One of our founder friends tried it. He really did.

About six months into the AI wave, he sat down and did the math. Hiring a PR agency was too costly, factoring his startup stage and business operational expenses, so he figured now with Claude, ChatGPT, Perplexity, NotebookLM – tools that could theoretically do what took an entire PR team—he could basically cut costs and do his own PR.

One solid prompt gets you strategy. Another gets media monitoring. A third spits out outreach lists and pitch emails. A fourth generates press releases and article drafts. You scale from a full PR team to one person managing the AI outputs. You cut costs, reduce overhead, move faster.

So he experimented. Six months later, he dropped me a WhatsApp saying it just wasn’t working. That was when we sat down and tried to figure out what went wrong. And that’s when we realised the real gaps in AI.

What changed (The pros)

Speed, for starters. Press releases, pitch emails, and article drafts were generated in minutes. His startup’s social media calendar was practically built out for a full year and scheduled to go. Research that took a researcher two days? Done in under an hour. That part actually worked.

Cost-wise, yes, he saved money on the volume side. That’s not nothing for a bootstrapped startup. And on paper, it looked like a win, especially for other founders wondering if they could ditch the PR agency retainers.

Also Read: The classroom: An untapped testbed for human-centric AI

What didn’t change (and this is where it got real)

But when he reached out saying results weren’t materialising, I pulled up my sleeves to help him figure out why. And that’s when we realised something: AI can hand you the tools, but it can’t hand you the judgment. There’s a fundamental difference.

A good PR professional isn’t just a content machine; they’re a translator of nuance. They understand that some information transfers through coffee chats and off-the-record conversations. They know that certain journalistic relationships are built on years of trust and calibration, not just a well-crafted pitch email. They recognise that what works in a pitch to a tech journalist in Singapore might fall flat with a business editor in Indonesia because of cultural context, language, and heck, which platform or channel you’re using to reach out to that particular editor.

AI doesn’t have access to that. And that became very obvious very quickly.

What made us re-evaluate things

First: The media list that looked perfect but wasn’t

His AI-generated outreach list looked pristine. Categorised perfectly. Emails formatted flawlessly. But when he actually started using it, he started hearing back: “This person doesn’t work here anymore.” “Wrong email.” “Not sure who this is.”

One producer was still listed because the online media registry hadn’t updated. But she had left for a new opportunity years ago, before I had even thought of starting SARAHÁ Advisory. Where was she now? Nobody would know unless they followed her LinkedIn or caught wind of it through industry chatter. That’s human knowledge. That’s the stuff you pick up over coffee with other journalists, or by noticing someone’s career move announced on social media.

Also Read: Pandai’s low-cost growth playbook puts the edutech startup on LSE’s 100x Impact radar

The AI agent pulled her old email and title. Our PR human, who stays plugged into the industry? She wouldn’t have even sent that email. That’s not a marginal difference, that’s the difference between a pitch that lands and one that gets lost in space.

Second: The article that was smart but generic

He had a real story to tell. A hard-won lesson in e-commerce store optimisation in a volatile economy.  A perspective that only he could speak to as a multi-entrepreneur. He gave the AI a brief: create a thought leadership piece that positions him as an industry voice.

But it never made it into the opinion editorial calendars where it mattered. Editors who he pitched it to said the same thing: too high-level, generic. It was lacking value. What could readers learn from this piece that they haven’t read on probably a million other pages? It lacked the hard-won insight, the “I’ve been in the trenches and here’s what I learned” energy that actually gets coverage in opinion sections.

Our founder could tell that story. It’s real to him. But he wasn’t a PR professional who was accustomed to knowing exactly what editors were looking for, the extra meat that made the news, so it became a blind spot that he skipped – his own customer use cases. And AI wasn’t smart either to call him out for more personal insights and experiences that couldn’t be found on the internet.

A human who’s been in this space knows what not to do. They know which journalists are burned out and unlikely to respond. They know which outlets have recently changed their editorial focus. They know the unspoken rules of how to approach different editors. That institutional knowledge isn’t in any database. It lives in people who’ve been paying attention.

Also Read: Alan Turing asked if machines could think. We asked if they could lie

The lesson for early stage startups

Your next PR hire might not be a large-scale PR agency. But it’s not AI either.

It’s a human, someone who can fact-check what the AI spits out. Someone who can look at a media list and know immediately if the contacts are real and current. Someone who understands your market, your founder’s voice, and the subtle differences between a pitch that lands and one that doesn’t.

You don’t necessarily need four PR hires as an early stage startup. But you do need the advice of one really good person who can be the quality filter between what AI produces and what actually goes out into the world.

That person isn’t managing AI agents. They’re auditing them. They’re the bridge between algorithmic efficiency and human judgment. They’re the reason your outdated contacts never make it into a pitch, and why your thought leadership actually sounds like it came from a real human who’s been in the trenches.

AI is the accelerant. This person is the steering wheel.

—

Image Credit: Matt Botsford on Unsplash

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.

The post Why startup founders shouldn’t trust an AI agent to replace a PR team appeared first on e27.

The first time an AI agent changed how I think about building a team, it was 10 AM on a Tuesday.

We were trying to fix a performance drop for a brand that was spending heavily on Meta, Google, and TikTok. My team had three dashboards open, ten Excel tabs, and the same old arguments: is this creative fatigue, bad targeting, or just the algorithm having a bad week? Then our internal agent flagged a small cluster of keywords and creatives that were quietly burning around 18 per cent of the monthly budget across channels, with almost no downstream conversions. It suggested three specific actions; we shipped them, and we started to see signs of recovery over the next few days.

That night, what hit me was not “this can replace a junior analyst”. It was “this gives my team another brain that can stare at data all night, at a level of speed and precision that no human wants to maintain”.

This is what the rise of AI agents looks like in real marketing teams. Less sci-fi, more about adding a layer of always-on intelligence on top of the people you already trust.

Industry data backs this shift. McKinsey estimates that companies using AI in marketing and sales see an average increase in marketing ROI of 15 to 20 per cent. Nucleus Research found that modern marketing automation delivers about US$5.44 in value for every US$1 spent, with payback in under six months. These gains are not coming from “smarter dashboards” alone, but from systems that watch, decide, and suggest what to do next.

Also Read: Why Southeast Asia’s edutech must go beyond chatbots to truly transform learning

Where AI agents already help marketers today

In growth marketing, I see three areas where agents already act as amplifiers of human intelligence and speed, instead of just being another tool.

Always-on diagnosis, not more dashboards

Most teams still live in dashboards. They refresh Google Ads, Meta, GA4, and maybe a BI tool, then manually connect the dots. Agents flip this. They sit on top of these tools, watch data 24/7, and push a ranked list of problems and opportunities. A good example is an agent that tracks thousands of search terms, spots the small set that quietly burns budget, and tells you what to pause, what to move, and where to reallocate spend.

Turning gut feel into measurable patterns

Strong media buyers feel when a creative is dying or an audience is tired. What they lack is time to prove it across thousands of rows.
A creative analysis agent can watch every ad, group them into “winners”, “money pits”, and “hidden gems”, and link that to patterns like hook, format, or duration. The human still decides the next concept, but they are no longer guessing which variant to scale or kill.

