For most travellers, the anxiety of booking a flight starts before the airport. A fare appears, the dates work, and then comes the harder question: can it be paid for today?

Laters.com, the Singapore-founded online travel agency previously known as Fly Fairly, is trying to build a business around that moment. The company has raised US$1.5 million in seed funding led by XBO Ventures, the investment arm of digital-asset platform XBO.com, while completing its rebrand from flyfairly.com to laters.com.

This round comes over a year after the firm acquired LFG, a social travel discovery engine known for its viral, Gen Z-focused user experience.

Also Read: What travel tech can look like for the travel industry’s revival

The startup sells flights across more than 650 airlines and lets customers pay through over 100 methods, including digital wallets, cryptocurrencies, and around 40 buy-now-pay-later and instalment options. Its pitch is not that it has cheaper fares than everyone else, but that it gives travellers more ways to complete a booking once they find one.

That distinction matters in Southeast Asia, where online travel demand has returned strongly after the pandemic, but payment behaviour remains fragmented. Credit card penetration varies widely across the region, while wallets, bank transfers, instalment products and local payment rails often dominate daily spending. For travel platforms, that creates a gap: demand may exist, but checkout can still fail if customers cannot use their preferred payment method.

Laters.com says the majority of its payment volume is already non-card, which it describes as the inverse of much of the wider travel industry.

“Nobody should lose the fare they found because payday is two weeks away,” said Alex Yardley, founder and CEO of Laters.com. “Family, work, a wedding: some trips cannot wait. Laters.com fixes the price today and spreads the cost, so the people who plan ahead are not the ones who pay the most.”

From Fly Fairly to Laters.com

Laters.com launched from Singapore in August 2024 and says it has been profitable every month since February 2025. The company claims it is now on a run rate of more than one million travellers a year searching for flights on its platform. The United States has become its largest market, despite the company being headquartered in Singapore.

That geographic mix hints at the nature of the product. Buy now pay later, or BNPL, is well established in markets such as the US, Australia and parts of Europe, but Southeast Asia has also become an important testing ground for alternative payments. Players such as Atome, Kredivo, Grab, Shopee and others have trained consumers to split purchases into instalments, even as regulators keep a closer eye on consumer debt and transparency.

Flights are a particularly sharp use case because prices can move quickly and the ticket size is often much higher than a typical e-commerce purchase. Laters.com says customers using local and flexible payment methods book 35 per cent more often and spend 24 per cent more per booking than card users, citing a Stripe case study. It also says BNPL customers generate an average order value more than three times higher than card users.

Those figures explain why travel companies are paying closer attention to checkout design. In a market where customer acquisition costs are high and margins can be thin, a failed payment is not a minor technical issue. It can be the difference between a booked trip and a lost customer.

Crypto at checkout, not as a gimmick

The other part of Laters.com’s proposition is cryptocurrency. The platform accepts stablecoins and more than 70 other cryptocurrencies, settled at checkout like any other payment method. According to the company, crypto customers spend more than twice as much as the average customer, typically on long-haul trips and higher cabin classes.

Also Read: The unsexy side of SEA traveltech: eSIMs, visas and hourly hotels win big

That does not mean crypto has become a mainstream way to buy airline tickets. In much of Southeast Asia, digital assets remain volatile, unevenly regulated and often associated more with trading than day-to-day payments. But stablecoins, which are designed to track the value of fiat currencies such as the US dollar, are increasingly being watched by fintech firms as a possible bridge between crypto users and practical spending.

For XBO Ventures, that appears to be the investment thesis.

“We back companies that turn digital assets into something people actually spend. Laters.com has done that at one of the sharpest points of friction there is: paying for a flight,” said Dor Maman, co-founder and CFO of XBO. “The majority of its volume already moves outside the card networks, and that is where we think this market is going.”

XBO Ventures has also backed payments infrastructure company Rapyd, participating in its US$500 million Series F round, according to the company’s notes to editors.

