The conversation around AI agents has moved quickly from demos to organisational design. Microsoft’s 2025 Work Trend Index for Singapore reported that 56 per cent of Singapore leaders were already using agents to fully automate workstreams or business processes, while 46 per cent expected their teams to build multi-agent systems.

I understand the appeal. I built an internal AI office with 21 defined roles across strategy, finance, marketing, sales, customer success, delivery, engineering, design, quality assurance, security, research and data.

I expected the hard part to be technical. It was not. The harder questions were managerial: Who owns the work? Who reviews it? What can an agent decide on its own? When does a human step in?

Most of my 21 roles are real and used in live work today, but most orchestration is still interactive. That distinction matters because the biggest lesson was not how to remove humans. It was how to design responsibility around machines.

The first mistake: Capability is not ownership

My earliest agents were too broad. Help with marketing or help with development sounded reasonable because modern models can do many things.

The problem was not output quality. It was ownership.

If one agent can research, write, publish and evaluate the result, who is actually responsible for the job? If something is wrong, where does the failure sit?

I eventually stopped defining roles by what the model could do and started defining them by what the role should own.

A writing role writes. A publishing role prepares distribution. A research role scouts for information but does not install what it discovers. A builder builds, but does not declare its own work production-ready.

That sounds obvious in a human company. With AI, it is easy to forget because the same underlying model may be capable of doing all of those jobs.

Also Read: When a slot opens, let the AI agent act – within limits

The second mistake: Letting one agent close the loop

I also learned that an agent should not be allowed to create, review and approve the same important output.

When a builder tests its own work and then declares it ready, the process can look efficient while removing the independent challenge that catches weak assumptions.

So I separated three responsibilities for higher-impact work: builder, reviewer and acceptor.

The builder produces the work. A separate role tests or reviews it. Final acceptance of important client-facing, production or irreversible actions still comes back to a human.

Not every task needs three layers. Drafting an internal note does not need the same controls as deploying software or sending something to a client.

The point is proportionality. The higher the impact of being wrong, the more important independent review becomes.

The third mistake: Treating autonomy as a maturity score

At first, I saw autonomy as progress. If an agent could do more without asking me, the system felt more advanced.

I no longer think that way.

The better question is: what happens if this action is wrong?

Singapore’s updated Model AI Governance Framework for Agentic AI uses a similar risk-based approach. One case study in the framework tiers actions by severity, reversibility and the feasibility of human oversight. Low-risk and reversible actions can be automated. Moderate-risk actions require human approval. High-risk actions with limited reversibility may be blocked entirely.

That logic changed how I design workflows.

An agent drafting an internal summary can have wide freedom. An agent changing permissions, deploying code, committing money or sending an external communication should face a much higher bar.

More autonomy is not always better. Appropriate autonomy is better.

The fourth mistake: Managing the agent but not its access

An agent is not just a prompt.

It may have access to tools, databases, files, connectors, scheduled jobs and external services. As my system grew, I realised the real management problem was not simply how many agents do I have. It was what can each of them touch.

That pushed me to maintain an inventory of the agents, tools and capabilities around them.

This aligns with Singapore’s agentic AI framework, which recommends bounding agents’ powers upfront, including their autonomy and access to tools and data.

For founders, this is easy to overlook because experimentation moves quickly. A useful prototype can become part of an operating process before anyone has paused to document what permissions accumulated around it.

I now treat that inventory as basic operating hygiene, not administrative paperwork.

Also Read: From KYC to KYA: how AI agents are reshaping payment risk

The most important lesson: Humans did not disappear

The fully unattended version of my AI office is still a work in progress.

Some workflows run automatically. Many still begin with a human instruction. Important outward actions still need approval.

I once saw that as a sign the system was unfinished. Now I think the better goal is not zero humans. It is humans only where judgement, accountability or relationships genuinely matter.

The AI roles can absorb repeatable work, prepare options, check outputs, monitor systems and move tasks forward. The human role becomes narrower but more important: direction, trade-offs, relationships and final accountability.

That is also why I am cautious with the term AI employee. An employee is not valuable because they can perform many tasks. They are valuable because there is clarity around what they own, what authority they have and how their work fits with everyone else.

What I would tell another founder

Before adding another agent, I would ask four questions:

• What job does this agent actually own?
• What systems and data can it access?
• Who independently checks important work?
• Which actions can it take without a human, and why?

If those answers are unclear, adding more agents does not create an AI workforce.

It creates more capability without more management.

That was the biggest change in my thinking. I started by trying to build better agents. I ended up redesigning the organisation around them.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post I built a 21-role AI workforce. The hardest part was management appeared first on e27.

Quantinuum, one of the world’s leading quantum computing companies, has already put a number on it: a trillion-dollar market waiting to be unlocked once fault-tolerant quantum computing arrives. That’s not a typo, and it’s not a crypto-style promise built on speculation — it’s an estimate built on real, quantifiable industries: drug discovery, materials science, chemicals, energy and finance, where a single better answer can be worth billions on its own. The uncomfortable question for founders, investors and policymakers in Asia is the same one many asked too late about AI and too late (or too early, and badly) about crypto: will you be building the picks and shovels for this boom, or reading about it after the fact?

Unlike the crypto cycle, this isn’t hype chasing a use case. And unlike the early days of AI, the economic bar for quantum computing is brutally explicit: DARPA and industry alike are converging on one test — a quantum calculation only counts commercially once it demonstrably saves more money than it costs to run. A pharmaceutical company that spends billions developing a single drug, for instance, could justify an extraordinarily expensive quantum calculation if it meaningfully cuts the odds of a failed candidate. That is the bar quantum computing has to clear before it can have its own “ChatGPT moment” — and it’s a far harsher bar than the one artificial intelligence had to clear.

Quantum computing is no longer purely theoretical. The harder question is whether it can become economically useful. For Asia, the opportunity may be less about owning quantum computers and more about building the industries, infrastructure and applications around them.

Why quantum computing’s commercial breakthrough won’t look like AI’s

Artificial intelligence had been around for decades before ChatGPT made it commercially legible. Neural networks existed. GPUs existed. Large language models existed. What changed was that several prerequisites converged at roughly the same time: sufficient computing power, enormous datasets, scalable cloud infrastructure, better algorithms, usable interfaces, and a business model that let millions of people access the technology without understanding the underlying mathematics.

Quantum computing may be approaching its own equivalent transition. But there’s an important distinction. AI became useful while remaining imperfect. Quantum computing must overcome a much harsher economic constraint: its output must be valuable enough to justify an extraordinarily expensive physical machine.

DARPA has put the question unusually clearly. Its Quantum Benchmarking Initiative is trying to determine whether a quantum computer can achieve “utility-scale” operation by 2033 — meaning its computational value exceeds its cost. That may ultimately be the only quantum benchmark that matters.

Quantum today looks a little like AI before the commercial explosion

There’s a useful, if imperfect, analogy. AI progressed roughly like this:

Academic research → specialist systems → cloud APIs → foundation models → consumer applications → enterprise infrastructure

Quantum computing could follow a parallel but distinct path:

Physics experiments → error-corrected logical qubits → specialist scientific applications → cloud-accessible accelerators → industry workflows → national computational infrastructure

The difference is that AI could run on increasingly commoditised silicon. Quantum computing requires extremely specialised infrastructure: superconducting systems may need temperatures close to absolute zero; trapped-ion machines need lasers and ultra-high vacuum systems; neutral-atom machines require sophisticated optical control. The result is that the quantum revolution is unlikely to put a quantum computer on everyone’s desk.