Cleaning up reporting and hygiene work

Every brand team has someone whose week disappears into screenshots, slides, and “quick reports”. Reporting agents can pull multi-channel data, assemble reports, and schedule them, while other agents keep accounts clean of duplicate keywords, dead ad sets, and obvious waste. This does not remove people; it lets them spend more time on decisions instead of copy-pasting.

Under the hood, none of this is magic. These agents rely on solid integrations, clear rules, and feedback from the people using them. But when those pieces are in place, they genuinely extend what a small team can handle.

Also Read: Designing for the employable workforce with AI

How “one person with agents” changes the math

There is a lot of noise about “one-person companies” powered by AI. Parts of that are hype. Parts of it are already very real.

If you are a solo founder or a thin in-house team running paid-heavy growth, you can now offload work that previously required a few full-time roles:

• Account hygiene and first-pass optimisation across multiple ad platforms
• Weekly reporting and narrative for leadership or clients
• First drafts of creative analysis and testing roadmaps

With the right setup, a small team can manage somewhere around US$50,000 to US$150,000 a month in ad spend with a much lighter operational footprint than a few years ago, as long as they are disciplined about process, guardrails, and oversight. The cost curve at the lower and mid tiers of spend is already shifting.

But there are clear limits:

• Strategy still needs humans: Agents do not own positioning, product, channel fit, or the question “Should we be on this channel at all?” They optimise the game you hand them. If the brief is wrong, they just get you to the wrong outcome faster.
• Cross-functional trade-offs are still social: In fintech, for example, we have seen campaigns that look great at the “approval” stage but fall apart at the “disbursal” stage, once you connect ad data with CRM and loan data. An agent can surface that pattern, but it cannot negotiate between growth, risk, and unit economics.

So yes, one person plus agents can now run a more complex growth machine than a full team could have a few years ago. The leverage is real, but it is leverage on execution, not a substitute for judgment.

Also Read: Pandai’s low-cost growth playbook puts the edutech startup on LSE’s 100x Impact radar

Where agents still fall short

The most painful failures I see come from assuming that if an agent can act, it should act everywhere.

Three common gaps show up in real deployments:

• Important signals live outside the tools

Agents work best when key signals live in clean, digital systems. They struggle when the truth is in sales calls, WhatsApp chats, or internal politics.

No agent today can fully capture the reality of a sales team that quietly avoids leads from a certain campaign because “they feel low quality”, even if the CRM is clean.

• Vague goals lead to weird behaviour

If you tell an agent to “improve engagement,” you might get lots of cheap clicks and shallow actions. If you ask for “more approved loans that actually disburse” and wire in finance data, it starts to behave in a way that helps the business.
Clarity on the metric is still a human responsibility.

• Risk and context are still hard

Once agents get powers, changing bids, launching variants, or talking to customers, they will do something awkward sooner or later. In regulated spaces like finance or health, that is not acceptable. The safer pattern I see in these sectors is “agents recommend, humans approve” instead of fully closed loops.

In short, agents are strong where the work is repeatable, measurable, and bounded. They are weak where the work is political, ambiguous, or value-driven.

Also Read: Designing for the employable workforce with AI

Who holds the line when the agent is wrong?

The uncomfortable part about agents is that they no longer just suggest; they act.

They might pause a campaign, shift budget between geos, or prioritise one lead segment over another. When those decisions backfire, the question is simple: who owns it?

From what I have seen, responsibility lives at three levels:

• Metric owner: Every agent needs a named human owner for a specific metric: ROAS for a media agent, default rate for a lending agent, NPS for a support agent. Without that, the agent will end up chasing the easiest proxy it can move.
• Guardrail owner: Someone has to define what the agent is not allowed to do. That might include caps on budget swings, brand safety rules, or limits by region or audience. These are not engineering details; they are policy decisions.
• Explanation owner: When things go wrong, someone must be able to explain, in plain language, why the agent did what it did. That means logging decisions and surfacing reasons inside the UI, the way you would expect a junior team member to walk you through their thinking.

In my own teams, we often treat agents like named colleagues in review meetings. We walk through their decisions, question them, and then adjust the scope depending on how they did. Humans still decide when that scope grows.

The real question for founders

If you are a founder or marketing leader, the practical question is not “will agents replace my team”.

The better question is, “Which parts of my team’s week are too important to leave to tired humans scrambling through dashboards, and better handled by a machine that never blinks?”

Anywhere the work is repetitive, data-heavy, and tied to a clear metric, your next “addition” is likely an AI agent that watches the numbers all day and nudges your team when it finds something useful.

Anywhere the work is about choosing what matters, setting the bar, and holding a line on values or risk, you want humans, ideally humans who finally have the time and headspace to do that well.

—

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.

The post How AI agents are quietly rewriting the growth marketing playbook appeared first on e27.

Cube, a startup that tracks e-commerce markets for consumer brands, internet platforms and investment firms, has raised a US$3.7 million Series A round led by Betatron Venture Group, with follow-on backing from M Venture Partners and participation from strategic angels.

The funding lands at a moment when the economics of online commerce are becoming harder, not easier, to read. Brands have long relied on established data providers to understand what is happening in supermarkets, pharmacies and other offline retail channels. Online, that same visibility is far less developed. Product listings shift constantly, sellers bundle items in different ways, pricing changes by the hour, and marketplaces do not always present clean, consistent data.

Also Read: Cube Asia attracts US$1.5M to help e-commerce consumers make more data-driven decisions

That chaos is precisely where Cube has built its pitch.

Founded in 2022 and headquartered in Bangkok, the startup sells market intelligence to more than 20 enterprise customers, including global consumer goods groups, internet platforms and investment firms. Its core promise is simple: take fragmented e-commerce data, structure it, enrich it and turn it into something companies can actually use to make decisions.

For brands, that means tracking online market share, identifying where demand is growing and understanding how pricing, promotions and product visibility affect performance. For investors and platforms, it means building a clearer view of the size, shape and trajectory of digital commerce categories across emerging markets.

Why this matters now

The broader market backdrop is helping startups like Cube make their case.

In offline retail, measurement has been built over the course of decades. Categories are relatively stable, product attributes are standardised, and distribution channels are familiar. E-commerce is a different beast. Categories can fragment overnight — cross-border sellers muddy comparisons. A single item can appear in multiple formats, pack sizes or promotional bundles. In many high-growth markets, the data infrastructure around all of this remains thin.

That leaves brands trying to answer basic questions with imperfect tools: Which sub-category is actually growing? Where is the share being won or lost? Are consumers shifting to smaller packs, premium products or multipacks? What is happening to visibility on digital shelves as marketplaces tweak search and recommendation algorithms?

Cube is betting that enterprises will increasingly pay for answers. The company says its AI-enabled product tagging systems are central to its edge, allowing it to classify products at a deeper level than standard market dashboards typically allow. That includes breaking down dimensions such as pack size, primary benefit and target age group, and more recently, splitting bundled products to identify what is actually being sold inside combo offers.

That type of granularity matters because online shelves are not organised like physical ones. A shampoo is no longer just a shampoo; it might be a travel-size anti-dandruff bundle sold with a conditioner under a time-limited promotion by a third-party merchant. If a brand cannot see those layers, it can misread both consumer demand and competitive pressure.

As Cube co-founder Simon Torring put it, “enterprises would need more reliable and accurate market data and insights to win online, and this has proven even more true in the age of AI”.

