The AI agent problem that's holding back innovation
We are witnessing the emergence of truly agentic AI – systems that operate with increasing autonomy and goal-directed behaviour rather than merely responding to prompts. However, as these impressive capabilities mature, a critical infrastructure gap threatens to undermine their potential at scale.
The problem is straightforward but profound: most AI agents today are confined to proprietary technological stacks. They rely on platform-specific memory stores, orchestration logic, toolchains and interaction schemas that work perfectly well in isolated environments, but create serious friction when deployed across organisational boundaries.
When agents cannot communicate, share context or delegate tasks across these boundaries, they become technological silos. This growing collection of disconnected, brittle implementations cannot be composed into the broader system of intelligence that enterprises require for meaningful automation.
The cost of this non-interoperability is not theoretical or distant; it is immediate and practical. Without interoperability standards, agentic automation inevitably stalls at the boundaries of vendor ecosystems.
Consider what happens when a business process spans multiple systems, from customer interaction to order fulfilment, or across supply chain and finance departments. Agents built by different vendors cannot share crucial context, delegate subtasks or coordinate actions effectively. As a result, cross-domain AI-driven automation, which should be the primary value proposition of agentic systems, becomes unworkable in practice.
BT in your inbox
Start and end each day with the latest news stories and analyses delivered straight to your inbox.
Sign Up
Sign Up
AI agents must interoperate across nine dimensions
Solving agent interoperability requires addressing multiple interconnected layers, not just creating superficial application programming interface (API) connections. If agents are to coordinate effectively across organisational, technical and vendor boundaries, we need a common substrate for how they discover each other, delegate tasks, share knowledge and enforce constraints.
Tool use represents the most visible dimension, where agents must invoke external services, automation platforms and APIs without relying on brittle hard-coded logic.
Anthropic's Model Context Protocol, for instance, tackles certain aspects of tool access. Communication protocols are equally critical, as agents need structured dialogue mechanisms to assign roles, exchange information, and negotiate task handoffs between systems. Cisco and LangChain's AGNTCY standard and Google's brand new A2A framework provide initial approaches to how agents may talk to each other.
But the requirements quickly expand beyond these basics. Trust and security become foundational concerns in multi-agent environments, requiring shared models for identity verification and authorisation. Memory systems must provide persistent, cross-context storage that agents can access across organisational boundaries.
Knowledge sharing demands mechanisms not just for information access, but for verification and integration of information across different agent systems. Looking toward more advanced interoperability, transaction capabilities become essential for agents to negotiate, purchase and compensate across digital marketplaces.
Stripe has developed preliminary agent toolkits for economic exchange. Governance frameworks must move beyond static policies to become executable constraints embedded directly in agent workflows. Discovery protocols are needed so agents can dynamically locate and interact with other relevant agents in a secure way, without relying solely on preconfigured integrations.
Finally, distributed agent systems require standardised approaches to error handling, where failures can be systematically surfaced, escalated or resolved.
These are some of the promising early efforts addressing pieces of this puzzle. However, they remain point solutions in different stages of maturity, enterprise readiness and adoption levels, indicating that a more comprehensive and coordinated approach is needed for true interoperability.
AI agents are path forward for global innovation
Interoperability should not be viewed as the enemy of commercial differentiation but rather, as its enabler. Open ecosystems do not diminish commercial opportunity; they expand it dramatically. When cloud platforms adopted container standards and interoperable storage APIs, they did not sacrifice competitive advantage; instead, they created a larger surface area for value-added innovation. The same principle applies to agentic AI systems.
Vendors supporting interoperable agent frameworks can still compete vigorously on orchestration strategies, domain specialisation (for example, offering agents pre-trained for specific industries like healthcare or finance), tooling sophistication and developer experience.
But they compete within a viable, expanding ecosystem rather than through isolation tactics that ultimately limit total market growth. For enterprise customers, the advantages of interoperability are even more compelling, enabling automation that spans organisational boundaries without requiring wholesale commitment to a single vendor's vision.
Asia-Pacific could be a key proving ground for agentic AI
With its unique combination of cross-border trade complexity, regulatory diversity and vibrant consumer services sector, Asia-Pacific creates real-world conditions that demand adaptive, autonomous systems. This unique combination suggests that the Asia-Pacific region could serve as a critical proving ground for these interoperable agentic technologies.
Agentic AI will find some of its most valuable use cases in environments where integration across languages, infrastructure types and regulatory jurisdictions is not optional but essential for business operations.
Achieving true interoperability will require more than individual product road maps or incremental feature development. It necessitates a coalition approach spanning technology vendors, industry associations, standards bodies, regulatory agencies and enterprise technology buyers. Interoperability must be treated as strategic infrastructure, not relegated to an afterthought or post-implementation integration problem.
The fundamental choice facing the AI industry is not between control and openness. It is between creating composable, scalable intelligence versus isolated automation silos. Those who solve for the complex interoperability requirements of diverse markets like Asia-Pacific will likely establish the blueprint for agentic ecosystems globally. And the time to make these architectural decisions is now, before proprietary approaches calcify into barriers we cannot undo.
