance, agencies contribute based on the scale of their capital programs.

These models are not always popular when introduced, but they create a practical way to fund shared capabilities without requiring each agency to negotiate separately. They also clarify that some problems are genuinely whole-of-sector problems and should not be funded only by the agency most visibly associated with them.

The Digital Restart Fund was discussed as another important model. Its strength was not only that it created new digital funding, but that it introduced discipline: assurance, prioritisation and the ability to stop work that was not delivering.

A similar model for digital, cyber, AI or cross-government technology capability was suggested as a potential next step. Rather than every agency trying to solve the same issues alone, the sector could pool investment, centralise assurance and prioritise shared capability.

AI Archeology as a Legacy Modernisation Tool

One of the most concrete AI use cases discussed was the use of AI to understand legacy systems. In one example, an agency had systems built decades earlier, with much of the original expertise lost over time. An AI-enabled tool was being used to analyse how the technology had been constructed and effectively produce an instruction manual for the legacy environment.

Participants described this as “AI archeology”: using AI to reverse-engineer, document and eventually support improvements to systems that are poorly understood but operationally critical.

This is a strong example because it targets a real Government 3.0 blocker. Legacy systems are not only technical liabilities; they also prevent interoperability, slow service redesign and create risk when knowledge is concentrated in a small number of people. AI archeology offers a practical bridge between inherited complexity and future transformation.

Shared AI Agent Libraries for Common Government Processes

Another promising idea was the development of a shared AI use case or agent library. Participants noted that many agencies face the same process burdens, such as ministerial correspondence, investment assessment, document review, approvals, triage and policy analysis.

Rather than each agency independently developing similar tools, government could create a shared library of reusable AI agents, patterns or implementation guides. Use cases could be shared without sharing sensitive data. Agencies could learn from each other’s designs, controls, limitations and evaluation results.

Ministerial correspondence was raised as a clear example. Many agencies are likely exploring similar automation opportunities. Without coordination, they will duplicate effort, create inconsistent risk controls and potentially compete for scarce expertise.

A shared library would need governance, incentives and adoption rules. One suggestion was that investment or assurance processes should require agencies to check the library before developing new AI tools, and to contribute lessons back into it. This would move reuse from goodwill to standard practice.

4. Challenges and Barriers

Risk Aversion is Rational in the Current System

Risk aversion was not treated as a personality flaw. Participants recognised that government leaders operate in a system where mistakes can generate ministerial pressure, audit findings, media scrutiny, integrity investigations and public harm. In that context, avoiding risk is often rational.

The issue is that the system frequently treats risk in vertical terms. If an agency owns the decision, it owns the consequences. Shared initiatives can therefore look unattractive unless the accountability model is clear.

This is particularly problematic for AI. Participants recognised the productivity opportunity, but also warned against rushing into AI without understanding the outcome, safeguards or consequences. Robo-debt was invoked as a cautionary reference: automation without appropriate governance, human accountability and policy understanding can cause significant harm.

A more mature model would distinguish between reckless risk-taking and managed innovation. Government needs leaders to take appropriate risks in pursuit of shared outcomes, but those risks must be governed, transparent and proportionate.

AI is Being Pulled Between Hype, Productivity and Governance

AI featured heavily in the discussion, but participants were careful not to treat it as a universal solution. There was broad agreement that agencies should start with outcomes, not AI itself. The right question is not “how are we using AI?” but “what public value are we trying to create, and is AI the appropriate tool?”

Several potential use cases were discussed, including investment assessment, ministerial correspondence, legacy system analysis and repetitive administrative processing. These examples all point to productivity gains: reducing manual workload, increasing speed, improving consistency and freeing humans to focus on higher-value judgement.

However, participants also raised concerns about guardrails. If AI accelerates decision-making, government must be able to show that decisions remain accurate, lawful, explainable and accountable. This requires evidence trails, human review, testing, validation and clear ownership.

One participant argued that the largest guardrail is senior executive understanding. Leaders need hands-on exposure to AI so they understand both its potential and its limits. Reading about AI is not enough. Executives need to use it, test it and ask better questions.

Data Integrity is the Foundation Problem

Participants repeatedly returned to data quality and fragmentation. AI cannot overcome poor data foundations. If agencies have inconsistent, duplicated, outdated or poorly classified data, AI will only accelerate the consequences of that weakness.

The problem is not simply that data sits in different systems. It is that data is often collected for one purpose and later reused for another. Legal permissions, policy intent, metadata, classification and system context all matter. Data that is technically available may not be appropriate or lawful to use in a new context.

This is especially important in whole-of-government settings. Agencies may want to share data, but confidence breaks down when there is uncertainty about source quality, consent, classification or processing authority.

For Government 3.0, the data challenge is therefore not only integration. It is integrity: knowing what the data means, where it came from, what it can be used for, who is accountable for it and whether it is fit for purpose.