Expansion beyond Southeast Asia

Cube started with a focus on Southeast Asia, where marketplace-led e-commerce growth has created large but uneven pools of digital demand. That regional base remains important, but the startup is now pushing further into North Asia and Latin America, two markets it describes as priorities for expansion.

Part of that plan includes expanding operations in Hong Kong SAR, which the company says will serve as a hub for North Asia-based clients and strengthen support for investment managers using its Tradewinds strategic market data product.

The move is notable. Many startups serving Southeast Asia struggle to scale beyond the region because local market structures differ sharply across countries. Consumer behaviour, marketplace dominance, language, logistics and regulatory environments all vary. Expanding into North Asia and Latin America suggests Cube believes the underlying data problem it solves is broad enough to travel, even if local execution will still matter.

Also Read: Hyperspace is making stores think and act like websites

That is also where the new funding is likely to be tested. Raising a Series A is one thing; proving repeatability across multiple geographies is another.

Betting on infrastructure, not just dashboards

Betatron’s backing reflects a wider investor appetite for business-to-business software that sits underneath major industry shifts rather than merely riding them.

In Cube’s case, the bet is not just on analytics dashboards but on the infrastructure needed to make e-commerce data usable. That includes data collection, product normalisation, tagging, enrichment and interpretation. If the raw material is poor, the insight layer on top quickly becomes unreliable.

Matthias Knobloch, Managing Partner and CEO of Betatron Venture Group, bluntly framed the gap, arguing that brands have solid tools for brick-and-mortar performance but still lack the same visibility in digital channels, where data is “messy, fragmented, and constantly in flux”.

That assessment is hard to argue with. Legacy market intelligence firms remain powerful in offline channels, but online commerce has produced a different set of technical challenges. Marketplaces do not expose data uniformly. Sellers manipulate listings. Product taxonomies vary from one platform to another. Promotional mechanics are more dynamic. In frontier and emerging markets, those problems are often magnified.

Cube’s opportunity lies in turning those pain points into a subscription business that feels indispensable to large customers.

The startup says its revenue has more than doubled annually since launch, and its business model is built predominantly around enterprise subscription plans. That is encouraging on paper, although the release does not disclose absolute revenue, retention or customer concentration figures, which remain key markers for any software company claiming strong enterprise traction.

The AI angle, minus the hype

Like many startups raising capital in 2026, Cube is leaning into the language of AI. But unlike businesses that sprinkle the term over generic automation, its use case is at least rooted in a concrete problem: making chaotic commercial data more precise and searchable.

That distinction is crucial.

For all the noise around generative AI, many enterprise buyers still care more about whether a system can improve data quality, reduce manual classification work, and surface useful signals faster than about flashy interfaces. In e-commerce intelligence, those gains can directly translate into pricing decisions, assortment planning, category expansion, and investment strategy.

Cube co-founder Sarabjit Singh said recent advances in AI have helped the company push reporting into deeper levels of detail, including the ability to separate bundled products and examine what sits inside them. That may sound technical, but it points to a practical reality: better parsing of online listings can lead to better business decisions.

What comes next

Cube is now entering a more demanding phase. The company has moved beyond proving that there is demand for better e-commerce market data in Southeast Asia. The next challenge is scaling that proposition across regions while defending the quality and accuracy of its insights as datasets become larger and more complex.

Also Read: E-commerce profits spark funding shift in Southeast Asia’s tech scene

That will require more than good fundraising headlines. It will require sustained product performance, strong enterprise retention and the ability to show that its intelligence is not merely interesting, but operationally valuable.

Still, the direction of travel is clear. As consumer spending continues to migrate online and digital shelves grow more crowded, companies that cannot properly read e-commerce markets risk flying blind. Cube is trying to become the system that tells them where the market is moving before their competitors see it.

For a startup born in Southeast Asia, that is an ambitious play. The new capital gives it more room to make it.

The post Cube raises US$3.7M to fix e-commerce’s visibility problem appeared first on e27.

Across Asia, a quiet revolution is underway. Banks are piloting AI agents to triage customer queries. Manufacturers are wiring factories with autonomous “co‑pilots” that watch sensor data and adjust production lines in real time. Governments are experimenting with digital assistants to guide citizens through permits and benefits.

These systems look like chatbots on the surface. Under the hood, they are something more consequential: AI agents, software entities that can perceive, plan, and act toward goals with less and less human supervision. They can call other tools, talk to other agents, and make decisions on our behalf.

For Asian policymakers, investors, and executives, the question is no longer whether these systems will arrive. They are already here. The real question is, can the region scale AI agents in ways that cut emissions, create good jobs, and strengthen social resilience—or will we simply import a new layer of risk and dependency?

This article looks at the environmental and social implications of AI agents in Asia and asks a more important question: can the region scale these systems in ways that reduce risk, strengthen resilience, and support long-term sustainability rather than simply accelerating automation?

The new machine room: Energy, water, and materials

Most conversations about AI agents focus on productivity. Far fewer acknowledge what it takes to keep them running.

An agentic system does not stop at one reply. It might:

• Call a language model dozens of times in a single task.
• Consult search, databases, or corporate systems.
• Coordinate with other agents, debating and refining answers in the background.

Each of those steps consumes computing cycles, which in turn draw electricity and require cooling.

Also Read: China blocks Meta’s AI bet on Manus: What it means next

Recent sustainability studies of AI infrastructure paint a stark picture. Training frontier models already uses massive energy and water; inference—the day‑to‑day running of models—now represents a growing share of the footprint as usage explodes. Agentic architectures amplify this trend. They stretch interactions over time and multiply calls by moving from individual queries to long workflows.

In Asia, that matters for three reasons.

• First, much of the region’s power is still fossil‑heavy. Adding large, always‑on AI agent workloads to grids in countries where coal remains dominant risks locking in additional emissions just as climate commitments are tightening.
• Second, water stress is rising. Many hyperscale data centres rely on water‑based cooling. Locating agent‑heavy workloads in already stressed basins—from northern China to parts of India and Southeast Asia—raises real questions about trade‑offs between digital ambitions and local water security.
• Third, hardware and materials are not neutral. Manufacturing the chips and networking gear that underpin AI agent platforms carries a global footprint, from mining to fabrication to e‑waste. Asia sits at several points in this chain—as producer, user, and often as the final destination for discarded electronics.

The uncomfortable truth is that energy‑blind and environment‑blind AI agents could quietly erode the very sustainability gains they are supposed to enable.

The Social Ledger: Work, Inequality, and Trust

Environmental impacts are only half the story. AI agents also reshape social and economic landscapes.

Also Read: Bridging the last mile: How AI can transform agriculture, health, and education in SEA

Work redesigned—from co‑pilot to overseer

Agentic AI changes not only which tasks can be automated, but also how work is organised. Instead of replacing an entire role, agents increasingly:

• Draft and refine content before a human sees it.
• Monitor activity streams and flag anomalies.
• Allocate work among humans based on rules the system learns over time.

In Asian service sectors—business‑process outsourcing, call centres, back‑office operations—this is already visible. Some tasks become easier or faster; others become more fragmented, more closely monitored, and less meaningful.

Research on automation and well‑being suggests that this kind of partial automation can create a peculiar mix of relief and strain. Routine burdens shrink, but so does autonomy. Workers may become supervisors of systems they neither understand nor control, bearing responsibility without agency.