The writer is a principal analyst at Forrester
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Business Times
an hour ago
- Business Times
BlackRock hits record US$12.5 trillion in assets amid tariff chaos
[NEW YORK] BlackRock pulled in US$46 billion to its investment funds, and assets hit a record of US$12.5 trillion as clients rode out the volatility of US President Donald Trump's tariff policies in the second quarter. Investors added US$85 billion to exchange-traded funds and US$29 billion to equities overall, New York-based BlackRock said in a statement on Tuesday (Jul 15). Net flows into long-term investments missed the US$61 billion average estimate of analysts surveyed by Bloomberg, as a single institutional client redeemed US$52 billion from a lower-fee index product. 'Our expanding client relationships are resonating in higher, more diversified organic base fee growth,' chief executive officer Larry Fink said in the statement. Overall net flows into the company's funds were US$68 billion, including US$22 billion to cash-management and money-market funds. At the start of the second quarter, Trump's announcement of unexpectedly stringent tariffs sent global stock markets plunging and led to convulsions in bond markets – at one point rivaling the volatility of the 2008 financial crisis and onset of the pandemic in 2020. About a week later, the president issued a 90-day pause on tariffs for dozens of trading partners. After the anxiety subsided, investors added to stocks and bonds as the S&P 500 and global indexes quickly recovered. BT in your inbox Start and end each day with the latest news stories and analyses delivered straight to your inbox. Sign Up Sign Up BlackRock also added US$9.8 billion in alternative investments in the period. Known for years primarily for its low-fee stock and bond funds, the world's largest asset manager continues to push into private markets and compete with the likes of Blackstone, Apollo Global Management and KKR . to reach institutional and, increasingly, retail investors. In his latest annual letter to investors, Fink said the firm is no longer a 'traditional asset manager' and pledged to bring private assets to the masses. Just after the quarter ended, BlackRock completed its US$12 billion acquisition of private-credit shop HPS Investment Partners, its third major purchase in about 18 months. As part of that US$28 billion buying spree, BlackRock also struck deals to buy Global Infrastructure Partners and private-markets data firm Preqin. With the acquisitions, BlackRock will manage more than US$600 billion of alternative investments, and the company has set a goal to raise another US$400 billion in private assets by 2030. BlackRock's adjusted earnings per share in the quarter 16 per cent from a year ago to US$12.05. That beat the average analyst estimate of US$10.87. Revenue rose 13 per cent to US$5.4 billion from a year ago, reflecting higher fees tied to GIP and the positive effect of markets. That was partially offset by lower performance fees. Shares of BlackRock rose 8.4 per cent through Monday, beating the 6.6 per cent increase of the S&P 500. BLOOMBERG
Business Times
an hour ago
- Business Times
Investment myths in a fast-moving stock market
[SINGAPORE] Welcome to the most volatile decade in recent memory – and we're not even halfway through. The 2020s have already recorded 440 trading days with daily stock movements of 1 per cent or more, according to wealth manager Ben Carlson. To put that in perspective, an entire decade typically averages just 507 such days. Said another way, the 2020s have crammed nearly 10 years of stock market volatility into less than five years – an unusually heavy dose of ups and downs. So, if you feel the stock market is moving faster, you now know why. When volatility teaches the wrong lessons When markets are volatile, the common advice is to simply tune out the noise, but that's easier said than done. The problem lies in the way our brain processes information. In the book Your Money and Your Brain, author Jason Zweig said our brains are wired to recognise patterns even when none exist. Here's the rub: It's not a mechanism you can switch off at will. In other words, by watching daily stock price movements, you may be receiving the wrong signals and end up learning the wrong lessons. That's a big problem as key misconceptions may start to take shape. Take the sudden rise of China's DeepSeek AI model back in January. When the news broke, AI-related stocks were hammered, with Nvidia bearing the brunt, suffering a 17 per cent fall in a single day. BT in your inbox Start and end each day with the latest news stories and analyses delivered straight to your inbox. Sign Up Sign Up Such a sequence suggests you need to react fast to avoid such a mishap. Furthermore, it also implies the need to instantly assess new developments and act to prevent further losses. But that's not how it played out. In both cases, investors were misled into thinking that speed is the difference between making and losing money. The misguided need for speed In the hours and days after DeepSeek became mainstream, the narrative quickly centred around the Chinese AI model's development cost – under US$6 million. Why was this figure important? This low expense stood in contrast to the billions of dollars US companies were pouring into artificial intelligence (AI) models and data centres, with Nvidia taking the lion's share. Hence, the prevailing narrative suggested that the company would have the most to lose. Yet, six months later, it's becoming clear that the hasty assessment on DeepSeek is premature in more ways than one. One, DeepSeek's US$6 million development cost does not cover any prior research and experiment on data, architectures and algorithms. The actual cost is higher but unknown, effectively knocking out the main narrative. Two, major tech companies such as Alphabet, Amazon, Meta Platforms and Microsoft have continued to pour billions into data centres. More importantly, much of this spending has flowed to Nvidia. For the past two quarters, the GPU provider reported revenue gains of 78 per cent and 69 per cent year on year, respectively. To top it off, Nvidia shares have risen by 39 per cent from their January low. In other words, those who were looking to save a few dollars by exiting fast ended up leaving a lot