How your AI agent may level the playing field against companies

AGENTIFICATION. Remember the word. It describes the moment artificial intelligence stopped merely answering our questions and started acting on our behalf – negotiating, cancelling, buying, replacing workers.

On the consumer side, that shift is already in people’s pockets. On the corporate side, it shows up in firms’ payrolls. Together, these two forces are converging on economies from opposite directions, and the collision will determine who captures the value AI creates: households or firms.

Consumers can now have their personal digital assistants in the form of Muse, an app recently released by Meta, which you can get on your phone. It can trawl the internet for you, negotiate deals, complete transactions and even make phone calls on your behalf – and it is free.

xAI has released its own personal digital tool called Grok Bot, ChatGPT offers something similar in its “agent mode” and Google is reportedly working on its own Muse-type app.

On the corporate side, companies can build their own agents from large language models (LLMs) to replicate or augment employee tasks. The field is crowded with several software companies offering agentic capabilities, including Microsoft, IBM and Salesforce.

Agent on your phone is coming for “inattention profits”

Personal digital agents like Muse target inattentive consumers – probably the majority.

These are people who neglect to cancel their auto-renewal subscriptions or fully redeem their frequent flyer miles and reward points, who do not refinance their mortgages at cheaper rates when they can, who fail to negotiate negotiable bills such as medical charges, credit card fees and rents, and who do not find discounts when they are available.

Working in the background 24/7, personal digital agents are able to correct these oversights, bringing the transactions of inattentive consumers closer to those of “maximising consumers” – the minority who are more vigilant.

In a rave review of Muse in The New York Times, Eli Tan described it as “the most useful AI app I had ever used”.

He found that it could efficiently fill out tedious online forms, make phone calls using different personalities, request and obtain refunds, and buy items at half their retail price.

The biggest impact of this optimising behaviour will be to help customers claw back some money from firms and financial institutions, which often make fat profits from consumer inattention, effectively lowering the cost of living and leading to mild disinflation – at least initially.

But firms will not simply absorb such losses; they, too, have tools at their disposal.

Some will block agentic shopping, as online retailer Amazon has done on the grounds that it represents undisclosed third parties using customer accounts and circumvents Amazon’s own recommendation engine. Others may raise prices to make up for lost margins on inattentive consumers.

Some firms may deploy their own agents to ferret out agentic purchases and then charge a different price for those compared with a human browsing the same page.

Alternatively, they could degrade services for agentic purchases through slower fulfilment or fewer options surfaced, which would also protect margins without price discrimination.

So, the money-saving benefits of personal agents may turn out to be transitory, depending on how fast firms adapt their business models and pricing.

Inside firms, different types of agents are already at work

Agents targeted at businesses are more established than personal agents. Some can be created directly from LLMs, as an alternative to packaged platforms like Salesforce’s Agentforce.

This explains why the share prices of software-as-a-service (SaaS) companies like Salesforce and Workday dropped sharply in 2026 after investors worried that AI agents would replace traditional SaaS products, in what came to be called the “SaaSpocalypse”.

But direct-from-LLM agents know nothing about how a particular business runs, which means companies have to do the engineering work to ensure proper governance, permissions and security as well as maintenance. Once this was realised, SaaS stocks recovered much lost ground. The SaaSpocalypse was short-lived.

Salesforce points out that using raw LLMs to create agents has other flaws.

In an interview with me, Paul Carvouni, its senior vice-president and general manager for Asean, said: “LLMs are probabilistic; they predict the next answer. That’s fine for consumer-oriented tasks such as booking flights. But in enterprise work you can’t be ‘probably right’.”

“If a company has to deal with a refund or a banking transaction, it needs to be 100 per cent right. Salesforce brings in all the context, data, rules and governance to transform intelligence into business outcomes.”

The company has released a variety of its own agents with specialised skills that businesses can deploy out of the box and even given them names: for example, Fin for customer experience, Hunter to manage outbound sales, Paige for IT and HR-related work and Marshall to manage supply chain issues.