The region faces additional challenges:

• High shares of informal employment mean many workers who feel AI pressure have little social protection.
• Power imbalances between multinational clients, local firms, and workers can turn “AI augmentation” into a vehicle for intensified surveillance.

Without deliberate choices, AI agents could widen gaps between high‑skill and low‑skill workers–between those who design systems and those who are managed by them.

Also Read: Creating a safe digital world: Protecting kids from cyber crimes and preventing cyberbullying

Social intelligence at scale

Most safety discussions have focused on single models hallucinating. Multi‑agent systems introduce a different class of risk. When agents interact, they can:

• Reinforce one another’s errors in long chains of reasoning.
• Converge rapidly on misleading narratives.
• Exhibit emergent “group behaviours” that their designers did not anticipate.

In markets where online information is already polarised or polluted, this matters. If news feeds, moderation systems, and political campaigns lean on swarms of semi‑autonomous agents, errors and biases can propagate faster and further. For countries balancing digital growth with fragile social contracts, that is not a theoretical concern.

Inequality of access and dependency

Finally, there is the question of who gets to own and steer these systems.

At present, the most capable agent platforms are being developed and hosted by a small set of global firms. Asian companies and governments increasingly rely on these platforms for critical functions—from software development and cloud operations to citizen services.

This raises familiar questions:

• How easy is it to switch providers if terms become unfavourable?
• Who sets the rules for data use, logging, and model updates?
• How can public regulators inspect systems that are only partially under their jurisdiction?

At the same time, there is a real opportunity for Asia’s own innovators—especially in countries such as Vietnam, Indonesia, and India—to build lightweight, local‑language agent frameworks tuned to regional needs. Whether that opportunity is seized or squandered will depend on choices taken now about standards, open tooling, and capacity‑building.

Also Read: How to navigate through the vast opportunities in the finance industry

Thinking Like the Earth: A different starting point

In our book Thinking Like the Earth: How Synthetic Intelligence Saves Our Planet, we argue that the central question is not whether AI can be “made green” through efficiency tweaks. The deeper question is whether we design AI systems—including agents—to behave as if they were part of living, interdependent systems rather than abstract optimisation engines.

That implies three shifts in mindset:

• From throughput to sufficiency. Not every task that can be automated should be. The right metric is not “maximum usage” but “enough usage to achieve social and ecological goals.”
• From isolated tools to ecosystems. AI agents sit within networks of people, institutions, infrastructures, and environments. Governance must take that whole system into account, not just the software.
• From global templates to local wisdom. Asia’s ecological, cultural, and economic diversity is an asset. AI governance that ignores this richness will fail in practice.

Also Read: How to navigate through the vast opportunities in the finance industry

Building practical governance for AI agents

The challenge for Asia is not whether AI agents will be adopted, but whether governance can keep pace with deployment.

This means building practical systems for accountability before large-scale adoption becomes irreversible.

Organisations need clearer standards around environmental reporting, human oversight, decision traceability, and vendor accountability.

Regulators need tools that move beyond abstract principles and into operational questions: where agents are being deployed, how much infrastructure they consume, and how failures are handled when systems make decisions at scale.

Without this, governance remains reactive instead of preventative.

A regional hub for AI environmental sustainability standardisation

Asia needs a voice in the way global environmental standards for AI are designed. If the region simply imports metrics, labels, and reporting formats devised elsewhere, two things may happen:

• Local environmental priorities—such as river health, air quality in dense cities, or climate resilience in deltas—will be underweighted.
• Smaller firms and public agencies may be overwhelmed by compliance demands that were never tailored to their context.

Through the Sustainable AI Portal, we are trying to work with partners to:

• Pilot energy‑aware and water‑aware metrics for AI workloads, including agentic systems, in real data‑centre and enterprise settings.
• Contribute Asia‑specific perspectives to ongoing discussions on AI sustainability in international standard‑setting bodies.
• Bring practical insights into multilateral venues, including the UN Global Dialogue on AI Governance, as it begins to grapple with environmental dimensions of AI.

The goal is not to duplicate what others are doing, but to ensure that Asia’s experiences and experiments shape global norms from the outset.

Also Read: “The risk doesn’t go away; execution decides everything”: Altara’s Dave Ng

Open tools and science‑for‑sustainability

Finally, the Portal acts as an open infrastructure for researchers, practitioners, and communities:

• Shared case studies connect AI‑agent scenarios to SDG priorities such as climate adaptation, sustainable agriculture, and resilient cities.
• Collaborative “policy labs” bring together engineers, environmental scientists, lawyers, and community representatives to design governance interventions around concrete deployments.

This combination mirrors a broader conviction from Thinking Like the Earth: synthetic intelligence will only help save the planet if it is developed as part of a wider commons—of data, knowledge, and responsibility.

What leaders in Asia can do now

The environmental and social impacts of AI agents are not an argument for paralysis. They are a call for more grounded ambition.

For policymakers:

• Treat AI agents as infrastructure, not just apps. Require basic environmental and social risk assessments before large‑scale deployments in public services.
• Support regional governance and research hubs—including in emerging centres such as Ho Chi Minh City, Jakarta, and Bengaluru—that can study impacts locally and feed into global processes such as the AI Dialogue.

For executives:

• Ask hard questions about the energy, water, and labour implications of agent deployments, not just productivity gains.
• Build internal capability to instrument and monitor systems, including “kill switches” and clear lines of accountability when agents act unexpectedly.

Also Read: The one-person company was always possible. AI agents make it probable

For academics and civil society:

• Work across disciplines—computer science, environmental science, law, social sciences—to build a realistic picture of how agentic AI is reshaping specific sectors and communities.
• Cross-border collaboration between researchers, regulators, and industry leaders will be essential to building governance models that reflect real operating conditions rather than imported assumptions.

Asia stands at a fork in the road. It can become a passive consumer of agent technologies designed elsewhere, absorbing their environmental and social costs. Or it can lead to showing how AI agents, governed wisely and designed with the Earth in mind, might actually help the region—and the planet—thrive.

The difference will not be determined by a line of code in Silicon Valley or Shenzhen. It will be shaped in ministries and boardrooms, universities and communities across Asia, by leaders willing to ask a harder question: not “How fast can we deploy agents?” but “What kind of future do we want them to build with us?”

—

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.

Image Credit: Noah Buscher on Unsplash

The post From code to carbon: How Asia can harness AI agents without harming people or the planet appeared first on e27.

Netbank is betting that the future of fintech in the Philippines will be built not on wrappers, but on real banking infrastructure. Fresh off a Series B round led by Altara Ventures, the banking-as-a-service (BaaS) player is positioning itself as a fully regulated embedded finance platform—one that owns the ledger, not just the interface.

In a market where startups often run into the limits of legacy banks, Netbank’s pitch is simple: move at startup speed without sacrificing regulatory depth.

Also Read: The future of payments in Singapore: From outages to innovation with BaaS

In this conversation, CEO Gus Poston breaks down what that actually means — from scaling payments and lending to navigating risk, partnerships, and the realities of building financial infrastructure in the Philippines.

Excerpts:

Netbank positions itself as the Philippines’s first embedded finance platform operating on a full banking licence. In practical terms, what does that licence allow you to do that a typical fintech infrastructure player cannot?