A crowded field, but a narrower wedge

Laters.com sits at the intersection of several competitive categories. In online travel, it faces large platforms such as Booking Holdings, Expedia Group, Agoda, Traveloka, Trip.com and AirAsia MOVE, many of which already offer flights and have deep supplier relationships. In flexible travel payments, companies such as Alternative Airlines have long promoted instalment options for flights, while Travala is known for allowing crypto payments across travel bookings.

In Southeast Asia, superapps and local travel platforms also have the advantage of distribution, loyalty programmes and embedded wallets.

Laters.com’s narrower wedge is payment breadth: instead of competing mainly on inventory or price, it is trying to become the travel checkout that adapts to how younger consumers already pay elsewhere.

Yardley’s background also reflects that mix of travel and commerce. Before founding Laters.com, he spent two decades across travel and e-commerce, including roles on eBay’s EMEA leadership team, as Senior Director at Booking Holdings leading partnerships for Agoda and Booking.com across Asia Pacific, and most recently as Managing Director at ShopBack.

The company is also publishing two free tools aimed at travellers navigating the fragmented pay-later market. One is a country-by-country guide to fly now pay later providers, covering options such as Klarna, Afterpay, Zip and Atome. The other is the Laters.com Payment Score, which rates pay-later providers available at checkout out of 10 based on repayment flexibility, interest rate, eligibility and approval process.

Laters.com says no provider pays to be scored, ranked or placed, and that it earns the same margin regardless of how a traveller pays. That claim will matter if the platform wants to be seen as a neutral guide rather than another checkout funnel.

The new funding will go towards brand growth, expansion of the core flight business, and new products and verticals. Yardley hinted that the company’s ambitions may stretch beyond travel.

Also Read: Travel is back, and it’s more cutthroat than ever

“Millennial and Gen Z travellers have spent a decade adapting to a booking experience built for their parents,” he said. “We built Laters.com around how this generation actually pays, and it turns out what we built works well beyond flights.”

For now, Laters.com is making a focused bet: that the next wave of online travel growth will not only come from more destinations or cheaper fares, but from giving travellers more control over how and when they pay.

The post Laters.com raises US$1.5M to expand flexible flight payments appeared first on e27.

Southeast Asia’s Native AI companies raised US$4.1 billion in the first seven months of 2026, more than double 2025’s full-year total, according to Tracxn’s Southeast Asia AI Startup Landscape report.

On a headline chart, that looks like a region riding the same AI wave as everyone else. Strip out a single transaction and the picture changes completely: Kling AI’s US$2.8 billion Series D alone accounted for roughly 68 per cent of that total. Take it out, and Southeast Asia’s AI companies raised closer to US$1.3 billion, and the number of disclosed rounds fell from 41 in 2025 to just 23 this year.

Also Read: AI is eating the world and startups are riding the infrastructure wave

Bigger cheques, fewer of them, concentrated in fewer companies. That is not a funding boom. It is a filtering mechanism, and most of the region’s AI “wrapper” startups (the ones offering a thin, prompt-engineered interface over someone else’s foundation model) are on the wrong side of the filter.

The wrapper reckoning is already global

“Last year demonstrated that it’s difficult to survive as an AI wrapper company,” George Mathew, MD at Insight Partners, told Crunchbase News in a trend piece published in January.

That sentiment has hardened into investor consensus through 2026. PitchBook analysts have documented investors nearly halting funding for horizontal, undifferentiated AI platforms, favouring companies with a proprietary data advantage, genuine compute economics, or workflow lock-in that a general-purpose chatbot cannot replicate overnight. The pitch-deck advice circulating among Silicon Valley accelerators this year is blunt: if a reviewer’s first reaction is “OpenAI wrapper,” the meeting is already over.

The mechanism is straightforward. Every capability a wrapper startup builds on top of GPT, Claude, or Gemini can, in principle, be absorbed into the next release of that same model. A startup whose entire product is a nicer interface to somebody else’s intelligence has no moat, only a head start, and head starts in this market now measure in months.