Also Read: AI, quantum computing and the future of cyber threats

Instead, picture a future cloud architecture where a business application sits above an AI orchestration layer, which in turn routes work across CPUs, GPUs and QPUs — CPUs for business logic, GPUs for AI models, and QPUs for quantum chemistry or optimisation — before producing a single business output. Most users may eventually consume quantum computing without ever knowing a quantum processor was involved. That’s probably the real commercial end state.

The quantum roadmap: From laboratory to invisible infrastructure

Any timeline remains speculative, but current company roadmaps provide useful boundaries. IBM’s current hardware roadmap targets its Starling fault-tolerant system for 2029, with 200 logical qubits and 100 million quantum gates. Quantinuum’s published roadmap targets universal, fully fault-tolerant quantum computing by the end of the decade. These are company targets, not guarantees.

DARPA provides a useful external counterweight to vendor roadmaps: its 2033 utility-scale test is explicitly designed to separate industrial usefulness from hype. On the hardware side, Google’s Willow chip demonstrated “below-threshold” error correction in late 2024 — a long-sought milestone where adding more physical qubits makes a logical qubit more reliable rather than less, which is itself a prerequisite for any of these roadmaps to hold.

Five prerequisites have to converge before quantum computing pays off

• Fault tolerance must become economically practical. Physical qubits are noisy. Useful machines need logical qubits created using error correction. The commercial question isn’t whether error correction works in principle, but how much physical infrastructure, energy, control hardware and runtime are required to produce one reliable logical computation.
• Useful algorithms need to appear. Having a quantum processor without useful algorithms is analogous to having GPUs without deep learning. The most commercially important breakthrough may come not from hardware but from discovering algorithms that transform high-value industries.
• Classical computing has to lose somewhere. Quantum doesn’t compete against computers from 2020. It competes against whatever GPUs, supercomputers, AI models and optimisation software exist when fault-tolerant quantum systems arrive. The baseline is a moving target.
• Developers need abstraction. Most software engineers will never design quantum circuits. Commercial adoption requires APIs, orchestration layers and hybrid workflows that route only the structurally suitable part of a problem to the QPU.
• The economics must close. The final equation is simple: the economic value of the quantum result must exceed the cost of the QPU, HPC, energy, talent, integration, sampling and error correction. Quantum can be scientifically revolutionary while still being commercially irrational.

In practice, a hybrid workflow is emerging: AI decomposes a problem, classical HPC narrows the candidates, a QPU tackles the hard quantum subproblem, AI interprets the output, and a human validates the result before it’s acted on.

Also Read: The AI-quantum collision: Navigating the 2026 infrastructure inflection point

Where the money may actually be in quantum computing

The strongest early applications for quantum computing use cases share one characteristic: a single computational answer can be worth enormous amounts of money.

For the Philippines example, see DOST-PCIEERD’s ongoing STArQE quantum algorithms project for EV and renewable-energy planning. DOST-PCIEERD’s 2026 programme update also frames quantum technology as a national capability-building priority for future applications in energy, health, defence and information technology.

Who is leading the global quantum computing race?

There’s no universally accepted national ranking, because countries lead in different layers of the stack. The broad competitive map looks like this:

Singapore’s commitment is unusually concrete for Southeast Asia. The Singapore National Quantum Strategy has close to SG$300 million (US$233 million) set aside under RIE2025 for quantum research, talent and coordinated capability building.

What does a quantum-enabled Singapore actually look like?

Singapore’s geography makes it an unusually good candidate for early application. It’s small, urbanised, capital-rich, highly networked and concentrated around sophisticated sectors.

A bank doesn’t need to own a quantum computer. Its risk platform could send an unusually hard computational subproblem to a regional quantum service, sitting on top of a national HPC and quantum cloud. A pharmaceutical laboratory could use AI to generate molecules, conventional computing to eliminate obvious failures, and quantum processing for the subset where classical modelling becomes prohibitive. A port system could combine AI forecasting, classical optimisation and quantum algorithms only for hard search spaces where there’s a demonstrated advantage.

Also Read: Quantum computing’s double-edged sword could threaten cybersecurity: Report

Singapore’s advantage would be density: finance, biotech, government, universities, data centres and advanced infrastructure exist within one compact geography. Its national play may therefore be to become the application, finance, governance and integration layer for quantum computing in Southeast Asia, rather than to manufacture every layer of the stack itself.

The Philippines needs a very different quantum computing strategy

The Philippines should probably not copy Singapore. Its competitive advantage — and problem set — is almost the opposite: a geographically fragmented archipelago of more than 7,600 islands, uneven infrastructure, and large national-scale coordination problems across energy, logistics, and climate and disaster response.

Consider energy. The Philippines must coordinate island grids, renewable generation, storage, transmission limits, typhoon exposure, fuel imports, electricity demand and future EV infrastructure. AI can forecast demand, weather and failures. Classical optimisation can solve most operational decisions. A quantum service should be invoked only when a specific combinatorial or physical problem demonstrably exceeds the economic performance of classical computing.

The country is already building capability rather than waiting for mature hardware. DOST-PCIEERD has identified quantum technology as a national R&D priority and is currently funding projects including quantum-based forecasting and optimisation for the electric power grid.

The AI lesson quantum computing cannot afford to ignore

AI created enormous value. It also gave us benchmark inflation, hallucination, opaque models, exaggerated capabilities, concentration of compute and deployment ahead of governance. Quantum has an opportunity to avoid some of this by being more precise about what “advantage” actually means.

Only the fourth consistently creates a sustainable commercial market. A machine that performs a specialised calculation one million times faster but costs one billion times more has produced an impressive experiment, not an economic revolution.

Also Read: Quantum’s inflection point: Why the smart money is watching now

What to actually watch over the next decade

The most important quantum headline will probably not be “Company X reaches one million qubits.” A more consequential headline would be: “A pharmaceutical company identifies a commercially successful molecule using a quantum calculation that could not economically have been performed classically.”

At that point, investment follows. Software ecosystems emerge. Cloud providers commoditise access. Consultancies build implementation practices. Universities produce specialists. Startups stop selling “quantum” and start selling better batteries, new medicines, cheaper chemicals and lower financial risk.

The commercial endgame for quantum computing

The likely end state is a division of labour between human, AI, CPU, GPU and QPU — each solving the class of problem for which it has a structural advantage. AI determines what to investigate. Classical computing handles the overwhelming majority of computation. Quantum processors tackle narrow problems where quantum mechanics provides an economically meaningful advantage. Humans determine whether the answer should be acted upon.

For Singapore, that could create a new regional infrastructure and services industry. For the Philippines, it could offer access to computational capabilities that would be prohibitively expensive to build domestically.

The commercial quantum era begins not when the machines become powerful. It begins when someone can prove that not using one costs more.

Author’s note: Company roadmaps are forward-looking statements and are presented as targets, not independently guaranteed outcomes. The commercial timeline in this article is an analytical scenario based on current roadmaps and public evidence, not a prediction of certainty.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post Quantum’s ChatGPT moment is coming — and it’s worth trillions appeared first on e27.

Singapore has placed 25 companies on the Best Workplaces in Asia 2026 list, the highest count in Southeast Asia, at a time when employers across the region are fighting harder to hold on to skilled workers.

The list, published by Great Place To Work, ranks 200 companies across Asia based on employee survey responses. Of the Singapore-linked companies recognised this year, 20 are in the large-employer category, covering firms with at least 500 staff in Asia, while five sit in the small and medium category, for companies with 50 to 499 employees.

Also Read: Why building a people-first work culture in HR tech matters more than ever in Southeast Asia

That makes Singapore the fifth-largest contributor to the Asia list by company count, behind the UAE with 42, India (40), Greater China (39), and Japan (34). Great Place To Work’s Asia region includes the Gulf, which explains the UAE’s position at the top.

