By year-end 2030, 80% of frontline workers employed by international companies will be assisted by physical AI systems
“Your physical AI co-worker is coming,” Plummer says. “We have all talked in the last few years about AI co-workers. They’re sitting in a box somewhere and they’re acting digitally, with an avatar.
“Now we’re seeing the world on the verge of having those physical robots walking around our offices amongst us.”
Various viral YouTube videos feature humanoid robots, but Plummer says, “It’s actually more likely to look like Star Wars with lots of tiny little mouse robots and the little box robots.”
Plummer adds, “Routine tasks are going to be delegated to robots, drones, and sensors. So they essentially become a working class to support us over time, and we get them to do things that we don’t want to do.”
Many industries are already using physical AI to improve safety, automate repetitive and hazardous work, and gain operational insights.
Gartner recommends organisations take a strategic, safety-first approach by investing in scalable platforms, robust governance frameworks and the skills needed to deploy and manage physical AI systems effectively.
There will be privacy challenges and business optimisation opportunities around the huge volumes of data collected by physical AI systems.
“Your best actions are to prioritise investments in modular, reusable control policies, edge processing, virtual simulation, and physical AI accessibility. How does a box robot get down the stairs?” Plummer said.
(Kiwi-led start-up Antioch recently raised $54m for its software platform that allows for virtual testing of physical AI robots.)
“These are things that you should prioritise now because they’ll get you there faster.
“Your worst action is that you will replace people with robot workers expecting a return on investment. Replacing human beings with AI or robots is always a bad choice.”
Plummer added, “So physical AI is going to create new human and machine relationships.
“Do you want a human-formed robot coming into your laundry at home and using your washer and dryer to wash clothes, to fold and press and fold them and hang them, or would you rather have a washing machine that you just throw dirty clothes at? It catches it, washes it, folds it, and spits it out in the basket.
“You have to decide which one you’ll like, because the relationship could be either one when you try to anthropomorphise the robot, or one where you just see it as a tool. Seeing it as a tool is not a bad thing. If we anthropomorphise it, we will fall in love with it eventually. That’s going to happen.”
2. Agent slop overwhelms public services
By the end of 2030, more than 10 billion autonomous agents created by people, companies and governments will clog public services.
Today, it tends to be only geekier power users who are setting up AI agents, which can be set a task that they carry out themselves.
Even now, this can have surprising results. Plummer points to an incident in Australia last month where a man, “Andrew”, asked an AI agent to book him into a gym class.
When the agent discovered there was a wait-list, it exploited a weakness in the gym’s booking software to delete people who were ahead of him in the queue.
But in years ahead, as agents become easier and easier for members of the public to unleash, Gartner sees an avalanche of agents.
It predicts a significant increase in demand for public services by enabling autonomous agents to identify opportunities, determine eligibility and submit requests on behalf of people with minimal human effort.
This surge in applications, claims and transactions will put pressure on government systems and require stronger approaches to identity, trust and accountability.
Gartner recommends governments prepare by modernising digital infrastructure, strengthening verification capabilities and planning for higher volumes of AI-mediated interactions.
It’s not just public services that will get swamped. Plummer also sees someone after tickets to a big game or concert sending hundreds of agents to act on their behalf.
Plummer said Government departments and retailers would have to deploy thousands of “counter-agents” to grapple with the flood of hurdle- and queue-jumping agents coming their way.
But at the same time, retailers will have to retool their sites to attract tireless AI agents as much as humans, Plummer says.
Changes are already underway. Last year – while still chief digital officer – Air New Zealand boss Nikhil Ravishankar said that with upgrades, he now had to consider how his airline’s website and app appealed to AI agents as much as humans.
3. Cost-exhaustion attacks will become a critical threat to AI systems
By year-end 2030, 80% of organisations with public-facing AI will have experienced a cost-exhaustion attack creating excessive AI cost.
Today, we have denial-of-service attacks, where hackers send swarms of bots to connect to a site, crowding out legitimate users like so many sheep blocking a road.
Now imagine thousands or millions of AI agents tie up your site at once, not only blocking your human customers but costing you tokens (billable units of AI consumption) as you deal with each.
