What You’ll Learn
- How should organizations measure AI agent value?
Organizations should shift from measuring cost per token to cost per business outcome, evaluating whether agent benefits justify the expenses. - Why do AI agents need assigned business owners?
Every significant AI agent requires a business owner who understands its purpose, data access, permitted actions, monitoring, and lifecycle. - What infrastructure supports responsible AI agent deployment?
Organizations need agent control towers, AI gateways, and governed APIs that provide cost visibility, enforce role-based access controls, and enable human approval for consequential actions.
Over the past year, I have spoken with customers, analysts, and industry leaders across Asia Pacific and Japan about the opportunities and challenges of bringing AI into everyday business operations.
That conversation is changing.
If 2025 was about preparing for AI and 2026 was about embedding agents into enterprise workflows, 2027 will be about proving their value and establishing who is accountable for what they do.
The urgency is clear across APJ. Research by Omdia, commissioned by Boomi, found that 74% of organisations across Australia, New Zealand, Singapore, Malaysia, and the Philippines have active AI initiatives. Yet only 46% have adopted a platform-led approach to integration, highlighting the gap between AI ambition and the infrastructure needed to support it.1
Meanwhile, financial constraints remain the region’s leading business challenge. Organisations are investing in AI while facing increasing pressure to demonstrate measurable returns.
As we look towards 2027, I believe two developments will define the next chapter: the emergence of AI Economics, and the need for every agent to have an accountable owner.
Prediction 1: AI Economics Becomes a New P&L Line
Enterprise AI has largely been managed as another technology expense, tracked through licences, consumption, and the number of projects launched. That won’t be enough in 2027.
As agentic AI moves into core operations, boards will demand the financial discipline they already apply to cloud computing. Businesses must understand what each task costs and whether the outcome justifies it. The FinOps Foundation found that 98% of surveyed practitioners now manage AI spending, compared with 31% two years earlier. Yet visibility into AI costs, allocating expenses to business units, and demonstrating financial returns remain significant challenges.2
The gap is especially apparent in APJ. Half of Philippine organisations have defined KPIs for AI success, against a regional average of 37%. Meanwhile, 28% of Australian and 34% of New Zealand organisations cannot effectively measure whether their AI initiatives are working.3
That’s why I believe 2027 will mark a turning point for AI Economics, as organisations move beyond tracking what AI consumes to proving what it delivers. The key measure will shift from cost per token to cost per business outcome.
Consider what happens when an AI agent moves into production. A task that looked cheap in a demonstration might require multiple model calls, several enterprise systems, and repeated attempts to act. Add monitoring, governance, and human intervention, and the economics change considerably. The question becomes whether all that intelligence is necessary at all.
In a recent media interview, I raised a simple example: why would I want an AI agent deciding whether I get paid at the end of the month? Payroll is deterministic. AI earns its place when work involves ambiguity or unstructured information. For predictable transactions, conventional automation may remain faster, cheaper, and more reliable.
New Zealand infrastructure company HEB Construction shows the more deliberate approach. Its agents turn employee CVs into reusable, trusted information and help tender teams tailor bids. HEB saves hours on each tender, and employees still approve the final content.4
I particularly like this example, because the value comes from solving a specific problem, not from deploying AI for its own sake.
In 2027, I expect more APJ enterprises to make similar distinctions. They will combine frontier models, smaller models, deterministic automation, and human judgement according to each task’s cost, risk, and complexity. AI gateways will show usage, latency, and operating costs while routing requests to the right model.
The financial consequences extend beyond consumption. A Forrester Consulting study commissioned by Boomi found that organisations in a state of “agentic chaos,” deploying agents without adequate governance, integration, and operational controls, risk an average of US$2.1 million in additional costs.5 The true cost of AI is not only what an agent consumes but also what happens when it gets something wrong.
Some high-profile organisations will scale back their AI initiatives, not because the technology failed but because they could not show the economics made sense. AI stops being magic in 2027. It starts being math.
Prediction 2: Someone Has to Own the Agent
Organisations have given agents growing access to applications, customer data, and financial systems. Agents are moving beyond generating answers to initiating actions, which creates an entirely different set of risks.
A July 2026 Forrester Consulting study commissioned by Boomi found that 86% of organisations had moved beyond agent pilots, yet only 34% trusted the actions their agents were taking.6 The question is no longer only whether a model gives an accurate answer, but who is responsible when an agent’s actions have unintended consequences.