A typical fintech infrastructure player (often a middleware provider) acts as a “wrapper” around someone else’s bank account. They are beholden to the bank’s uptime, risk appetite, and legacy settlement cycles.

Netbank owns the ledger. They don’t need to ask permission to open a sub-account or move money; they perform the settlement themselves. This means that we can provide a much broader range of services, from accounts to loans to payments. There are synergies between these services, enabling us to provide a ‘tailored, simple and complete’ service for our fintech and tech company partners.

You say fintechs in the Philippines eventually “hit the same wall” and need a bank that can move at startup speed. What exactly is that wall: regulation, settlement, compliance, legacy integrations, or the unwillingness of incumbent banks to support newer business models?

I’m describing a confluence of rigid legacy systems and risk aversion in traditional banks: most incumbents see fintechs as high-risk, low-margin experiments. When a startup scales, the incumbent’s manual compliance checks, ‘standard ways of working’, or basic mistrust become a bottleneck. We adapt to the fintech’s needs so we can keep developing alongside the partner.

Also Read: The banking revolution: Balancing convenience and security in the digital era

A press release says Netbank grew revenue by 88 per cent YoY in FY2025 and was profitable. What drove that performance most sharply: payments volume, accounts growth, lending, or a few large partners scaling quickly?

Payment volume was the main driver: QR.Ph grew rapidly last year, and we were a significant part of that growth. We expect this will continue. We are also proud of the growth we saw in accounts-as-a-service and embedded lending: these will drive our future growth.

A lot of BaaS players talk about becoming the infrastructure layer for digital finance, but margins can get thin if they are moving money in the background. How does Netbank make money in a way that is both scalable and defensible?

This is indeed a risk. However, Netbank aims to be more than a ‘regulated utility’; we are a genuine partner, jointly developing innovative solutions. We use the analogy of a meal: utility BaaS providers serve ‘ingredients’, and we aim to combine them into a ‘full meal’.

You are expanding into real-time disbursements, collections, cross-border rails, embedded lending, cards, and accounts. How do you avoid becoming too broad and losing focus, especially in a market where execution risk is high?

This is indeed a risk; we have a complex business. We use partnerships where possible and grow only when there is a justified business case. By building Lego bricks for businesses focused only on the back end, we avoid large marketing spend and customer support overhead.

Who exactly is Netbank building for today: fintechs, marketplaces, payroll platforms, SME software providers, lenders, or larger enterprises? And which customer segment has turned out to be more commercially attractive than you originally expected?

We build for a wide range of tech companies, including fintech and non-fintech companies and lenders. In the mid-term, we aim to enable more tech companies to offer financial services embedded in their products.

Embedded lending is attractive, but it can go wrong quickly if underwriting discipline slips. How are you thinking about credit risk as you scale lending products through partner platforms rather than direct customer relationships?

Ultimately, the fundamentals of credit apply: understand your clients and offer them a loan they want to repay. It is hard to pull meaningful data from clients, but partner selection can help identify good clients, simplify the lending process, and build loyalty. We typically aim to work closely with our partners, who often operate under a risk-sharing approach, allowing them to benefit from their ability to identify good clients.

Also Read: Why plug-and-play should be the new standard for embedded finance

The Philippines is full of promise, but also operational friction. What has been harder than expected in building regulated financial infrastructure there: winning partner trust, navigating compliance, talent, or changing how companies think about banking integration?

The main challenge is that partnerships are slow to build and scale: the industry is just starting to appreciate the benefits of partnership and the possible products. We are now getting the momentum that pulls in clients.

What is Netbank’s real competitive moat as larger banks, digital banks, and regional infrastructure players all push deeper into embedded finance?

It’s our ‘open attitude’; we are willing to work with partners, listen to their needs and collaboratively design good banking solutions. It’s hard for a bank focused on its own clients to achieve this level of integration.

The post ‘We want to be the full meal of fintech, not just the ingredients’: Netbank CEO appeared first on e27.

Markets closed with a collective sigh of caution on Tuesday as major US indices retreated and the crypto market followed suit, reflecting a broad reassessment of risk ahead of the Federal Reserve’s pivotal interest rate decision. The Nasdaq Composite fell 0.90 per cent to 24,663.80 while the S&P 500 slipped 0.49 per cent to 7,138.80 and the Dow Jones Industrial Average edged down a modest 0.05 per cent to 49,141.93.

This synchronised pullback signals more than routine volatility. It reveals a market grappling with the twin pressures of scepticism about artificial intelligence spending and geopolitical friction, all while awaiting clarity from central bank policymakers.

The trigger for Tuesday’s equity slide came from renewed doubts about the AI investment boom. A report indicating that OpenAI missed internal growth and user acquisition targets sparked a reassessment among AI-dependent firms. Oracle and CoreWeave each fell approximately five per cent while chipmakers Nvidia, Broadcom, and AMD also moved lower.

This reaction underscores a critical inflexion point. Capital allocated to AI infrastructure must now demonstrate tangible returns rather than speculative promise. From my perspective, this scrutiny is healthy. It pushes the ecosystem toward sustainable innovation rather than valuation inflation driven by fear of missing out.

The market is beginning to distinguish between companies building durable AI advantages and those riding a momentum wave. That differentiation will define the next phase of technological and financial evolution.

Also Read: Bitcoin’s US$77,000 test: What the next 48 hours mean for your portfolio

Energy markets added another layer of complexity as oil prices surged amid renewed tensions in the Middle East. Brent crude reached US$110.75 a barrel while West Texas Intermediate traded near US$99. Disruptions in the Strait of Hormuz continue to threaten global maritime trade, injecting supply-side uncertainty into an already fragile macro picture. Higher energy costs ripple through corporate margins and consumer spending, particularly affecting logistics and transportation firms.

This geopolitical dimension reminds us that financial markets do not operate in a vacuum. They reflect real-world friction, and when trade routes are disrupted, risk premiums widen across asset classes. For investors focused on decentralised systems, this reinforces the value of resilient, borderless infrastructure that can operate despite regional instability.

Corporate earnings provided mixed signals amid the macro noise. Coca-Cola gained nearly four to five per cent after beating expectations and raising its annual outlook, demonstrating the enduring power of brands with pricing power and global reach. General Motors advanced 1.3 per cent on a strong quarterly profit beat, suggesting resilience in cyclical sectors as long as execution remains sharp.

In contrast, UPS fell three to four per cent as rising fuel costs offset underlying operational improvements, while Spotify dropped over 10 per cent due to disappointing Q2 profit guidance. These divergent performances highlight that company-specific fundamentals still matter, even when macro headwinds dominate headlines. Investors are rewarding clarity and penalising uncertainty, a dynamic that favours transparent, well-capitalised enterprises, whether in traditional or digital markets.

Also Read: While the Fed offers only 7 basis points of hope, Bitcoin marches toward US$80K

All eyes now turn to the Federal Reserve, which prepares to announce its interest rate decision at 2:00 PM ET today, with markets widely expecting rates to remain unchanged at 3.75 per cent. The real focus lies on Chair Powell’s 2:30 PM ET press conference for signals about the future policy path. Economic data releases, including durable goods orders and building permits, will add context, but the tone of forward guidance will drive immediate market direction.