The rise of public-facing AI is creating a new cybersecurity risk called cost exhaustion attacks, where malicious actors deliberately drive excessive AI use to increase operational costs.
As AI becomes more embedded in customer-facing applications, organisations will need to treat token consumption and AI use patterns as both cost management and security concerns.
Gartner recommends organisations make AI token costs a cybersecurity indicator. A sudden spike in token use might indicate an employee goofing around trying to give a bald prime minister 100 different hairstyles. It could also be a marker of an external cost-exhaustion attack.
It also says to implement cost-focused security controls, and extend monitoring capabilities across all AI technologies.
4. A dedicated function for tracking AI costs to value
By 2029, 60% of organisations deploying AI will establish a dedicated function responsible for mapping AI total cost to value or profit.
Gartner says at the moment, many organisations see the likes of:
- A single task can end up costing 100 times more than expected because of endless retries
- “Context window inflation”, where poorly written or sequenced prompts mean an AI has to reread the entire history of a conversation and all attached documents
- Frontier model overprovisioning: Using the most powerful AI model for simple jobs wastes money on power you don’t actually need, when the likes of an open-source, free or cheap Chinese model or a specialised “small language model” would be a better choice.
Plummer says the answer is to install “gateway” software that helps choose the right AI for the right task and “guardian” agents that keep tabs on other agents’ behaviour.
5. The rise of ‘burner’ apps
By 2029, 80% of new applications will be intentionally disposable – used for less than one year – revolutionising the software lifecycle.
AI is making application development so accessible that employees will increasingly create temporary applications to meet short-term business needs.
Organisations face new governance, security, compliance and records management challenges, particularly when they influence decisions or access sensitive data.
Gartner recommends organisations establish risk-based governance frameworks, monitor business-created applications through automated registries, and update records retention policies to address AI-generated applications and agents.
6. Private power providers
By 2030, US$10 trillion in enterprise-owned energy will make the world’s 2000 largest companies unexpected power providers – selling to grids and AI data centres, reshaping the utility industry.
“Energy is a rising concern,” Plummer said.
“Around Atlanta, where I live, power bills have gone up 26% on average because data centres are sucking up all the energy and driving the prices.
“By 2030, global 2000 firms [the world’s 2000 largest companies] will be power providers because it’s cheaper for them to do so.
“They’re going to start generating their own power and sell the excess to grids and AI data centres and reshape the utility industry.”
Projects spanned from the big – Microsoft’s plan to reboot long-abandoned nuclear reactors on Three Mile Island – to the more everyday scale (solar panels on the roofs of company warehouses and factories) to the emerging field of businesses creating their own banks of batteries, controlled by increasingly sophisticated software for buying energy or selling it back to the grid at the best time.
Surging electricity demand from AI and data centres is driving enterprises to invest heavily in energy generation, storage and management assets, blurring the traditional line between energy consumers and power providers, Plummer said.
As energy becomes a strategic, software-defined asset, organisations will need new capabilities to optimise production, storage and consumption.
Power was no longer the domain of a firm’s facilities manager. The CFO and CTO also had to get into the mix.
Plummer said some of the power predictions were best applied to the US, but there is a worldwide trend towards self-generation with solar, and batteries like those recently launched by Aotea Energy in New Zealand that treat the battery as a commodity to store power (with solar panels not necessarily in the mix) and sophisticated software for power consumers to act as their own gentailer as they buy, store or sell electricity at different times of the day.
7. Insurers drive AI governance
By 2030, insurers – not regulators – will drive AI governance, as strict underwriting standards for AI liability insurance will be required to reduce insurance costs.
Organisations are shifting from policy-based AI governance to operational governance that embeds controls directly into AI systems and workflows, Gartner says.
As AI risks and liability concerns grow, insurers are expected to influence governance practices by encouraging stronger oversight, risk management and technical controls.
Gartner recommends organisations implement AI governance technologies, test controls in real environments, and establish runtime oversight to support both innovation and accountability.