Collaboration Often Relies on Threat Rather Than Trust

A concerning pattern discussed was the use of escalation threats to force cooperation. Examples included one agency threatening to go to another agency’s minister, secretary or CEO to secure action.

While this may resolve an immediate transaction, it damages trust. The targeted group may comply, but future collaboration becomes harder. Teams retreat, become defensive and avoid voluntary partnership.

This reflects a deeper issue in the public sector system: some environments operate through fear and threat rather than trust and growth. When poor behaviour is not called out, it becomes normalised. Bad actors remain in place, and collaboration becomes a matter of political leverage rather than shared mission.

Government 3.0 requires a different leadership culture. The system needs stronger norms around constructive escalation, transparent decision-making and shared problem-solving. Collaboration should not depend on coercion.

5. Actionable Outcomes

Immediate Actions

Start with the outcome, not the tool.

Before launching AI, automation or cross-agency digital initiatives, teams should define the public value being pursued. Is the goal faster service delivery, improved accuracy, reduced cost, better compliance, stronger inclusion or safer decision-making? AI should be tested against that outcome, not treated as the outcome itself.

Clarify decision rights in every collaborative forum.

Cross-agency groups should be explicit about what they can decide, what they can recommend and what must be escalated. Without decision rights, collaboration forums become discussion groups rather than delivery mechanisms.

Identify the air cover early.

For cross-agency work, teams should identify the authorising environment at the start. Which minister, secretary, central agency or executive sponsor has the authority to unblock issues? If that sponsorship is absent, the initiative may be limited to local experimentation.

Use shared examples to reduce duplication.

Agencies should actively share use cases, patterns and lessons learned, particularly around AI. Ministerial correspondence, legacy system analysis, investment assessment and administrative triage are likely common enough to justify shared patterns.

Medium-Term Goals

Create a shared AI use case and agent library.

A cross-government library could capture reusable AI agents, solution patterns, governance controls, evaluation methods and implementation lessons. Agencies should be required to check the library before funding new AI work and encouraged to contribute back after implementation.

Build proportionate AI governance pathways.

Not every AI use case should require the same level of governance. Agencies should develop low, medium and high-risk pathways, with common patterns automatically classified as low risk where appropriate. High-risk use cases should receive stronger assurance, but low-risk productivity tools should not be trapped in excessive process.

Develop a digital equivalent of infrastructure assurance.

Participants suggested that digital, cyber and AI investment could benefit from a model similar to infrastructure assurance, where initiatives pass through a common gateway before reaching major funding decisions. This would help prioritise investment, reduce duplication and stop weak initiatives earlier.

Invest in executive AI literacy through hands-on exposure.

Senior leaders need practical understanding of AI. Short, hands-on sessions with small executive groups can build better judgement than passive briefings. Leaders who understand AI’s limits are more likely to ask the right questions about risk, value and accountability.

Long-Term Vision

Move from relationship-based collaboration to institutional collaboration.

Personal relationships will always matter, but Government 3.0 cannot rely on “friends in high places”. The system needs durable mechanisms for cross-agency cooperation, decision-making and funding that survive changes in people, ministers and machinery of government.

Design funding models for shared capability.

Whole-of-sector digital problems should not be solved agency by agency. Funding models could require proportional contributions based on usage, risk, fleet size, capital program scale, transaction volume or service dependency. This would create sustainable investment in shared platforms and assurance.

Treat data integrity as core infrastructure.

Before government can scale AI or agentic services, it must address data quality, classification, lineage, lawful use and system interoperability. Data integrity should be treated as a public sector capability, not a technical housekeeping task.

Create a leadership culture that rewards shared outcomes.

Performance measures, executive expectations and governance structures should reward leaders for contributing to outcomes beyond their own agency. Without this shift, vertical delivery will continue to dominate, even where everyone agrees that connected government is the better model.

6. Conclusion

The Sydney discussion made clear that Government 3.0 is not primarily constrained by technology. The greater challenge is leadership: how government authorises, funds, governs and rewards collaboration across institutional boundaries.

Participants identified familiar barriers — risk aversion, machinery-of-government churn, weak decision rights, fragmented data, unclear accountability and duplication of effort. But they also surfaced practical pathways forward. Shared funding models, central assurance, executive AI literacy, reusable agent libraries, AI archeology and outcome-led governance all offer ways to move from collaboration by exception to collaboration by design.

The most important shift is cultural and structural. Government must stop treating collaboration as extra work and start treating it as the operating model for complex public outcomes. That means creating the conditions where leaders are not punished for sharing risk, teams are not forced to reinvent common solutions, and agencies are not left to solve whole-of-government problems alone.

Government 3.0 will succeed when collaboration is no longer dependent on crisis, personality or political pressure. It will succeed when the system itself makes connected delivery the easiest path.