Among Southeast Asian markets, Singapore narrowly led Vietnam, which had 24 companies on the list, followed by the Philippines with 20, Indonesia with 19, Thailand with 18, and Malaysia with 15. Together, the six Southeast Asian markets accounted for 121 of the 200 recognised workplaces.

The ranking arrives in a year when workplace culture is no longer a soft metric for companies in the region. Aon’s 2025 Salary Increase and Turnover Study, which surveyed more than 700 businesses across six Southeast Asian markets between July and September last year, projects that 19.3 per cent of skilled Singapore workers will change jobs during 2026. That is the second-highest expected skilled-worker turnover rate in the region, behind only the Philippines at 20 per cent.

The pressure is made sharper by salary expectations. Singapore employers are budgeting pay rises of 4.3 per cent, the lowest among the six Southeast Asian markets covered by Aon and below the regional average of 5.3 per cent. For founders, operators, and HR leaders, that raises a familiar problem: when companies cannot compete on pay alone, they have to compete on trust, fairness, flexibility, and the quality of day-to-day management.

Real estate makes a rare appearance

One of the more striking features of Singapore’s showing is the presence of two real estate companies, a sector that barely appears in Southeast Asia’s workplace rankings.

Co-working operator JustCo ranked fifth among small and medium workplaces in Asia, while Pontiac Land Group placed 97th in the same category. They were the only real estate entrants from Southeast Asia. Indonesia, Malaysia, the Philippines, Thailand, and Vietnam had none.

Across the whole Asia list, real estate accounted for just 11 places, six of them held by UAE companies. Singapore was the only market outside the Gulf and Japan to have more than one real estate company recognised.

That matters because property has not typically been viewed as a talent magnet in the same way as technology, financial services, or consumer brands. In Singapore, however, the sector sits close to several major shifts: hybrid work, premium office demand, hospitality-linked real estate, and the changing use of commercial space after the pandemic. JustCo’s appearance also points to how flexible workspace operators are trying to position themselves not simply as landlords, but as workplace experience companies.

Biotechnology and pharmaceuticals formed another visible Singapore cluster. AbbVie ranked fourth in Asia, Merz Aesthetics ranked ninth among small and medium workplaces, and Amgen came in 50th. Thailand matched Singapore’s count in this sector, with the same three companies appearing on its national list.

Also Read: Growth meets purpose: Rethinking impact in startup culture

IT remained the largest single group in Singapore’s tally, with six companies. These included Cisco, which ranked third in Asia, Visa at 39th, and Mastercard at 65th. Manufacturing and production followed with four Singapore entrants, while hospitality and biotechnology had three each.

Overall, Singapore’s 25 companies covered 10 industries, slightly less broad than Vietnam’s 14 industries from a similar total.

The companies that climbed

Several Singapore entrants improved their regional positions from last year.

Visa posted the largest gain among Singapore companies, rising 19 places from 58th in 2025 to 39th in 2026. Micron Technology climbed 12 places, from 19th to seventh, entering Asia’s top 10. Amgen also gained 12 spots, moving from 62nd to 50th.

Other climbers included AbbVie, which rose from seventh to fourth; Cisco, which moved from fifth to third; Capella Hotels and Resorts, which went from 12th to 10th; and Marriott International, which edged up from third to second.

Hilton held first place in Asia for the second consecutive year and also appears on Singapore’s list. It was followed regionally by Marriott International, Cisco, AbbVie, and DHL. All five have Singapore operations recognised in the 2026 ranking, reflecting the Republic’s role as a regional base for many multinational groups.

At the same time, Singapore’s list was not only a story of established global names. Ten of its 25 companies were new to the Asia ranking: Allianz, Experian, Zim Integrated Shipping Services, Mastercard, Jebsen & Jessen, Heineken, JustCo, TC Acoustic, Home Nursing Foundation, and Pontiac Land Group.

Four of the five small and medium entrants were new, with Merz Aesthetics the only carry-over in that category. JustCo, TC Acoustic, Home Nursing Foundation, and Pontiac Land Group appeared on the Asia list through Singapore alone, rather than through multiple national markets. That makes their inclusion more locally specific than the listings of large multinationals that qualify through operations across several countries.

Why the ranking matters

Workplace lists can be easy to dismiss as employer branding exercises, especially in a region where companies often use awards to signal stability to customers, partners, and prospective hires. The Great Place To Work methodology, however, is built around employee responses rather than a judging panel.

Employees answer 60 statements on a five-point scale, along with two open-ended questions. The survey covers issues such as whether leaders are accessible and honest, whether pay and promotion are seen as fair, and whether employees feel their work has meaning.

A key part of the ranking is consistency. Great Place To Work measures how much responses vary within the same organisation. A company where one department is highly engaged but another is disengaged will be marked down compared with one that produces more even scores across teams and job functions.

This is particularly relevant in Southeast Asia, where companies often manage large differences between headquarters staff, frontline workers, technical teams, regional offices, and outsourced functions. A polished culture at the management level does not necessarily translate into trust across the organisation.

Also Read: The work culture paradigm in a hybrid-first world

This year’s Asia list drew on surveys of companies employing more than 8.9 million people, with more than 3.8 million individual employee responses across 36 countries and territories.

For Singapore, the result reinforces its position as a regional hub for talent-intensive industries, from enterprise technology and payments to biotech, hospitality, real estate, and manufacturing. But the wider labour market signals are less comfortable. With skilled-worker turnover projected to remain high and pay increases trailing the regional average, employers will need more than brand recognition to keep people.

In a market where employees have options, the companies that stand out may be those that can make fairness and trust feel consistent, not occasional.

The post Singapore tops Southeast Asia on Best Workplaces in Asia 2026 list appeared first on e27.

Beyond Border, the Singapore-founded, tech-enabled US immigration platform for founders and highly skilled professionals, has added former Airbnb China COO Kum Hong Siew to its advisory bench as Business Advisor.

The appointment comes as the company, which says it has stayed net profitable for two years while hitting a multimillion-dollar annual revenue run rate, pushes further into the US and prepares to expand into Latin America.

Also Read: Late-stage deals revive in Southeast Asia, but early-stage founders remain under pressure

Siew’s résumé reads like a masterclass in scaling across regulatory complexity: one of Airbnb’s earliest Asia Pacific hires, he went on to run the company’s regional business before becoming COO of Airbnb China, a market notorious for grinding down foreign operators. Before that, he cut his teeth at law firm Rajah & Tann and at Yahoo!.

e27 sat down with Siew (and on two questions specifically about Beyond Border’s market strategy, with CEO and co-founder Fred Ng) to unpack what “operational scaling” actually means for an immigration platform, why the O-1 visa’s “Einstein” reputation is mostly myth, and what a law-and-tech background taught him about vetting founders before agreeing to advise them.

You went from Airbnb to immigration tech. What drew you to Beyond Border specifically, and what parallels do you see between scaling Airbnb’s regional operations and helping founders navigate US immigration?

Kum Hong Siew: What drew me in was the founders first. Fred Ng and Arnold Ip are strong founders who are building something special. Silicon Valley is the indisputable epicentre of the AI revolution, and ambitious founders and builders around the world increasingly understand that they need to move there to have truly global impact.

I also see the mission in broader terms than a typical visa-processing pitch. When someone moves to the US and succeeds, that success flows back home too, especially true for folks hailing from emerging markets like Latin America.

The Airbnb connection, though, is about something more specific than mission alignment for me. When it involves cross-border movement of people, local nuance and understanding is critical. And it can be challenging to maintain this local mindset as you scale a business. I learnt this from my time at Airbnb, and I look forward to helping Fred and Arnold do this as well.

You went from one of Airbnb’s earliest APAC hires to COO of Airbnb China, a market known for regulatory complexity. What lessons from operating there apply to Beyond Border’s move into Latin America?