Earlier this year, I heard about one organisation that gives each AI agent an employee number and registers it in its HR system. Agents are onboarded, overseen, and offboarded like staff. It may sound unusual, but employees have identities, managers, defined roles, and permissions that change and are revoked when they leave. Why should agents operating in those same systems be any different?
Every agent operating at meaningful scale will require a named business owner who understands its purpose, its data access, its permitted actions, how it is monitored, and when it should be restricted or retired. Accountability cannot exist only on paper. Organisations will need agent control towers, along with AI gateways and governed APIs that enforce identity, access permissions, rate limits, and human approval at runtime.
In a recent Asia Tech Podcast discussion, my colleague Amy McNee noted that the Terminator was reduced to searching a phone book because he lacked reliable information about which Sarah Connor he needed to find.7 Human employees compensate for poor data with experience and judgement. Agents do not necessarily have that instinct, and they can act confidently on bad information at machine speed. Trusted data, real-time integration, and observability are therefore essential.
Governments are responding. Singapore’s Infocomm Media Development Authority introduced a Model AI Governance Framework for Agentic AI in January 2026, updated in May, emphasising human accountability and approval checkpoints for consequential actions.8 Malaysia is moving in a similar direction, with a proposed AI Governance Bill9 and a Boardroom Primer urging directors to take responsibility for AI risk.10 The message is becoming clear across APJ: Responsibility for autonomous AI must extend from technology teams into business leadership.
The principle I’ve advocated for is straightforward: Agents should inherit their permissions, not invent them.
Agents need their own identities and narrowly scoped access, drawn from approved business policies. By the end of 2027, I believe the measure of a mature AI programme will be whether every significant agent has a defined role, a named owner, an auditable identity, and an accountable lifecycle.
A Business Case and a Business Owner
In 2027, I believe the conversation will shift from making AI more powerful to making it accountable, both financially and operationally. An agent without a clear cost picture becomes an expensive experiment. An agent without defined ownership and trusted data becomes a business risk. A more powerful model solves neither.
This is where enterprise AI infrastructure will play an increasingly important role. AI gateways will help organisations understand and control model consumption and costs. Agent control towers, governed APIs, and integration platforms will provide the visibility and operational controls needed to manage agents across different systems and environments.
For APJ enterprises navigating diverse markets, economic pressures, and evolving governance expectations, the ability to combine financial discipline with responsible autonomy will become a competitive advantage.
In 2027, every AI agent will need two things: a business case and a business owner. The winners will not be the organisations running the most agents or the largest models. They will be the ones that can confidently explain why each agent exists, what value it delivers, and who is responsible for what it does.
——
- Omdia, AI Ambition Meets Data Reality: APAC Technology Priorities and Challenges 2026 (research commissioned by Boomi, 2026), https://boomi.com/content/report/apac-tech-priorities-ai-2026 and companion blog, AI Ambition Meets Data Reality for APAC: Data Infrastructure Must Support Growth, David Irecki, https://boomi.com/blog/ai-ambition-meets-data-reality
- FinOps Foundation, State of FinOps 2026, https://data.finops.org
- Omdia, AI Ambition Meets Data Reality.
- Boomi, “HEB Construction Optimizes Project Bidding With Specialized AI Agents,” https://boomi.com/customer/heb-construction
- Forrester Consulting, The Agentic AI Readiness Gap (commissioned by Boomi, July 2026), https://boomi.com/resources/resources-library/agentic-chaos-forrester-study
- Forrester Consulting, The Agentic AI Readiness Gap.
- Asia Tech Podcast, “What Happens When an AI Agent Confidently Acts With Wrong Data?” featuring David Irecki and Amy McNee, September 16, 2026, https://atpinsights.substack.com/p/what-happens-when-an-ai-agent-confidently
- Infocomm Media Development Authority, Model AI Governance Framework for Agentic AI (Singapore, January 2026; updated May 2026), https://www.imda.gov.sg/resources/press-releases-factsheets-and-speeches/press-releases/2026/new-model-ai-governance-framework-for-agentic-ai
- Malaysia Ministry of Digital and National AI Office, “Public Consultation on the Proposed AI Governance Bill,” July–August 2026, https://upc.mpc.gov.my/view-consultation/264
- AI Malaysia, Boardroom Primer to AI Adoption and Governance (July 2026), https://ai.gov.my/governance/boardroom-primer-to-ai-adoption-and-governance