Having analysed central bank communications for years, I believe the Fed faces a delicate balancing act. It must acknowledge persistent inflation pressures without derailing economic momentum. For crypto and decentralised finance, the stakes are equally high. A hawkish tilt could strengthen the dollar and pressure risk assets, while a more neutral stance might provide room for alternative financial systems to attract capital seeking yield and innovation.

The crypto market mirrored traditional risk assets, declining 0.96 per cent over 24 hours to a total market capitalisation of US$2.55T over 24 hours. Bitcoin led the weakness, falling 1.02 per cent to approximately US$76,344 and accounting for over 60 per cent of the market’s total decline.

This move triggered US$46.38M in long liquidations concentrated near the US$76,000-US$77,000 range, illustrating how leverage can amplify downturns during periods of macro uncertainty. The Coinbase Premium Index turned negative for the first time in three weeks, signalling waning US institutional demand.

Simultaneously, the Bank of Japan’s hawkish tilt revived fears of a yen carry-trade unwind, pressuring global liquidity conditions. These dynamics confirm that crypto has matured into a macro-sensitive asset class, correlated with traditional risk indicators and still capable of independent innovation.

Also Read: US$8.5B Bitcoin options expire today: Why US$72,000 is the magic number

Looking ahead, the near-term trajectory hinges on two key factors.

• First, Bitcoin must hold above the US$75,000 support level to prevent a deeper test toward the US$2.46T Fibonacci support for the total market cap.
• Second, the Federal Reserve’s messaging on April 29 will set the tone for risk appetite across equities, commodities, and digital assets.

If Powell strikes a balanced tone that acknowledges data dependence without committing to premature tightening, markets could stabilise and even rebound. Any unexpectedly hawkish surprise could extend the selloff as traders de-risk portfolios. From my vantage point, this environment favours disciplined capital allocation.

It rewards projects with clear utility, strong treasury management, and genuine user adoption over those relying on speculative narratives. The convergence of AI and blockchain, a theme I explore deeply in my work, will benefit from this clarity as resources flow toward architectures that enhance decentralisation rather than centralise control.

In conclusion, the current market posture reflects a healthy recalibration rather than a fundamental breakdown. The pullback in AI-related equities, the pressure on crypto leverage, and the cautious stance ahead of the Fed decision all point to a market digesting complex inputs and seeking equilibrium.

For those of us building the next iteration of the internet, this period of consolidation offers a strategic opportunity. It allows us to focus on technical robustness, regulatory clarity, and user-centric design without the distraction of irrational exuberance. The correlation between traditional and digital markets underscores our shared exposure to macro forces, but it also highlights the unique value proposition of decentralised systems that operate with transparency and resilience.

—

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 post The US$75,000 line in the sand: What happens to markets if Bitcoin breaks below appeared first on e27.

For years, the promise of AI has been framed around productivity, faster workflows, leaner teams, better decisions. But there’s a less discussed second-order effect emerging: AI agents may not dismantle power structures. They may reinforce them.

Not at the bottom. Not at the top. But right in the middle.

The most exposed group isn’t manual labour or visionary leadership. It’s the layer in between, the management and knowledge workers whose roles revolve around processing, validating, coordinating, and repeating structured decisions.

It includes skills that once took years to build:

• Drafting technical drawings
• Running financial models and analysis
• Producing investment memos and research reports
• Structuring presentations and strategic recommendations

These were once considered hard-earned capabilities. Today, they are rapidly being commoditised.

There was a time when being an analyst meant something.

Also Read: The hidden link between FMCG and healthcare in Philippines

Working at firms like McKinsey & Company or Boston Consulting Group wasn’t just a job; it was access to elite thinking frameworks, proprietary insights, and structured problem-solving. They defined how industries thought.

Today, that advantage is eroding.

Their frameworks are public. Their thinking is widely distributed. And more importantly, AI can now:

• Replicate their structured outputs
• Synthesise cross-industry insights
• Generate tailored strategies based on specific contexts

What used to be “top-tier thinking” is now:

• Searchable
• Learnable
• Reproducible
• Customisable on demand

A founder, junior analyst, or even a solo operator can now generate outputs that resemble what top consulting firms once charged millions for, but faster, and often more tailored.

So the question is no longer: Who has access to the best thinking?

It’s: Who controls what gets accepted?

Also Read: China blocks Meta’s AI bet on Manus: What it means next

When skills become commodities

As AI agents flatten the skill curve, the market gets flooded.

More people can:

• Build financial models
• Produce architectural drafts
• Write investment theses
• Conduct market research

The barrier to doing drops dramatically.

Which sounds like progress.

But here’s the catch: when supply increases, incumbents don’t just compete. They defend.

If technical skills are no longer scarce, the defence shifts to something harder to quantify.

We’re already seeing:

• More emphasis on ethics and governance frameworks
• Stricter compliance layers
• Additional certifications and approvals

On the surface, these are safeguards.

In practice, they are filters.

Also Read: The autonomous agent paradigm: Meta’s Manus acquisition, MCP integration, and the disruption of SaaS

Because unlike technical skills, these criteria are:

• Hard to measure
• Open to interpretation
• Controlled by insiders

And that’s where guanxi comes in.

Guanxi becomes the real moat

When output quality is no longer the differentiator, access becomes the game.

Who gets approved?
Who gets trusted?
Who gets the mandate?

Not necessarily the most capable, but the most connected.

AI agents reduce the importance of what you can produce. They increase the importance of who can vouch for you.

This is how guanxi quietly strengthens:

• Not through explicit favouritism
• But through ambiguous systems that reward familiarity over merit

The irony is uncomfortable:
The more meritocratic the tools become, the less meritocratic the system may feel.

Also Read: Why investors are betting big on Asia’s social impact startups

The macro reflection: Systems that thrive on intermediation

Zoom out, and this dynamic doesn’t just exist within companies.

It shows up at the country level.

Take Singapore.

It doesn’t compete on scale manufacturing or raw output. Instead, it thrives as a:

• Financial hub
• Regulatory bridge
• Trust intermediary between East and West

In a world where AI lowers production barriers, this positioning becomes even more powerful. This explains why Singapore move fastest in:

• AI regulation
• Institutional controls
• Usage boundaries in sensitive environments like education

Not to stop AI — but to shape who benefits from it.

AI agents were supposed to level the playing field. In many ways, they already have. But when everyone can produce, the game shifts to who gets recognised. We might not get a more open system; we get a more subtle one.

—

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.

Image Credit: ika ika on Unsplash

The post When AI agents replace the middle class, Guanxi gets more important appeared first on e27.

Thailand’s consumer lending market has become one of Southeast Asia’s more closely watched fintech battlegrounds, and Siam Digital Lending has just added fresh fuel to that race.

The Bangkok-based lender said it has raised US$7.8 million in Series A funding from a group of international investors, including a German fund manager, two German family offices, and a Hong Kong-based investment house. They include existing shareholders Santo Venture Capital and Cloudberry Ventures.

Also Read: Bridging the financial gap: How digital lending is powering financial inclusion in Southeast Asia

The company said the round was oversubscribed.

SiamDL is not simply another digital lender chasing scale in a crowded market. Its pitch is that artificial intelligence (AI) can make small-ticket lending faster, cheaper and more accurate in a country where access to formal credit remains uneven, especially for consumers and micro-entrepreneurs with thin or inconsistent financial records.