Kum Hong Siew: I’m careful not to overclaim a one-size-fits-all playbook here. Every country really is different. But there are some fundamental truths in running and scaling a business. For example, succeeding in China requires a relentless focus on the customer, and that translates everywhere else.

Beyond Border frames immigration as sitting “at the intersection of legal, operational and business strategy.” Can you unpack that with a concrete example? How does a founder’s visa status actually shape decisions like where to hire or raise capital?

Kum Hong Siew: Most of Beyond Border’s customers have already made the harder decision before they ever speak to us. They’ve already decided that they want to be in the US, usually because that’s where funding, talent and customer demand are frequently the deepest and densest in the world. And then Beyond Border helps them get the right visa to be there.

Physical presence isn’t optional if a founder actually wants to build a team or raise capital in the US. It’s very hard to do those things successfully if you are constantly flying in for two weeks before leaving, and coming back only in a few months’ time.

Also Read: Why the smartest founders are interviewing investors before investors interview them

I’ll offer a data point from my own angel investing. I have a portfolio founder from Hong Kong who now spends more than half their time in San Francisco. Since they started doing this late last year, they’ve more than 4x-ed their annualised revenue.

As Business Advisor, you’ll work on legal operations and operational scaling. What does “operational scaling” mean for an immigration platform — internal processes, attorney networks, client experience, or all three?

Kum Hong Siew: I’ll be blunt about the industry’s starting point. Immigration service providers have historically been heavy on manual processes and operations. With AI, there is now an opportunity to reinvent how such services can be delivered, to enable faster, cheaper, and higher-quality outcomes.

But I push back on treating AI as a bolt-on. It’s not about taking an existing workflow and automating select steps with AI, while essentially doing the same steps in fundamentally the same way. Instead, it’s about designing and building a completely new workflow from first principles using the technology that is now possible, while maintaining or even improving the quality of the deliverables.

My bet is on combining Beyond Border’s talented workforce and attorney networks with technology to serve far more clients without diluting the experience.

You’ve worked across law, tech, and now immigration. How has that cross-disciplinary background shaped how you evaluate a company like Beyond Border before agreeing to advise it?

Kum Hong Siew: My background allowed me to quickly understand the problem they were solving, and the different ways in which they could attack it as they scale. What sealed the decision for me was less the numbers than the founders themselves. They have already built a good business, but want to make it even bigger and better. They had a really good understanding of their business and the levers for further growth, and were also open and hungry to learn. At this stage, everything starts with and flows from the founders. So what I saw made the decision easy.

Beyond Border has grown to a multimillion-dollar revenue run rate while staying net profitable for two years. What told you this was a fundamentally sound business rather than just a fast-growing one?

Kum Hong Siew: Those metrics, together with my understanding of how such businesses operate, show that Beyond Border could grow to be a great business at scale.

Singapore-founded startups often face the question of when and how to internationalise. What do you make of Beyond Border’s decision to headquarter in Singapore while its core product serves the US market?

Fred Ng: I’d correct the framing a little there. Beyond Border was founded in Singapore, but today we are headquartered across the US and Hong Kong. Singapore remains an important part of our story, and Asia continues to be a core market for us.

I see our geographic spread as a philosophy rather than an accident. One of our core beliefs is that where you come from should not define where you can go. Singapore gave us a strong launchpad, with its concentration of talent and its position as a regional business hub, but the company has grown organically into an increasingly international business.

Also Read: Connecting founders across SEA used to be the easy part of the job, and now it’s becoming the whole job

The numbers back that up: more than 60 per cent of enquiries now come from the US, with the rest split across Asia, Europe and Latin America. For us, internationalisation has therefore been less about choosing between Singapore, Asia or the US, and more about building the company in the markets where our customers are today and where they want to go next. I think it also shows that a consumer company founded in Singapore can build from Asia and find meaningful traction with a global audience.

A press release cites that immigrants co-founded 59 per cent of America’s billion-dollar startups. From your vantage point, what’s the biggest misconception people have about who actually needs O-1 or EB-1A pathways?

Fred Ng: I’ll go straight for the myth that keeps eligible applicants from even trying. The biggest misconception is that you have to be “Einstein-smart” to qualify for an O-1 or EB-1A. These pathways have developed a reputation as the “Einstein visas,” which leads many founders and professionals to assume they are out of reach.

I’d point to approval data as the counter-argument. O-1 approval rates, for instance, have remained above 90 per cent even across both Trump administrations. What matters is whether you can demonstrate a strong track record of professional achievement and recognition within your field. As a rule of thumb, if you have achieved meaningful professional milestones that are recognised by your industry, it is worth assessing your eligibility rather than ruling yourself out because of the “extraordinary ability” label.

Beyond Border is expanding into Latin America alongside its US and Asia footprint. What operational or cultural challenges do you anticipate in building trust with founders and professionals in a market quite different from Southeast Asia?

Kum Hong Siew: I’ll return to a theme I raised earlier: there’s no universal playbook, only a universal starting point. I always advise founders to start with the customer. When you do that, you can use first principles to figure out the best way to build trust with and to serve the customer. The way to do that will vary across countries and cultures, hence local nuance and understanding is critical. That is why Beyond Border’s obsession with being customer-centric has served and will continue to serve them well.

The post Why Beyond Border thinks visas are now part of the founder playbook appeared first on e27.

The AI community keeps asking: “Where’s crypto’s killer app?” They may be asking the wrong species. Blockchain never found product-market fit with humans. It may find it with machines. AI creates intelligence. Blockchain creates verification.

For years, one of the strongest arguments from the AI community has been that blockchain never found a meaningful product-market fit. While artificial intelligence is transforming software development, healthcare, education and enterprise productivity, much of the crypto industry continues to be associated with speculative trading, meme coins and short-lived narratives. From this perspective, blockchain appears to solve few problems that traditional software cannot already address. Ironically, however, the rapid rise of AI agents may become the strongest argument yet for why blockchain infrastructure matters.

The key distinction is that humans and AI agents operate under completely different economic constraints. Most people already have access to bank accounts, credit cards, digital wallets and payment processors such as Visa, Stripe or PayPal. AI agents have none of these privileges. They cannot open bank accounts, complete Know Your Customer (KYC) verification, own payment cards or initiate cross-border bank transfers as legal entities. Yet they are increasingly expected to purchase APIs, rent GPU resources, subscribe to SaaS products, hire freelancers, negotiate cloud computing prices and interact autonomously with other software agents. Intelligence alone is insufficient; autonomous systems also require programmable identity, ownership and payments.

Also Read: EVs gain ground in the Philippines as fuel costs squeeze drivers

Blockchain provides precisely these missing primitives. A wallet serves as a native financial account for software, stablecoins enable instant global settlement without relying on banking hours, smart contracts automate payment conditions, decentralised identity establishes verifiable credentials, and tokenised assets create programmable ownership. Instead of replacing AI, blockchain increasingly functions as the financial infrastructure layer that allows autonomous agents to participate in the digital economy.

“AI makes everything fake. Crypto makes things real again,” said Balaji Srinivasan.

This transition is already visible across the industry. Coinbase recently introduced x402, an open protocol that enables AI agents to pay for APIs automatically using stablecoins by reviving HTTP status code 402 (“Payment Required”). Visa launched Intelligent Commerce, allowing AI assistants to make purchases within predefined spending limits, while Mastercard introduced Agent Pay to support secure AI-driven transactions. Stripe has expanded programmable stablecoin payment infrastructure, and startups such as Skyfire and Payman AI are building dedicated financial rails for autonomous software agents. None of these initiatives focus on retail crypto speculation; instead, they address a practical question: how can software become an economic participant?