In the here and now, however, Plummer said AI insurance was still very much an emerging area. Only 1% of large organisations had dedicated AI liability insurance. It’s more typical for the new technology to be covered between cyber security, general liability and practice liability policies.
It was also a moving feast, with AI agents entering the mainstream over just the past few months, and the rise of physical robot AI pending.
But there was also a more timeless theme.
“Insurance companies will fight to not pay. We know the underwriters will drive the premiums up. We know that the risk may be higher than we expected,” Plummer said.
“But we have to put it in place to potentially handle those white swan, black swan events.”
8. Burner business: Fast followers made obsolete
By 2029, 25% of Fortune Global 500 companies will continuously innovate componentised AI-powered offerings, creating a competitive moat that obsoletes fast-follower strategies.
The idea here is that market leaders will reinvent themselves every six months. That means firms who today adopt a “fast-follower” strategy will no longer quickly catch up after watching the first-mover do all the legwork, Plummer says.
As AI-native competitors introduce more customised, efficient and scalable offerings, established organisations face growing pressure to innovate or risk losing customers, talent and market share, Gartner says.
Although you want to rein in token costs, you also don’t want to overly prioritise controls and internal efficiency gains at the expense of innovation.
9. Tracking token consumption in real time
By 2028, 60% of Fortune Global 500 companies will embed AI FinOps (financial operations) control in inference (AI-speak for when a AI prompt is entered), shifting cost governance from reactive reporting to real-time optimisation.
The Herald asked Gartner analyst Robert Naegle, an expert on AI costs, a couple of simple questions: “What is a token?” and “Is the price of tokens going up or down?”
“We did a piece of research looking at the top 20 AI tools,” he said.
“What we found is they all use the word ‘token’, but for all 20 it means something different.
“So the price per token is different. The consumption requirement of token is different.
“For a typical task, it’ll use a certain number of tokens, input tokens and output tokens, and so the tokenisation of consumption is really complex.”
A single prompt typically involves many tokens – or billable units, as AI models continue their shift to consumption-based pricing rather than a set per-user, per-month fee.
The net result: “AI spending is becoming harder to predict and control as organisations scale AI tools and agentic workflows, driving a shift from retrospective cost reporting to real-time AI FinOps [financial operations] governance.”
As AI becomes a larger operational expense, organisations will increasingly focus on measuring cost per task and token efficiency to maintain margins and maximise value, Gartner says.
Gartner recommends organisations implement runtime cost controls, deploy inference-path telemetry and make cost governance a core requirement of AI platforms and applications.
In plainer language: keep tabs on things so people don’t use expensive AI tools for simple tasks.
“One of the things we recommend to clients is that you budget based on value rather than based on cost of input. So think about what’s the value AI creates. What’s that value worth to you, and then spend according to value.”
In practical terms, that could mean capping individual employees’ monthly spending on AI tokens or other charges.
Meanwhile, staff at big organisations with consumption-based AI pricing also had to realise that careless consumption costs money, Plummer said.
So does polite consumption. If you always type “thank you” after an AI delivers a response or completes a task, that’s costing your employer money. (This author does wonder if it’s a worthwhile hedge for the day when robots take over.)
10. New evidence custodians
By 2030, 80% of the Fortune Global 500 will contractually make their CIO (or CAIO – chief AI officer) the “evidence custodian” for AI accountability.
As AI becomes embedded in critical business processes and decisions, organisations face increasing pressure to ensure accountability, transparency and oversight of AI actions, Gartner says.
This is expanding the role of technology leaders in overseeing AI accountability and ensuring governance guardrails are applied consistently across the organisation.
Gartner recommends organisations assess their digital evidence management capabilities and establish clear accountability for AI actions to support responsible AI governance.
What’s the end point, in an age when AI can cause commercial damage or even potentially death?
“Somebody’s going to jail,” Plummer said, channelling a lyric from Don Henley’s New York Minute.
“You need to know who’s going to jail – even if you plan to defend them in court so they don’t go to jail.”
Chris Keall is an Auckland-based member of the Herald’s business team. He joined the Herald in 2018 and is the technology editor and a senior business writer.
Chris Keall travelled to the Gold Coast courtesy of Gartner.