Salesforce clients report positive results from using its agents. For example, Maxicare, which handles healthcare claims in the Philippines, found that tasks that used to take 20 minutes are now completed by agents in 30 seconds.

“That is profound when you have tens of thousands of requests per month,” said Carvouni.

Anthropic has resolved 79 per cent of its customer experience issues through Fin.

In Singapore, insurer Singlife uses a general insurance e-mail agent designed to help customers get the support they need more quickly and easily. The agent handles over 20 per cent ofcustomer inquiries received via e-mail.

But companies need to do much preparatory work to be “agent ready”. Most importantly, they need to unify their data rather than having it sit in departmental silos.

For example, a finance department would have finance data, a sales department would have sales data, and a human resources department would have HR data. All that data needs to be consolidated into one data set that covers the entirety of the company’s customers, employees and business.

Second, companies need to reimagine their workflows. For example, Carvouni points out: “If you’re a healthcare insurer and your biggest business issue is claims processing, what could that look like if you redo the process from scratch? The process needs to be reimagined.”

Third, companies need to train people to work with agents, which will be embedded in reimagined workflows.

The jobs question nobody can dodge

A big issue relates to jobs. If a single enterprise agent can replicate the work of multiple employees, as seems to be the case, would it not follow that the widespread deployment of such agents would lead to mass job losses?

Techno-optimists think not. They point to how past waves of technological advances have eventually led to new jobs.

This was true of, for example, the Industrial Revolution, electrification, the mechanisation of agriculture, the advent of automobiles, the deployment of ATMs as well as computing and the internet.

In all these cases, new jobs were created in previously unknown areas and economic output also increased, leading to even more jobs. So will it be with AI agents, according to the optimists.

This may be true, but if so, job creation will be a second-order effect, and even there, the pessimistic view is that many of the new jobs will also be done by AI.

But the first-order impact of agentic AI points to job losses, and there is evidence for this.

For example, in its US Securities and Exchange Commission filing in June, tech giant Oracle explicitly named AI as a cause of roughly 21,000 job cuts in 2026.

A Gartner report released in May, based on a survey of 350 executives at companies with revenues of at least US$1 billion that were piloting or deploying AI agents, found that 80 per cent of the organisations represented had reduced headcount.

Singapore’s second-quarter labour market report may also be revealing.

It found that retrenchments increased and were concentrated in sectors that were growing, such as manufacturing and information and communications, as well as financial services.

Moreover, PMETS (professionals, managers, executives and technicians) were the occupational group most affected. The main reason cited for the layoffs was “business reorganisation or restructuring”. It is plausible that piloting or deployment of AI was part of this.

Two revolutions, pulling in opposite directions

In sum, agentification works in different ways. Personal agents like Muse operate on the demand side of consumer markets – they attack firms’ ability to extract rents from consumer inattention and transfer resources from firms to consumers – until firms fight back.

Their impact will be felt sooner, as app downloads can accelerate quickly. They also involve few onboarding challenges for consumers, although there will be privacy concerns: Some may be reluctant to grant agents access to their e-mail accounts, payment credentials or other personal data.

Enterprise agents operate on the supply side – inside firms, where they substitute for labour. They will take longer to be deployed at scale across economies because they need a lot of preparatory work.

Their main impact will be to lower costs and raise productivity, but the benefits will reach consumers only if competition forces firms to pass on the savings in the form of lower prices.

Agentification, in other words, is not one story but two, pulling in opposite directions. Personal agents like Muse attack firms from the outside, arbitraging away the profits companies have long extracted from consumer inattention – a quiet, fast-moving transfer of wealth to consumers, for as long as firms take to fight back.

Enterprise agents attack from the inside, substituting capital for labour inside firms – slower to arrive, but potentially more consequential once they do because they impact livelihoods.

Whether the gains from either revolution ever reach the ordinary household will depend on how much competition survives the age of the agent.

If firms retain their pricing power even as their costs fall, agentification will enrich shareholders, not consumers. If they do not, the AI agent may turn out to be the most quietly redistributive technology in a generation. THE STRAITS TIMES