This shift also changes the way blockchain should be evaluated. During the previous cycle, most discussions revolved around transaction throughput, token prices and decentralised finance yields. In the AI era, the more relevant question may be whether blockchain can become the operating system for machine-to-machine commerce. Autonomous vehicles paying charging stations, AI researchers purchasing proprietary datasets, robots ordering replacement components, and software agents negotiating cloud infrastructure all require trusted payments, programmable ownership and verifiable identity. These are infrastructure problems rather than consumer applications.

“Crypto startups shouldn’t pivot to AI. Crypto is foundational infrastructure for AI,” said Brian Armstrong, Coinbase.

The debate therefore may have been framed incorrectly from the beginning. Artificial intelligence and blockchain solve fundamentally different problems. AI produces intelligence by enabling machines to reason, generate content and make decisions. Blockchain produces trust by enabling ownership, payments and coordination without centralised intermediaries. Rather than competing for the same market, the two technologies increasingly appear to be complementary layers of the same digital economy. The next decade is unlikely to be defined by AI versus blockchain. It is far more likely to be defined by AI powered by blockchain, where intelligence and trust evolve together to support an economy increasingly driven not only by humans, but also by autonomous machines.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post AI doesn’t need crypto, AI agents do appeared first on e27.

A startup changes the model behind its AI feature. The new model is faster, cheaper and performs better on public benchmarks. The engineering team runs its tests, deploys the update and waits for the improvement.

Instead, support tickets begin to arrive. The assistant is less accurate on short questions. It misunderstands customers who mix languages. Nothing is completely broken, but the product is noticeably worse.

This is one of the difficult realities of building with generative AI: a system can pass every conventional software test while its behaviour deteriorates. The cheapest protection is not another monitoring platform or a more powerful model. It is a small, carefully maintained collection of real examples known as a golden dataset.

AI regressions are different

Traditional software is usually tested against predictable outputs. Give a function a particular input and it should return an exact result.

Generative AI does not work that way. Two answers can use entirely different words and still be equally correct. Conversely, an answer can sound fluent and professional while omitting a critical fact or inventing something the system does not know.

This makes informal testing dangerously attractive. Someone tries five prompts in a staging environment, reads the answers and concludes that the new version “looks better”.

That judgement becomes unreliable as the product grows. A support assistant serving several Southeast Asian markets may encounter English, Thai, Vietnamese and code-switching within the same conversation. Users submit fragments, screenshots, misspellings and requests that the product team never anticipated.

The polished questions used in demonstrations rarely resemble this traffic. A golden dataset makes quality visible. It contains representative user inputs together with a description of what a successful response must and must not do. The team runs the same examples whenever it changes a model, prompt, retrieval pipeline or tool.

It does not replace production monitoring, user feedback or A/B testing. It catches problems before those slower signals arrive.

Begin with real behaviour

The first version does not need thousands of examples. Thirty to fifty inputs from actual usage are enough to begin testing one important product behaviour.

For a customer-support assistant, that behaviour might be answering refund questions, for a financial product, it might be explaining a transaction without offering unauthorised advice, and so on.

Production examples are far more valuable than questions invented during a workshop. They contain the ambiguity, incomplete information and unusual phrasing that expose weaknesses in the system.

Also Read: SEA’s AI boom has a water problem it cannot offset away

Naturally, using production traffic requires appropriate consent, access controls, retention rules and removal of personal information. If the product has not launched, examples from internal testing or a closed beta can be used temporarily. Synthetic examples are useful for getting started, but they should gradually be replaced by real interactions.

The objective is not to construct a perfect benchmark. It is to capture a small but recognisable sample of how people actually use the product.

Test failures, not just the happy path

Many teams create evaluation sets that resemble product demonstrations: clear questions, correct terminology and complete information. Unsurprisingly, the system performs well.

A useful dataset should contain the situations most likely to cause damage. These might include ambiguous requests, unsupported languages, missing account information, contradictory documents, attempts to override instructions or questions that should be escalated to a person. Support tickets and user complaints are often the best source of such examples.

Every production failure should leave something useful behind. Once the immediate problem has been resolved, the interaction should become a new evaluation case. That ensures the same class of failure is less likely to return quietly after the next update.

Over time, the dataset becomes a record of what the team has learned about its users and its product.

Evaluate outcomes, not identical wording

For open-ended AI outputs, requiring an exact answer is usually the wrong approach.

Consider a user asking for a refund. A successful response might need to mention the refund policy, correctly identify the order, avoid promising an outcome and call the account lookup tool before answering. Many different responses could satisfy those requirements.

The evaluation should therefore describe the outcome:

• What information must appear?
• What must never appear?
• Which tool or source must the system use?
• When must the request be escalated?
• Are there limits on length, latency or cost?

Also Read: Thailand’s mobility future will be decided by data, not just vehicles

Some checks are inexpensive and objective. A test can verify whether the response contains a required fact, follows a defined structure, calls the correct tool or stays within a token limit.

Subjective qualities such as tone or clarity may require human review or another model acting as a judge. But expensive AI-based scoring should not be the default. Use the cheapest test capable of detecting the problem.

Both OpenAI’s evaluation guidance and Anthropic’s work on agent evaluations emphasise structured, task-specific evaluation rather than relying on general benchmarks alone.

Keep the dataset small enough to trust

A dataset becomes useless when it is too large for anyone to inspect. Each regression report should show the affected example, the previous response, the new response and the reason it failed. A dashboard announcing that quality fell from 84 to 81 per cent is not enough. Engineers need to see what became worse.

The dataset should also be versioned alongside the product. When expected behaviour changes, the reason should be recorded. A small portion of the examples can be held back from everyday development so the team does not unconsciously tune the system only to the cases it sees.

Finally, someone must own the dataset. Shared responsibility often means no responsibility. The owner should add newly discovered failure modes, remove obsolete examples and ensure evaluations continue to run as the product changes.

A practical starting point

A startup can establish a useful regression process within a week. Choose the single AI behaviour that matters most to customers. Collect 30 to 50 representative inputs. For each one, write down two to four conditions that define a successful outcome. Automate the inexpensive checks and run them whenever the relevant system changes.

AI teams will never eliminate uncertainty completely. Models change, products evolve and users find new ways to surprise us. But a team should always be able to answer one basic question before releasing an update: did this make the product better or worse? A golden dataset is the cheapest reliable way to find out.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post The cheapest way to stop your AI product from regressing appeared first on e27.

In Southeast Asia’s heat and humidity, keeping fruit, vegetables and ready-to-eat food fresh is not just a logistics problem but an economic one, too. Produce can lose value at every stage of the chain, from farm to wholesaler to supermarket shelf, while restaurants and households often throw away food long before it should have spoiled.

Singapore-based N&E Innovations is betting that part of the answer can come from the waste stream itself.

Also Read: Deeptech’s secret: Ignore the market, master the engineering, and let opportunity find you

The deeptech company has raised approximately US$1.7 million in Series A funding led by Australian agrifood investor Tundra Capital. SGInnovate, The Radical Fund, Archipelago VC and SG7 Group also joined the round, alongside existing backers Cercano, SEEDS Capital, Elev8 Capital and Qian Hu Corporation.

Founded in 2020 by biomedical scientist Didi Gan, N&E has developed ViKANG99, a patented edible antimicrobial ingredient made from upcycled agricultural by-products, including discarded cashew nut husks. The company says the ingredient can help fresh produce last up to four times longer by slowing the growth of bacteria and mould.

“Food waste is usually seen as something we need to get rid of. We see it as a resource,” Gan said. “We can take something like a discarded cashew nut husk, extract the compounds that naturally fight microbes and turn them into an ingredient that can help protect food.”

