AI capability is advancing faster than most organisations can absorb it. While many enterprises are experimenting with copilots and agents, far fewer have transformed how work gets done or how value is created. The challenge is no longer access to AI. It is organisational readiness.
Microsoft introduced the Frontier Firm concept in its 2025 Work Trend Index: an organisation structured around on-demand intelligence and human-agent teams. Its 2026 report placed 19% of surveyed AI users in an individual “Frontier” group, where individual and organisational readiness are both high and reinforce one another. That is not the share of companies that are Frontier Firms. [1, 2]
Becoming a Frontier Firm requires more than deploying the latest technology. Business strategy, technology, data, work, knowledge, and governance need to evolve together. Progress in one area quickly exposes constraints in another.
Drawing on emerging research and enterprise adoption challenges, I see 8 interconnected shifts enterprises need to make.
FROM AI USE CASES → TO BUSINESS OUTCOMES
The first wave of enterprise AI asked: Where can we use AI? It created no shortage of ideas. But a catalogueue of use cases is not an AI strategy. The harder question is: Where can AI materially change how the organisation creates value today and tomorrow?
Start with outcomes such as growth, customer experience, operational capacity or risk reduction, then work backwards to the technology. Assess opportunities against value, feasibility and readiness. Is the outcome worth pursuing? Can AI deliver it reliably and economically? Is the organisation ready to make the change?
Saved time doesn't automatically become business value. As an illustrative example, reducing preparation time could let a customer team serve more customers without adding headcount. Measure that extra capacity, not just the hours saved. Evaluate total implementation and operating costs, including integration, maintenance and human review.
The goal isn't to build the biggest AI portfolio. It's to build the right one.
FROM PILOTS → TO SCALABLE OPERATING MODELS
Most enterprises no longer lack AI experiments. The harder problem is turning experimentation into production. A proof of concept can show that AI performs a task. It does not prove that the solution can operate securely, reliably and economically across an enterprise.
Discover → Prioritise → Prove → Industrialise → Adopt → Measure → Scale
This is an iterative pathway, not a one-way procession. In a Stanford Digital Economy Lab study of 51 successful deployments across 41 organisations, 61% included a previous failure. While the selected cases are not representative of every AI project, they reinforce the value of learning and iteration. [3]
Failure becomes useful only when the organisation learns from it. Set success criteria, ownership and the path to production early. Make disciplined decisions about what to stop, change, or scale. A successful pilot is evidence for an investment decision, not the finish line.
FROM DATA READINESS → TO AGENT READINESS
“Get your data ready for AI” remains good advice. But agents raise the bar. When AI can access applications, call APIs, trigger workflows and act, readiness extends to identity, permissions, security, integration and observability.
My recommendation is to modernise these foundations alongside bounded adoption, rather than wait for everything to be perfect. Start within a clearly defined scope. Australia’s ASD Australian Cyber Security Centre recommends phased access and autonomy, least privilege, and limiting agentic AI services to low-risk, non-sensitive tasks. That is a deliberately cautious starting point, not an endorsement of broad autonomy. [4]
A useful readiness test is: Would you be comfortable giving an AI agent access to the systems, information and permissions your employees have today? If not, the constraint may sit elsewhere in the enterprise.
AI does not remove organisational weaknesses. It reveals them.
FROM ISOLATED TOOLS → TO CONNECTED INTELLIGENCE
The next risk is creating another generation of technology silos. Copilots, agents and custom AI need to operate as part of a connected enterprise, not as isolated experiences. That means moving from deploying individual solutions to orchestrating intelligence across systems and workflows.
Many organisations are deploying AI in isolated pockets. A customer insight captured in CRM can inform proposal creation, trigger workflows, support approvals and improve service delivery, all while maintaining governance and role-based access controls. Connected does not mean unrestricted.
The question becomes less “Which model should we standardise on?” and more “How does the right intelligence reach the right work, with the right context and controls?” Design for changing models and capabilities, rather than locking the enterprise around today’s technology.
Frontier Firms will not win because they have one superior model. They will win by connecting intelligence to work more effectively than their competitors.
FROM INDIVIDUAL PRODUCTIVITY → TO HUMAN-AGENT TEAMS
The first wave asked how AI could help people do existing jobs faster. The Frontier Firm asks: If humans and agents could work together, would we design the work the same way? In an illustrative ten-step process, AI might accelerate three steps. Redesign could remove unnecessary hand-offs and reserve human judgement for the decisions that matter.