In many ways, Thailand is a natural test bed for this model, with a large population, deep smartphone penetration, a mature digital payments ecosystem and a regulator, which has already opened pathways for licensed personal and nano-loan providers. At the same time, millions of consumers still sit in the grey zone — between traditional bank credit and informal borrowing. That gap has created a sizeable opportunity for tech-led lenders promising quick decisions and transparent pricing.

SiamDL claims its lending apps have recorded more than 300,000 organic downloads, while borrowers have applied for more than US$100 million in financing since launch. The company operates in Thailand under both personal loan and nano-loan licences from the Bank of Thailand.

Why Thailand’s consumer lending fintech sector matters

Consumer lending technology is important in Thailand because it sits at the intersection of two stubborn realities: strong demand for liquidity and uneven access to formal credit.

For years, banks have dominated retail lending, but their underwriting models have traditionally favoured salaried workers and customers with established credit histories. That leaves out a large pool of self-employed workers, gig earners, small merchants and younger borrowers whose incomes may be real but irregular. In a digital economy, those people still need working capital, emergency loans and short-term financing. Fintech lenders have stepped in to serve that demand.

The sector has grown on the back of several structural factors. One is mobile-first behaviour. Consumers in Thailand are highly engaged with smartphones, digital wallets and app-based financial services. Another is the rise of alternative data, which gives lenders more signals to assess risk beyond salary slips and formal banking records. Payment data, app behaviour, device information and repayment history can all help build a clearer picture of a borrower.

Regulation has also helped. Thailand’s central bank has spent years shaping frameworks for digital financial services, including nano-finance and personal lending, allowing newer entrants to compete within defined rules rather than operating in regulatory limbo. Add in a vibrant e-commerce economy, rapid digital adoption since the pandemic, and ongoing pressure on household budgets, and the result is a market where demand for faster, smaller and more flexible credit continues to grow.

Also Read: e-Conomy SEA 2025: Digital lending hits US$91B, QR networks go regional

SiamDL CEO Andy Thienkosol framed the problem bluntly: “Until now, cost has been a primary barrier to entry for Thais seeking access to credit through online platforms.”

That observation helps explain why investors are still willing to back lending fintechs even in a tighter funding climate. In Thailand, the opportunity is not just about displacing banks; it is about making smaller loans economically viable at scale.

AI is becoming the core operating system of digital lending

This is where AI enters the picture, and where SiamDL is trying to differentiate itself.

In consumer lending, AI’s real value lies in making underwriting and servicing more efficient. In markets like Thailand, where many borrowers are under-documented, machine learning models can analyse wider sets of data to estimate repayment capacity and default risk more precisely than rigid rule-based systems. That can shorten approval times, reduce manual checks, improve fraud detection and lower operating costs.

For lenders, the benefit is obvious: smaller loans become more profitable if the cost of assessing and servicing them falls. For borrowers, the best-case outcome is quicker decisions and fairer pricing. AI can also improve collections by identifying early signs of stress and prompting softer interventions before delinquency worsens.

SiamDL says its proprietary AiTHENA system analyses thousands of factors to build customer credit profiles. That fits a broader industry shift. Across Asia, lenders are increasingly using AI not just at the point of approval, but across the full credit lifecycle, from marketing and risk segmentation to customer support and recovery.

There is, of course, a caveat. AI in lending only works as advertised if models are well-governed. Regulators and consumer advocates are paying closer attention to bias, explainability and data privacy. Faster lending decisions are attractive; opaque or discriminatory ones are not. Any Thai lender scaling aggressively with AI will eventually have to prove that its models are not just efficient, but defensible.

SiamDL is not alone in Thailand’s AI lending push

The competitive backdrop matters here because SiamDL is entering a space that already has several data-led players.

Among the better-known names is MONIX, the operator of the FINNIX digital loan app, which has built its model around mobile access, alternative-data scoring and automated credit decisions. Abacus Digital, another prominent Thai fintech, has also positioned itself around AI-enabled credit assessment and digital lending products. Ascend Nano, linked to the broader TrueMoney ecosystem, has targeted underserved borrowers and micro-merchants using digital data to underwrite customers often overlooked by traditional institutions.

That does not make the field overcrowded so much as validate the thesis. Thailand’s lending opportunity is large enough to support multiple models, especially as providers target different slices of the market: salaried consumers, first-time borrowers, merchants, informal workers and regional users outside Bangkok.

What investors appear to be betting on is that the winners will be the firms that can combine licensing, distribution, disciplined risk management and low-cost technology. Fancy algorithms alone do not build a durable lender. Cheap funding, responsible collections and regulatory credibility still matter a great deal.

A vote on execution

SiamDL founder Maxwell Meyer said the company sees room to expand access to “fair rates” for Thai borrowers. That ambition is easy to pitch; executing it is harder. Digital lenders often look impressive in growth mode, only to run into credit quality problems when underwriting is tested across a full economic cycle.

That is why the Series A is notable. International investors are not just backing a narrative around AI. They are backing a licensed lender in a market where scale, risk controls and compliance have to move together.

Also Read: Why digital lending is the future for SMEs in India

For Thailand’s fintech scene, the round is another sign that consumer credit remains one of the sector’s most investable themes, particularly when paired with AI infrastructure and a clear regulatory route. For SiamDL, the harder part starts now. Raising capital is one thing. Proving that AI can expand credit access without amplifying risk is where the real test begins.

In Thailand’s lending market, that is the difference between a flashy fintech story and a durable business.

The post Exclusive: SiamDL lands US$7.8M as AI reshapes Thailand’s lending market appeared first on e27.

When it comes to testing AI in the real world, many instinctively look to boardrooms or innovation labs. But it turns out the real proving ground is schools. Classrooms sit at the crossroads of unpredictable human behaviour, whether it’s diverse needs, learning styles, or developing real emotions. This makes them one of the best places to see whether AI works in everyday life or is just impressive in theory.

In Southeast Asia (SEA), edutech adoption is rising alongside the spread of Generative AI (GenAI), signalling a shift in how teaching and learning are approached. Deloitte finds that SEA ranks second out of nine for GenAI usage, with 9 out of 10 students having tried it.

As SEA works to build strong AI ecosystems, responsible edutech is poised to become a foundation for long-term digital growth. The World Economic Forum finds that technology skills, including AI, are expected to see rapid growth in demand – but at the same time, human skills, such as creative thinking, resilience, flexibility and agility, will remain critical.

Therefore, AI must be taught with thoughtful framing from the start, so students develop the right mix of digital skills and ethical awareness to engage with the technology confidently and safely, while their views on how to use technology, behave online, and judge information are still taking shape.

Where GenAI in education can meet educators

Efforts to incorporate AI use within the classroom are encouraging, with studies showing that thoughtfully integrated, vetted platforms can improve learning outcomes and meaningfully support children’s cognitive development. This reinforces the importance of getting the  foundation right, and positioning technology as an enhancement in their day-to-day efforts, rather than replacing critical thinking. Meeting educators where they are is essential to unlocking this potential.

Edutech companies must rethink whether their innovation is designed to truly help children learn better, more responsibly, and with greater agency. For instance, platforms may generate polished student work or assess assignments without a teacher’s input. At the baseline, educators remain wary of tools that oversell the merits of efficiency and reinforce passive automation rather than active guidance. 