From cashew husks to cling wrap

ViKANG99 works by extracting naturally occurring antimicrobial compounds from agricultural waste, refining them into a food-grade ingredient, and embedding or applying that ingredient across different use cases.

One of N&E’s first commercial products is The Orange Wrap, which the company describes as the world’s first antibacterial cling wrap. Conventional cling film acts largely as a barrier. N&E’s version incorporates ViKANG99 into the material, allowing it to actively inhibit microbial growth on food surfaces.

The same core ingredient is also being used in KeepWell, N&E’s plant-based cleaning and hygiene range, and in post-harvest agriculture products that can be applied to fruit and vegetables after harvest. These include a post-harvest wash and guard system designed to suppress mould and microbial growth during storage, transport and retail.

That matters in a region where supply chains are often fragmented and temperature control can be inconsistent outside premium channels. In markets such as Vietnam, Indonesia and the Philippines, produce may move through multiple intermediaries before reaching consumers. Even in Singapore, where retail standards are high, the country’s reliance on food imports makes shelf life a national resilience issue as much as a supermarket concern.

Also Read: The environmental ethics of AI should be a product decision, not a sustainability footnote

Food waste is also a stubborn problem for Singapore. The city-state generates hundreds of thousands of tonnes of food waste each year, according to official waste statistics, while importing more than 90 per cent of its food. Any technology that can keep produce usable for longer without adding heavy infrastructure could therefore have relevance beyond a niche sustainability story.

Commercial traction beyond the lab

N&E is not entering this round as a company still searching for its first use case. Its customers already include Singapore Airlines and supermarket chain Sheng Siong. Its KeepWell hand sanitiser was also selected for inclusion in the official 2026 Singapore National Day Parade fun pack, putting the company’s technology in front of a mass local audience.

The new capital will be used to expand ViKANG99 across three areas: active antimicrobial food packaging, plant-based cleaning, and post-harvest agriculture. N&E also plans to support regulatory programmes, international market entry and team expansion.

Australia is shaping up as one of its key growth markets. Kinoya was appointed exclusive Australian distributor for the KeepWell range in June, with selected IGA supermarkets in Sydney lined up as an initial retail channel. KeepWell products are already being trialled by hospitality businesses including M Bar Thai Eatery, Show Nom Dessert and Show Neua Thai.

The Orange Wrap has also entered Australia’s foodservice market through Perth-based distributor Familiar Goods, which supplies hotels, restaurants and other foodservice operators.

For N&E, Australia offers a useful test bed. It is a sophisticated retail and foodservice market with strict food safety expectations, but it also faces long supply routes and high spoilage costs across fresh produce. Success there could strengthen N&E’s case with partners in other developed markets, including the UK and Germany, while giving it credibility in regional export markets such as Vietnam and Taiwan.

A crowded but growing shelf-life race

N&E is operating in a global market where several startups and established players are trying to extend food shelf life without relying solely on cold-chain expansion or traditional preservatives.

US-based Apeel Sciences, one of the best-known companies in the space, developed plant-derived coatings that slow water loss and oxidation in produce. Hazel Technologies, also from the US, works on packaging inserts that regulate ripening and reduce spoilage.

AgroFresh, a more established post-harvest technology player, provides freshness solutions for fruit supply chains, while startups such as Mori and Sufresca are exploring edible or biodegradable coatings.

N&E’s differentiation lies in its use of upcycled agricultural waste as an antimicrobial source and its attempt to apply the same active ingredient across packaging, cleaning and post-harvest treatment. That breadth could open several revenue channels, though it also means the company must navigate different regulatory regimes, customer buying cycles and product-performance expectations at once.

The company’s model is described as licensing-led, with ViKANG99 embedded into partner products rather than only sold through N&E’s own branded lines. If executed well, that could help it scale without building a large manufacturing footprint in every market. But licensing in food and packaging is rarely quick; partners typically require evidence on safety, performance, cost, durability and consumer acceptance.

Why investors are paying attention

Tundra Capital Managing Partner Timothy Hui said the investment fits the firm’s focus on technologies that address structural challenges in food systems.

“N&E Innovations is one of those startups that ticks every box; their technology transforms agricultural waste into a valuable ingredient, their antimicrobial active ingredient is applicable at various points across the food value chain, and they have demonstrated customers globally love their product,” Hui said.

The round also reflects a broader investor interest in climate-adjacent technologies that can show near-term commercial value. Food waste reduction has often been framed as an environmental goal, but for supermarkets, restaurants and growers, the argument is simpler: less spoilage means better margins.

Also Read: Why Southeast Asia’s next climate unicorn might be built from farm waste

For Southeast Asian startups, that commercial framing is important. Sustainability products can struggle when they are priced as a moral choice. They stand a better chance when they solve a cost, compliance or operational problem. N&E’s challenge now is to prove that its food-waste-derived antimicrobial can do so consistently across geographies and applications.

The company’s story is also unusually circular: agricultural waste becomes an ingredient that helps prevent more food from becoming waste. If N&E can translate that logic into scalable products and partnerships, it could turn a Singapore lab-born idea into a practical tool for food systems well beyond the region.

The post N&E Innovations nets US$1.7M to turn cashew husk waste into fresh produce protection appeared first on e27.

For decades, talent management has operated on a seemingly logical premise: identify your top performers, your A-players, and invest in them disproportionately. This approach, popularised by McKinsey’s War for Talent in the late 1990s, promised that organisations could secure competitive advantage by systematically differentiating their workforce. The logic was seductively simple: measure performance annually, rank employees on a curve, and focus resources on the highest-rated segment.

The universal A-player framework spread across industries because it offered HR leaders a simple, comparable metric for talent quality. If every role could be measured against the same scale, then talent could be managed like capital, allocated efficiently, tracked rigorously, and optimised continuously. This aspiration reached its logical conclusion in the talent supply chain movement, where frameworks like the R7 model (Recruit, Rate, Retain, Redeploy, Redevelop, Release, Rehire) attempted to apply manufacturing discipline to workforce management.

The problem is that the universal A-player never actually existed. Consider a global bank that rated a call centre agent and an investment banker on the same 1-5 scale. The agent who resolved 95 per cent of calls on first contact received a 3 because her manager was a tough rater. The banker who closed a mediocre deal received a 5 because his manager was lenient. The bank then used these ratings to allocate training budgets and promotion opportunities. Research has found that individual rater biases can account for up to three times as much variance in performance scores as actual employee performance. When a manager rates an employee, the result reveals more about the manager’s evaluation style than about the employee’s capabilities.

This bias problem becomes exponentially more dangerous when flawed historical ratings are used as training data for AI systems. Consider Amazon’s AI recruiting tool: trained on resumes submitted over a decade, it learned to penalise resumes containing the word women’s (e.g., Captain of women’s chess club) because most past hires were men. Algorithms trained on biased human decisions do not eliminate bias. They scale it with terrifying efficiency.

Why the universal A-player cannot survive the AI-driven supply chain

The talent supply chain metaphor demands real-time visibility into quality, throughput, and defects. Yet the universal A-player provides none of these. It suffers from three terminal pathologies.

First, it is static. Annual performance reviews look backward six to twelve months. Consider a cybersecurity firm whose annual reviews in December rated engineers on skills relevant to pre-AI threat landscapes. By January, generative AI had transformed the work entirely. Those A-players from December lacked the capabilities needed for February’s critical incident response. The organisation knew who was excellent yesterday, not who was right for today.

Second, it is retrospective. The universal A-player relies entirely on lagging indicators, completed projects, closed deals, past ratings. Consider a retail chain during the pandemic shift to e-commerce. Their A-player store managers, rated highly on in-person sales, struggled to adapt to digital fulfilment. Meanwhile, a previously B-player assistant manager who had built the store’s social media presence in her spare time turned out to be the ideal candidate. The lagging indicators missed her entirely.