In Stanford’s selected deployments, escalation-based models, where AI handled at least 80% of work autonomously and people reviewed exceptions, were associated with 71% median productivity gains, compared with 30% where people approved every output. While task type and risk profile influence results, the findings suggest autonomy can unlock greater value when applied appropriately. [3]
Evidence also exists beyond self-reported deployments. The peer-reviewed study Generative AI at Work examined 5,172 customer-support agents and found 15% more issues resolved per hour on average, with less experienced workers improving speed and quality. It is evidence from one setting, not a universal forecast. [5]
Decide deliberately where people should create, decide, approve and supervise, and where agents can research, reason, coordinate and execute. The goal is not maximum automation. It is better work design.
FROM KNOWLEDGE REPOSITORIES → TO ORGANISATIONAL INTELLIGENCE
Not everything an organisation knows exists in a database. It lives in decisions, customer relationships, business rules, conversations and people’s experience. AI creates an opportunity to make more of that knowledge reusable.
But learning is not automatic. An agent receiving feedback does not mean the organisation has learned, or that the agent should retain everything. Useful feedback must be captured, validated and incorporated into governed knowledge systems, with permissions, accountable maintenance and a way to update or retire outdated material.
Work → Insight → Intelligence → Better work
That flywheel is something to design, not assume. Competitors may use similar models, but they do not share the same history, expertise, processes and relationships. Knowledge becomes an advantage only when it can be reused reliably.
FROM COMPLIANCE → TO TRUSTED AUTONOMY
Governance changes when AI moves from answering to acting. What can an agent access? Which tools can it use? What decisions can it make? When must a person intervene? And who remains accountable?
Risk → Autonomy → Controls → Oversight
Make those boundaries tangible: least-privilege access, action logs, approval thresholds and tested rollback or recovery procedures. Define who reviews exceptions and who can stop the system. Greater autonomy may suit low-risk, recoverable work; significant financial, regulatory or customer consequences call for stronger controls and explicit human approval.
The ACSC’s cautious guidance is an important counterweight to the productivity argument. National AI Centre implementation guidance also supports an organisation-wide AI register and human control. Together, these help make responsibility visible rather than assumed. [4, 6]
The objective is trusted autonomy, not autonomy at any cost. As AI moves from answering to acting, governance must move with it.
FROM AI PROJECTS → TO AN ADAPTIVE ORGANISATION
The first seven shifts will not sustain themselves without an organisation designed to support them. That means clear executive ownership, cross-functional decisions, funding, skills and incentives that move AI beyond individual projects. Business, technology, data and security teams need to connect continuously, not only when a project requires them.
Microsoft’s 2026 Work Trend Index found organisational factors were more strongly associated with self-reported AI impact than individual factors. These survey findings do not establish causation, but they reinforce the importance of the organisation around the technology. [2]
There is no permanent state of AI readiness. Models will improve. Agents will become more capable. Today’s frontier will become tomorrow’s baseline. The enduring capability is not becoming ready for today’s AI. It is building an organisation that gets better every time AI does.
Start with one valuable workflow. Establish a baseline, appoint an accountable owner and define autonomy boundaries. Measure outcomes, costs and risks before expanding. Use what you learn to revisit value, feasibility and readiness, then take the next step.
The organisations that succeed will not necessarily be those that adopt AI fastest. They will be those that learn fastest. Every new model, agent and capability creates new opportunities and new constraints. The path to Frontier will look different for every organisation, but the destination is an enterprise capable of continuously turning what AI makes possible into sustained business value.
[1] Microsoft. 2025 Work Trend Index: The year the Frontier Firm is born. https://www.microsoft.com/en-us/worklab/work-trend-index/2025-the-year-the-frontier-firm-is-born
[2] Microsoft. Agents, human agency, and the opportunity for every organisation. 5 May 2026. https://www.microsoft.com/en-us/worklab/work-trend-index/agents-human-agency-and-the-opportunity-for-every-organisation
[3] Pereira, Graylin and Brynjolfsson. The Enterprise AI Playbook: Lessons from 51 Successful Deployments. Stanford Digital Economy Lab, April 2026. https://digitaleconomy.stanford.edu/app/uploads/2026/03/EnterpriseAIPlaybook_PereiraGraylinBrynjolfsson.pdf
[4] ASD Australian Cyber Security Centre. Careful adoption of agentic AI services. 1 May 2026. https://www.cyber.gov.au/business-government/secure-design/artificial-intelligence/careful-adoption-of-agentic-ai-services
[5] Brynjolfsson, Li and Raymond. Generative AI at Work. Quarterly Journal of Economics 140(2), 2025, pp. 889–942. https://academic.oup.com/qje/article/140/2/889/7990658
[6] National AI Centre. Guidance for AI adoption: implementation guidance. 5 May 2026. https://www.ai.gov.au/staying-safe-and-responsible/essential-ai-practises/guidance-ai-adoption-implementation-guidance
Turn AI potential into business value. Talk to SoftwareOne about your readiness, governance and first workflow to scale.
Turn AI potential into business value. Talk to SoftwareOne about your readiness, governance and first workflow to scale.