more money on the table. Simply waiting for a couple of quarters would have done the trick. Avoiding the pitfalls of volatile markets But if moving fast is not the answer, then what should investors do instead? In my mind, there are three useful guidelines to follow. For starters, most major declines are accompanied by negative headlines. To be clear, it's not about the bad news itself. The real problem lies in the avalanche of negativity that follows, ranging from detailed articles on what went wrong to podcasts and rapid-fire social media takes. Amid the onslaught, any good news is buried and often left unmentioned. This predicament is best exemplified when Netflix lost a million subscribers in H1 2022. In the aftermath, the negative news was unrelenting, with many pointing to the crowded field of competitors and the lack of quality content. Impatient investors reacted by dumping Netflix shares, sending the stock down by 75 per cent from its 2021 peak. But here's the truth: If you believe that you shouldn't fall in love with a stock, then you should also accept that you shouldn't fall into hate with a stock. Amid the pessimism, the naysayers overlooked the positives. In particular, Netflix had a dual strategy in place to launch an ad-supported plan while cracking down on password sharing. Within six months, its new ad-supported option was launched. By the end of 2024, the company's subscriptions surpassed 300 million users, adding over 80 million new members since losing a million in H1 2022. In fact, if you zoom out and look at the loss of subscribers between January and June 2022, it would look like a minor blip in the upward trend. The lesson here is clear. When everyone is eager to tell you the bad news, you have to look for the good news yourself. Let time be the ultimate judge Not every piece of news is worthy of your attention, even when everyone's talking about it. So, here's the next thing to keep in mind: There's a limit to what you can focus on, so don't give your attention away cheaply. The current developments in the AI space are a good example. When OpenAI's ChatGPT caught fire in late 2022, the reactions went on overdrive, prophesying on the future outcomes. The problem, however, is that the outcomes suggested often veer towards the extreme. For instance, some commenters predicted the demise of Google search. The situation is further exacerbated by Microsoft's launch of an AI-powered search. When faced with such a dilemma, remember this: Let time be the judge. In my experience, disruptions do not happen immediately. In addition, there is always time for the incumbent to respond. Furthermore, the AI landscape could also play out in ways we cannot imagine ahead of time. OpenAI's recent deal with Google Cloud services is a good example. This outcome shows there can be more than one winner. You simply have to be patient and let time be the judge. Get smart: There's always time to add a winner Much of the rush to act is based on the fear of missing out, that if you don't invest now, you are sure to miss out on a huge winner in the future. Nothing could be further from the truth. For instance, you could miss the iPhone's launch by a decade and still net a nearly 470 per cent return by investing in Apple 10 years later. Sure, the gain wouldn't be as good as investing at the time of the introduction of the iPhone. But have you seen an unhappy investor with a 4.7x gain? To summarise, as the stock market fluctuates, don't let your emotions flutter along with the ride. Instead, keep these principles in mind: It's not about giving your attention to every single bit of news, it's about figuring out which new development is worth your time. It's not about how fast you react to a new business development, it's about whether or not these developments have a lasting impact. It's not about whether you have a strong opinion on technological trends but whether you can validate these trends at the business level over a long period of time. Simply said, let time do its work. The writer owns all the stocks mentioned. He is co-founder of The Smart Investor, a website that aims to help people to invest smartly by providing investor education, stock commentary and market coverage
Business Times
an hour ago
- Business Times
Society has to choose security over convenience when dealing with scams: MAS chief
[SINGAPORE] As a society, there is a need to choose more security over convenience especially when dealing with scams, said Chia Der Jiun, managing director of the Monetary Authority of Singapore (MAS). This comes as measures to fight scams inevitably introduce more friction and delays to users. 'While the industry will do its best to minimise the inconvenience and give time for customers to get used to the new measures, as a society we will have to make the choice for more security and less convenience,' he said at the release of MAS' annual report on Tuesday (Jul 15). Chia noted that MAS has been working with banks and payment service providers to implement stronger safeguards to disrupt scam transactions. For example, banks have phased out SMS one-time passwords for digital token user account logins and authentication of card transactions, while some major retail banks have implemented pre-transaction warnings for large value transfers made via digital banking. MAS has also seen growing take-up of measures to safeguard deposits, such as money lock. BT in your inbox Start and end each day with the latest news stories and analyses delivered straight to your inbox. Sign Up Sign Up As at May 31, 2025, at least 350,000 bank customers have signed up to use money lock and set aside close to S$28 billion of savings, Chia said. Meanwhile, MAS has been looking to further strengthen the user authentication toolkit to defend against phishing. It has been working with banks to use hardware tokens that are compliant with the Fast Identity Online Alliance (Fido) for higher risk Internet banking payments and transfers. To approve a transaction, the token will need to be plugged into the device that an individual uses. This will make it difficult for scammers to remotely perform unauthorised transactions. In the meantime, Chia noted the need to do better collectively as a society to combat all types of scams. 'Vigilance by each of us as consumers remains an important first line of defence,' Chia said.