Third, it is universal. Traditional A-player frameworks apply generic traits, leadership, initiative, strategic thinking, across wildly different roles. Consider a manufacturing company that rated a warehouse supervisor and a data scientist on the same strategic thinking competency. The supervisor, whose role demanded real-time operational decisions about shift scheduling, scored poorly. The data scientist, who spent weeks developing long-term forecasting models, scored highly. The company concluded the supervisor was a B-player and redirected development funds. Six months later, the supervisor had reduced warehouse defects by 40 per cent through a simple reorganisation. The universal scale had measured the wrong thing.

Also Read: Are you a human resource?

The new model: Dynamic talent matching for the AI era

The end of the universal A-player opens the door to dynamic talent matching based on real-time, role-relative fit. This new model solves both the measurement problem and the bias problem.

• Pillar one: From static to dynamic

Instead of annual ratings, the new model measures talent continuously against specific role demands. A global consumer goods company transformed its recruitment process by abandoning CVs and annual reviews entirely, replacing them with game-based assessments and video interviews analysed by AI. Candidates completed neuroscience-based games measuring problem-solving and learning agility in under 25 minutes, with results evaluated against role-specific benchmarks, not universal scales. The approach increased new hire diversity while reducing screening time by 75 per cent.

Research from Heliyon demonstrates that skill assessments can be recalibrated using neural networks that combine expert knowledge with real-time data, ensuring measurement adapts as roles evolve.

• Pillar two: From retrospective to predictive

The new model uses predictive signals instead of lagging indicators. In one documented implementation, candidates completed open-ended questions like How do you know you’ve understood what someone said? without seeing skill labels. The AI mapped free-text responses to the five to ten core tasks of the specific role. This approach prevented self-screening bias, candidates who doubted their qualifications applied anyway, and eliminated the correct answer bias of multiple-choice tests. The result: retention increased by 25 per cent and time-to-hire dropped by 23 per cent.

Research from Taylor & Francis on adaptive job recommendation systems demonstrates that such approaches achieve 92 per cent accuracy in job matching while reducing algorithmic bias by 15 per cent through adversarial debiasing mechanisms.

• Pillar three: From universal to role-relative

The new model recognises that different critical roles require different measurement criteria. A manufacturing company implemented role-specific validation by studying 226 employees, correlating assessment scores with actual job outcomes. For warehouse associates, top scorers performed at the 64th percentile on the job, while bottom scorers performed at the 33rd percentile. Top scorers were also three times less likely to be involved in safety incidents. For data analysts in the same company, the predictive criteria were completely different: pattern recognition speed and intellectual curiosity, not physical safety indicators.

Also Reda: The app worked, the product didn’t: Can we install judgement into AI agents?

Solving the bias problem through dynamic matching

The universal A-player made bias invisible. Dynamic talent matching makes bias detectable and correctable through three mechanisms.

First, role-relative measurement breaks proxy discrimination. Consider a health system where universal ratings penalised nurses who took family leave, marking them as lower commitment. Under role-relative measurement, a nurse’s fit for an intensive care role depends on clinical judgement and rapid response time, not attendance patterns from three years ago. The proxy of leave-taking no longer leaks into the prediction.

Second, continuous auditing replaces one-time validation. A technology company’s engineering team documented that the most effective bias mitigation is continuous fairness auditing using the EEOC’s four-fifths rule: a model’s scoring rate for any demographic group must be at least 80 per cent of the highest-scoring group’s rate. Leading implementations audit intersectional groups, Black women, Latinx men, where bias is often most severe. When a financial services firm ran these audits, they discovered their model systematically downgraded candidates from historically Black colleges. They retrained the model, removing the proxy signal, and hiring diversity improved without reducing performance.

Third, human-in-the-loop governance ensures accountability. Under the EU AI Act, hiring algorithms are classified as high-risk systems requiring full risk-management programmes. Employees and candidates must have the right to understand what signals influence fit scores and to contest inferences. Consider a logistics company where an algorithm flagged a driver for low reliability based on GPS data showing frequent stops. The driver contested, explaining the stops were required safety checks for hazardous materials. The system was adjusted, and the driver became one of the highest-performing team members.

Conclusion: The universal A-player is dead

The universal A-player was a useful simplification for a stable world. That world no longer exists. In an AI-driven talent supply chain where roles evolve continuously and critical talent must be redeployed in days, static, retrospective, universal ratings are worse than useless.

The new model asks a fundamentally different question: not who are our A-players? but who fits this role, right now? By shifting from static to dynamic measurement, from retrospective to predictive signals, and from universal to role-relative standards, organisations can finally escape the bias trap and manage talent with real-time discipline.

Organisations that cling to the universal A-player will find themselves measuring what no longer matters. Those that embrace dynamic talent matching will redeploy capability as quickly as they respond to changing demand. The supply chain taught us to manage inventory in real time. It is time to manage talent the same way.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post The end of the universal a-player: Dynamic talent matching in the AI-driven supply chain appeared first on e27.

AI is remarkably good at producing answers. It is even better at sounding certain. Ask a difficult question and, within seconds, a system can gather information, connect ideas and return a polished explanation. Yet a harder question arrives later: what happens when reality proves the answer wrong?

I met that problem while building OnTheRice, a Singapore-based AI research publication. Part of its work involves making time-bound directional calls, saving the evidence available at the time, and checking the result later. The experience changed how I think about AI products. Producing an answer was the easy part. Preserving an honest record of it was much harder.

Freeze the answer before reality arrives

A prediction is easy to admire while its outcome is unknown. It is also easy to repair after the fact. A changed sentence, a missing timestamp or a quietly removed failure can turn poor judgement into a convincing success story.

We began locking each call before its result was known: the direction, entry price, source set, publication time and evaluation time. As at 1 September 2026, the public Founder Ledger showed 52 correct and 34 incorrect results across 86 resolved calls, or 60.47 per cent. 10 older records remained visible but excluded because they could not be verified properly.

Those figures are not proof that the system is exceptional. They are useful because they are incomplete, imperfect and inspectable. Once the misses remained on the page, wrong stopped being one category. Sometimes the reasoning failed. Sometimes the information arrived too late. Sometimes the market had already moved. Sometimes an event changed the conditions after publication.

The practical lesson is simple: save the original output, the evidence, the timestamp and the scoring rule before the outcome arrives. Otherwise, learning can become hindsight wearing a lab coat.

More links do not mean more evidence

A second problem appeared when the system began reading large numbers of reports. 10 websites may cover the same event, yet nine might trace back to one wire story. That is not ten witnesses. It is one witness with excellent distribution.

AI systems can confuse information volume with independent confirmation. Counting URLs rewards duplication. It can make a thin claim look strong simply because it travelled far.

Also Read: SEA’s AI boom has a water problem it cannot offset away

We started treating source origin as part of the evidence. Reports were grouped when they repeated the same underlying account, while genuinely independent reporting carried more weight. The practical rule is to trace claims backwards, not merely count how many pages repeat them. Three independent reports can tell us more than 100 copies.

Time belongs inside the evidence

Suppose a report at 8am says oil is falling and another at 5pm says it is rising. Which is wrong? Possibly neither. The world happened between them.

An AI system that ignores time can flatten both statements into one moment. Worse, it can use information published later to explain a decision made earlier. This creates the illusion that the system knew more than it could have known.

Useful records therefore need more than a source link. They need the time the source was published, the time the system found it, the time the output was made and the time it was assessed. This applies beyond markets. A business cannot honestly link a competitor’s price change to falling sales without knowing which came first.