Southeast Asia’s version of the squeeze

The regional data bears this out with uncomfortable precision. Singapore alone accounts for roughly US$9.3 billion of Southeast Asia’s cumulative Native AI funding since 2019. On the other hand, Vietnam, Malaysia, Indonesia and Thailand have collectively raised less than US$40 million combined.

AI infrastructure (foundation models, compute platforms, and the picks-and-shovels layer) was the single most heavily funded segment in 2026, pulling in US$4.3 billion across 56 rounds, led by Kling AI and MiniMax’s US$1.2 billion round.

Also Read: Fintech, DeFi and applied AI define Southeast Asia’s new venture discipline

Investors are not walking away from Southeast Asian AI. They are walking straight past the application layer to write concentrated cheques into infrastructure and foundation-model plays domiciled almost entirely in one city-state.

That leaves a large, under-discussed population of genuinely useful but thinly differentiated GenAI tools — customer-service chat layers, document summarisers, marketing-copy generators built for SEA-specific languages and workflows — competing for a shrinking pool of smaller, earlier-stage cheques. Some of that work is legitimately valuable to the SMEs and enterprises using it. Very little of it, on current investor logic, is fundable as a stand-alone venture-backed company.

What actually separates a wrapper from a company

The startups clearing the bar globally share a pattern worth naming plainly, because it is achievable, not mystical. They own an exclusive dataset a general model cannot replicate, which is proprietary transaction, behavioural or domain data accumulated through actual usage. They have workflow lock-in deep enough that switching costs, not model quality, keep customers paying. And they can show a credible path to serving users profitably at scale, rather than assuming compute costs will simply keep falling in their favour.

For Southeast Asia specifically, that argues for leaning harder into precisely the terrain that is hardest for a Silicon Valley foundation model to serve well from the outside: hyper-local language data across Bahasa, Vietnamese, Thai and the region’s dozens of dialects; regulatory and compliance workflows tied to specific national frameworks; and vertical depth in sectors (logistics, agritech, healthcare compliance), where the value sits in proprietary operational data, not in the fluency of the underlying model.

The quiet exit ramp: consolidation, not collapse

It would be too simple to say wrapper startups simply fail. The more common outcome globally has been quiet absorption: talent acquihires, small tuck-in acquisitions by larger platforms wanting a distribution channel or a regional team, or founders folding a standalone product into a feature inside someone else’s suite.

Crunchbase News has tracked over 127,000 tech job cuts at US-based companies in 2025 alone, a chunk of it AI-adjacent restructuring rather than pure failure — talent being reallocated rather than simply let go.

Southeast Asia should expect the same pattern rather than a wave of dramatic shutdowns: wrapper startups that raised a seed round in 2024’s enthusiasm quietly becoming a feature at a larger fintech, super-app, or enterprise software company rather than an independent Series A story. That is a reasonable outcome for a founding team, but it is a very different one from the venture-scale exits the 2024 funding wave implicitly promised investors and early employees.

The uncomfortable question for founders and investors alike

None of this means Southeast Asia’s AI funding story is disappointing; US$4.1 billion is real capital, and infrastructure investment of this scale builds genuine regional capability over time. But founders currently raising on a “we built a nice interface to an LLM” pitch should treat 2026’s numbers as a warning rather than encouragement. The rounds are getting bigger for companies that have already proven defensibility, and smaller, or non-existent, for everyone else.

Also Read: Where AI money is made, and where SEA founders should actually compete

Investors, for their part, might ask themselves a harder question than “does this have a moat”: whether Southeast Asia’s own concentration of AI capital into Singapore-domiciled infrastructure plays is creating exactly the kind of regional imbalance the ecosystem has spent a decade trying to correct.

A funding boom that leaves Vietnam, Indonesia, Malaysia and Thailand collectively under US$40 million is not obviously healthier than the wrapper glut it is replacing; it is simply a different kind of concentration risk, one investors are currently far less inclined to name out loud.

The post The AI wrapper reckoning has reached SEA’s funding tables appeared first on e27.

The post The Podular future: Why AI demands a new organisational architecture appeared first on e27.

Ask AI how to improve a factory, a clinic, a logistics company, or a retail business, and it will have plenty to say.