Also Read: The future of work is here: The role of edutech in an AI-ready workforce

Responsible workflow design is a winning differentiator over flashy features. This includes:

• Multi-layer safety: Monitorable chat logs, in-built detection for inappropriate content which can quickly flag alerts to educators for their intervention, and safeguards against bias
• Pedagogical alignment: Tools must support “productive struggle,” enabling collaboration with AI, not outsourcing cognition to it
• Zero ambiguity in data use: Strict prohibitions on training models with student inputs
• Customisation: Toggles across grade levels, subjects, and accessibility features for students with different learning needs
  Building digital citizenship into the learning experience

The role of edutech in shaping digital citizenship adds another layer of responsibility in shaping how an entire generation learns to use AI ethically. Responsible behaviour should be embedded directly into the user experience, for example, reminders to fact-check the research claims made by AI, linkbacks to how certain answers are generated and disclosures against sharing sensitive data. Features that make learning accessible to students with different needs also contribute to healthier AI habits.

Transparency around its limitations is also equally important. These include unreliable plagiarism detectors and inaccessible features that can entrench bias or exclude learners.

How schools can put a human-first approach to AI into practice

Responsible AI deployment in classrooms often starts with choosing tools that can enhance teacher-student interaction rather than distance it. Some schools, including Stamford American International School, are approaching AI as an intentional enhancement to learning. This entails tapping on it to support and scaffold learning transparently and through safe exploration, while keeping human judgment at its core.

Examples of this in practice could include:

• Scaffold-first AI use: Tools that guide students through inquiry and problem-solving instead of delivering answers
• Safety-by-design systems: Transparent chat logs, content flagging, and teacher-intervention checkpoints
• Embedded AI literacy: Short primers before tool use, plus ongoing reminders to cite AI and avoid sensitive data
• Co-creation models: Students produce original work, then use AI for enhancement, for example, to visualise portfolios or create artwork for storybooks

These principles provide a blueprint for edutech founders, emphasising that AI should support pedagogy and enhance creativity while preserving the irreplaceable role of the teacher.

Also Read: Edutech in SEA is ripe for acceleration. This is why they can help build a more inclusive society

Such practices will also help students to learn more about AI use in a responsible, controlled manner. By learning to question outputs, cite AI use, and understand tool limitations within a safe and supervised environment, students develop the foundation for healthy AI habits that will shape how they use it, well beyond the classroom.

What’s in a successful school pilot?

For startups, the real test of their readiness lies in how well they navigate school environments. Start by engaging schools in sync with their planning cycles – a partnership is more likely to be successful when edutech vendors’ outreach coincides with curriculum planning, so that it can be meaningfully integrated from the start.

Offer modular packages. Schools respond best to providers that allow flexibility, with different offerings that schools can tailor for their specific needs, such as products to fit the region’s learning styles, cultures, and accessibility needs.

Moving into the evaluation stage, prioritise whole-community feedback. Assess opinions from everyone who uses the tool, such as teachers, students, and parents. Pilots tend to push through when data practices are kept clear.

If classrooms are the proving ground for human-centric AI, then edutech companies have an opportunity and an obligation to design with intention. Schools prioritise tools that uphold learning, amplify human judgement, and help students build the digital fluency they will need long after graduation. The future will belong to the products that understand the classroom — not as a market to enter, but as a community to serve.

Building an AI-ready generation without losing what makes us human

The promise of AI in education rests on how well it can strengthen, rather than substitute, the human elements of learning. That means designing tools that can support thinking and creativity without taking the reins of social and interpersonal skills, which no technology can replicate. Measures like device-free time, group tasks, and supervised collaboration remain essential, ensuring students continue to fail safely, build empathy, communication, and teamwork even as AI becomes more embedded in the classroom.

If SEA is looking to cultivate a generation ready for an AI-enabled future, the path forward lies in pairing technological progress with an unwavering commitment to people.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. Share your opinion by submitting an article, video, podcast, or infographic.

Enjoyed this read? Don’t miss out on the next insight. Join our WhatsApp channel for real-time drops.

Image credit: Canva

The post The classroom: An untapped testbed for human-centric AI appeared first on e27.

Malaysian agritech company Qarbotech has been named the Grand Prix winner of the SusHi Tech Challenge at SusHi Tech 2026, Asia’s largest global innovation conference, held at Tokyo Big Sight from April 27 to 29.

The startup beat out 17 other semifinalists to claim the top prize of JPY10 million (US$62,000), emerging from a field of 820 applicants representing 60 countries and regions, with 383 companies coming from Japan and 437 from international markets.

Qarbotech has developed a photosynthesis-promoting agent using nanocarbon technology designed to increase crop yields without forcing plants to overexert themselves.

“We are not making the plant work extra hard,” said Chor Chee Hoe, CEO and Co-Founder of Qarbotech. “We are just increasing their light energy usage. During overcast weather or the rainy season in our region, sunlight is insufficient, and farmers face a drop in productivity of up to 40 per cent. During sunny days, productivity is at its optimum level … we are bridging that gap, bringing those unproductive periods up to the same level as productive ones.”

When asked about the particular challenges the startup faced in promoting its solutions, the CEO spoke about the ageing farming population, which is often cautious about adopting new methods. To navigate this, Qarbotech works through established distribution partners rather than approaching farmers directly.

Also Read: Farmnet’s US$11.75M bet on a different kind of capital

“We work with partners like seed manufacturers, organic bio producers, direct contract farming companies, and chemical companies—leveraging their distribution channels and their credibility in the field,” he said.

The company also positions itself as resilient against commodity price swings. “We use agricultural-based materials, so our sustainable source will be long-term and will not be subject to cost volatility,” Chor noted.

On the exit front, the company is keeping its options open: “Plan A could be an IPO; Plan B could be an acquisition by a larger conglomerate in chemical or biotech.”

The startup announced a US$1.5 million funding round in 2024.

Malaysian startups taking on the global stage

Qarbotech’s win at SusHi Tech 2026 marks a notable moment for Southeast Asian agritech on the international stage—and a signal that solutions born in the tropics, where weather volatility directly threatens food security, are resonating with a global audience. However, Qarbotech was not the only Southeast Asian company to reach the semifinals of the SusHi Tech Challenge.

Midwest Composites, also from Malaysia, competed in the materials and bio category with a solution that processes agricultural waste into high-performance bio-composites, positioning them as alternatives to plastic and fibreglass.

While Midwest Composites did not take the Grand Prix, its inclusion alongside Qarbotech underscored Malaysia’s growing presence in the global sustainable technology space.

Also Read: The classroom: An untapped testbed for human-centric AI

Now in its fourth year, SusHi Tech—short for Sustainable High City Tech—has grown into the largest innovation conference of its kind in Asia. The 2026 edition ran across two business days on April 27 and 28, followed by a public day on April 29, a national holiday in Japan, at Tokyo Big Sight’s West Halls 1 through 4 in the Ariake district.

The event brings together startups, investors, large corporations, and universities under a shared mandate: using advanced technology to build more sustainable cities. The SusHi Tech Challenge, its flagship pitch competition, drew entries from across the globe, with the 18 semifinalists selected from that pool of 820 companies. In addition to the Grand Prix, 15 corporate partner awards were presented to other competing startups.

—

This coverage was produced as part of our media partnership with SusHi Tech Tokyo 2026.

The post Malaysia’s Qarbotech takes top honour at SusHi Tech 2026 global pitch contest appeared first on e27.