Documentation is not decoration

This approach is not unique to one product. The researchers behind Model Cards for Model Reporting proposed standardised records of a model’s intended uses, evaluation and limitations. The NIST AI Risk Management Framework also treats accountability, transparency, validity and reliability as central parts of trustworthy AI.

The need is growing. Stanford’s 2025 AI Index reported 233 AI-related incidents in 2024, 56.4 per cent more than in 2023. The figure does not mean every incident could have been prevented by better records. It does show why explanations offered only after something goes wrong are not enough.

For an applied AI product, a small receipt can carry the source, timestamp, system version, original output, confidence level, known limits, evaluation rule and eventual outcome. None of this looks as exciting as a smarter model demonstration. It is far more useful during a dispute, audit or failure review.

Also Read: The app worked, the product didn’t: Can we install judgement into AI agents?

Let the system say I don’t know

AI products are designed to answer. Silence looks broken, especially in a demonstration. But forcing a decision when sources conflict or evidence is missing creates artificial certainty.

One of our hardest lessons was to separate wrong from unverifiable. The first means reality contradicted a recorded call. The second means the evidence is too weak to score it honestly. Combining them hides different problems; counting either as a win is worse.

A mature system needs permission to abstain. I don’t know yet should be a valid output when confidence falls below a clear threshold. Teams should record why the system abstained, then test whether the rule was too cautious or appropriately restrained.

Trust needs evidence

The AI industry is understandably focused on better reasoning, larger context windows and stronger models. Yet applied systems face a less glamorous test: can another person inspect what happened?

Can they see the original source? Can they see what the system said, when it said it and what changed afterwards? Can they find the failures as easily as the successes?

Building an AI that keeps receipts taught me that mistakes are not embarrassing leftovers. Properly preserved, they are training data for the product team, evidence for the user and a guard against self-deception.

A system should not earn trust by describing itself as intelligent. Show the work. Keep the misses. Let the record speak.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post I built an AI that keeps receipts. The mistakes became the useful part appeared first on e27.

Every founder has opinions.

After all, they spend their days solving customer problems, navigating market uncertainty, raising capital, hiring talent and making decisions that shape the future of their business. Those experiences naturally produce perspectives on everything from emerging technologies and regulation to leadership, innovation and industry trends.

Yet expertise alone does not earn media coverage.

Every week, journalists receive pitches from founders who want to share their views on the latest developments in their industry. Most are ignored, not because the founders lack credibility, but because their perspectives have not been translated into stories that are timely, relevant or valuable to readers.

That is the difference between having an opinion and having a media-ready narrative.

The role of a startup PR agency or tech PR agency extends well beyond writing opinion pieces or arranging interviews. Effective thought leadership begins with understanding how journalists think, what audiences care about and how a founder’s expertise can contribute meaningfully to conversations already taking place.

When those elements come together, founder insights become more than commentary. They become stories that build credibility and strengthen a company’s reputation.

Start with the media landscape, not the founder

Many companies begin by asking what their founder wants to say.

A better question is where that conversation belongs.

Every publication serves a different audience and has its own editorial priorities. Business publications may look for commentary on market trends, while technology media often seeks practical insights into innovation, product development or investment activity. The same founder perspective can be highly relevant to one publication and completely unsuitable for another.

This becomes even more important for companies operating across Southeast Asia.

Although the region is often discussed as a single market, each country has its own media ecosystem, business priorities and cultural context. A narrative that resonates with journalists in Singapore may require a different angle in Indonesia, Malaysia or the Philippines.

For example, a founder discussing artificial intelligence could focus on regulatory frameworks for Singaporean business media, manufacturing transformation in Vietnam or digital inclusion in Indonesia. The expertise remains consistent, but the story evolves to reflect local priorities.

The best thought leadership is never one-size-fits-all. It respects regional nuances while remaining authentic to the founder’s perspective.

This is why an experienced corporate communications agency develops narratives with both editorial relevance and local market context in mind.

Understand what journalists are trying to write

Successful media relations is not simply about understanding publications. It is about understanding the journalists behind them.

Every reporter develops areas of expertise and recurring themes throughout their career. Some focus on breaking news, while others specialise in long-form analysis, policy developments or emerging technologies. Many spend years covering the same industries and are constantly looking for credible voices who can help explain what is happening beneath the headlines.

Also Read: The role of thought leadership in scaling beyond your first market

The strongest founder commentary supports those objectives.

Rather than asking, “What do we want to say?”, communications teams should ask, “What questions is this journalist already trying to answer?”

Can your founder explain why investment in a particular sector is accelerating? Have they observed changing customer behaviour before industry reports identified the trend? Can they provide practical context behind a new government policy or technological development?

When founder expertise helps journalists tell a better story, media opportunities become significantly easier to secure.

This is one reason companies work with a PR agency in Singapore. The media pool tends to be smaller in comparison, and strong media relations help you understand editorial priorities and shape founder insights to create genuine value for journalists and their audiences.

Genuine expertise is more valuable than manufactured opinions

There is a common misconception that thought leadership requires controversial opinions or bold predictions.

In reality, credibility consistently outperforms sensationalism.

Founders possess something that analysts and commentators often cannot replicate: first-hand operational experience. They understand customer challenges, market dynamics and competitive pressures because they navigate them every day.

Those experiences produce insights that are both practical and distinctive.

Instead of encouraging founders to manufacture provocative opinions, communications teams should identify recurring patterns, lessons and observations that have emerged through building the business.

Sometimes the most compelling narrative is not about predicting the future. It is about helping people understand the present.

Journalists value perspectives grounded in experience because they offer readers something genuinely useful rather than simply adding another opinion to an already crowded conversation.

Every narrative should strengthen your company’s positioning

Not every interesting opinion deserves to become thought leadership.

The most effective founder narratives reinforce the expertise that a company wants to become known for.

A cybersecurity company should consistently contribute to discussions around digital resilience and cyber risk. A fintech founder should become recognised for perspectives on financial innovation, regulation and customer trust. A climate technology business should build authority around sustainability and industrial transformation.

Also Read: Choosing the right tool: What works for news, thought leadership, influence

Over time, these repeated associations shape how journalists, customers, investors and industry peers perceive the organisation.

Visibility alone is rarely the objective.

Authority is.

Every interview, contributed article and expert quote should strengthen the same strategic positioning. This is where an experienced tech PR agency provides lasting value, ensuring media opportunities contribute to long-term reputation rather than short-term exposure.

Timing is often the deciding factor

Even exceptional insights can fail if they arrive at the wrong moment.

Media operates on relevance. Journalists are constantly looking for expert perspectives that help explain developments already shaping the news cycle, whether that involves funding announcements, new regulations, economic uncertainty, emerging technologies or changing customer behaviour.

The most successful communications strategies therefore connect founder expertise with conversations that are already happening.

Instead of waiting for inspiration, communications teams should actively monitor industry developments and identify opportunities where their founder can contribute a credible, differentiated perspective.

When expertise meets timing, media-ready narratives become significantly more compelling.

Turning expertise into influence

Founders rarely struggle with having opinions. More often, they struggle with communicating those opinions in ways that resonate beyond their own organisation.

Media-ready narratives are built at the intersection of expertise, editorial relevance, strategic positioning and timing. They are grounded in genuine experience, shaped around the needs of journalists and aligned with the company’s long-term business objectives.

That is why effective thought leadership is never simply about publishing more content. It is about consistently contributing meaningful perspectives that help audiences better understand their industry.

Whether you’re an emerging startup preparing for your first round of media engagement or an established technology company expanding across Southeast Asia, the goal remains the same.

Transform founder expertise into stories that journalists want to tell, audiences want to read and stakeholders remember long after the headlines have disappeared.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post How to turn your founder’s opinions into media-ready narratives appeared first on e27.