It can list ideas, explain trends, draft plans, compare options, and make a rough proposal sound persuasive. In a few minutes, it can produce the kind of first draft that once took a junior team days.

That is useful.

It is also easy to mistake useful information for good advice.

AI can suggest 20 ways to improve a factory. The person who has spent 10 years on that factory floor may know which 19 will fail by Friday.

That is not because the experienced person knows more facts. AI may have more facts than either of you can read in a lifetime.

It is because experience gives people judgment.

Answers are becoming cheap

For a long time, access to information was an advantage. If you knew where to look, which expert to call, or how to write a decent first draft, you could move faster than someone who did not.

AI is lowering that advantage quickly.

Research summaries, product ideas, basic code, marketing copy, customer emails, and business plans are becoming faster and cheaper to create. The cost of a first attempt is approaching zero.

Also Read: The transformation ecology crisis: How AI is exposing the hidden fragility of high-performing teams

This is good news for people starting with less money or fewer connections. A graduate can explore an idea without hiring a consultant. A small business can analyse feedback without a large research team. A retrenched manager can turn years of experience into a clearer plan for a new service or product.

But when everyone can get an answer, the value shifts to deciding which answer matters.

Experience is a filter

Domain expertise is not a collection of facts stored in someone’s head. It is a filter built through repetition.

It tells a nurse which symptom matters first. It tells a procurement manager which supplier promise will collapse under pressure. It tells a mechanic which sound is serious. It tells a founder which customer complaint is a real market signal and which one is simply a loud opinion.

This kind of knowledge is often hard to explain because it is practical. It is built from bad decisions, difficult customers, failed projects, missed deadlines, and work that went wrong when the presentation said it should go right.

AI can give an expert more options. It can help them see patterns, organise information, and test ideas. But it cannot know the local constraints unless someone who understands them provides the context.

That makes experienced people more important, not less.

The best prompts come from people who know the work

A weak question gets a weak answer.

Ask AI, “How can I improve my logistics business?” and you will receive a polished set of general suggestions. Ask, “How can I reduce failed same-day deliveries in Jakarta during peak rain, without adding drivers or breaking our margin?” and the answer becomes more useful.

The difference is not the tool. It is the person asking.

Domain experts know what to include in the question. They know which limits cannot be ignored. They know the customer, the workflow, the budget, the regulation, and the inconvenient detail that changes everything.

Also Read: Why Malaysia’s AI Nation 2030 plan matters for B2B startups

That is why a person with real experience can use AI as leverage. They can turn a broad suggestion into a practical test. They can spot an idea that will not survive the real world. They can notice a useful improvement that someone outside the industry would never have seen.

Creation may be cheap, judgment is not

As the cost of producing a first draft falls, creating something average will become easier.

The valuable work will be deciding what is worth creating at all.

This matters for graduates. The goal is not only to learn how to use AI. It is to get close to real work, real customers, and real consequences. Those experiences create the judgment that makes AI useful.

It also matters for workers who are retraining after a career change. Years spent in an industry are not obsolete because a chatbot can explain the industry. Practical knowledge is often the best starting point for a service, product improvement, or invention.

And it matters for SMEs. Their most valuable knowledge may sit with the people who talk to customers, run the machines, manage the suppliers, and fix problems when nobody else knows what to do.

If that knowledge leaves with an employee, it may leave without being captured. If it is noticed, documented, and developed, it may become a better product, a trade secret, a patentable invention, or a stronger way of working.

AI helps expertise travel further

The hopeful story is not that AI replaces the expert. It is that AI can help the expert do more.

A food producer can explore ways to reduce waste. A technician can turn a recurring repair into a design improvement. A logistics manager can use operational data to test a better route or handover process. A clinician can identify a problem in a care journey and explore a safer solution.

The tool reduces the cost of exploration. The expert supplies the judgment.

That combination is powerful because it lets practical experience travel further. It can move from a private insight to a documented process, a tested product, a protected invention, or a business that serves more people.

The future will reward people who know how to use AI. But it will reward even more those who know what AI should be used for.

AI has answers.

Experience knows which ones matter.

If you have practical knowledge that could become a product, process, or invention, start brainstorming it at IPGuru.ai.

—

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 has answers, experience has judgment appeared first on e27.

A great deal of ethical AI discussion still happens at a distance from the people who live with the system every day. It happens in governance forums, legal reviews, executive updates, risk committees, and product documents. All of that has value, but none of it answers the most revealing question.

When the system makes a bad call, can the person closest to the customer, patient, claimant, applicant, or case actually challenge it?

Contestability is the missing centre of ethical AI

Many organisations speak about fairness, accountability, transparency, and safety. Far fewer build properly for contestability. That matters because a system can be documented, monitored, and technically explainable, while still being very difficult to challenge in practice.

Contestability means more than having an override button buried somewhere in the workflow. It means the system is designed so that human disagreement is expected, legitimate, and operationally supported. It means people are not merely allowed to question an output in theory. They are able to do so without being punished by time pressure, managerial pressure, or the quiet cultural message that the machine is usually better.

This is the point many companies still miss. They think ethics is mainly about how the model behaves. It is also about how much room the organisation gives humans to resist the model when reality no longer fits the output.

Frontline employees often see harm before leadership does

One reason this question matters so much is that frontline employees are usually the first people to see where the system is breaking. They hear the confusion in a customer’s voice. They notice when a recommendation does not fit the case history. They see when a decision is technically consistent but practically absurd. They feel the human consequences before those consequences become a trend line in a monthly review.

Yet in many organisations, the frontline sits low in the hierarchy of trust. Their judgement is treated as anecdotal. Their objections are seen as local friction. Their escalations are sometimes tolerated, but not welcomed. The machine may be backed by data science, product, engineering, and leadership optimism, while the employee challenging it is backed only by experience and instinct.

Ethical AI becomes fragile when the people nearest to lived reality are expected to absorb the consequences of bad outputs without having the standing to challenge them. In those environments, the business still tells itself that humans remain in control. In practice, the human role has narrowed into delivery and damage management.

Also Read: AI agents could help Southeast Asian firms untangle cross-border payment costs

Override is not meaningful if it carries career risk

A lot of companies can point to formal override mechanisms. They will say the employee can escalate, pause the process, or route the case for review. On paper, that sounds reassuring. In reality, the value of override depends entirely on the surrounding culture.

Can a frontline employee challenge the model without being seen as inefficient?

Can they do it without creating a delay they will later be blamed for?

Can they do it without needing to prove the system wrong to a higher evidential standard than the system needed to make the recommendation in the first place?

Can they do it repeatedly if a pattern emerges, or only occasionally before they are labelled difficult?

Most organisations want the comfort of human judgement without the cost of supporting it

This is where the issue becomes uncomfortable in a useful way. Many firms want to claim that a human remains in the process, but they do not want to pay the operational price of making that human genuinely powerful.

Real challenge rights are expensive. They slow some decisions down. They require training. They require better case design, better escalation flows, and managers willing to back people who raise concerns. They require enough slack in the system for employees to think rather than merely process. They require leaders to accept that some machine recommendations will be questioned not because the model is broken, but because human reality is messy.

That is a much harder model than symbolic oversight.

So companies often settle for a compromise they do not describe clearly. The employee remains present, but not empowered. The organisation gets the reassurance of human involvement and the productivity profile of machine-led processing. The frontline becomes a moral buffer rather than a real decision maker.

Ethical AI depends on organisational, not just technical restraint

There is a tendency to frame ethical AI as a problem of model limits. Better testing, clearer thresholds, stronger policies, safer deployment. These all matter, but they do not solve the core institutional question.

Does the organisation have the courage to let people close to the work challenge the system, even when doing so slows things down, complicates reporting, or disrupts the story that the product is performing well?

Also Read: When AI starts thinking for us

That is the harder test because it touches power. It asks whether a call centre agent, claims reviewer, nurse, support specialist, operations analyst, or case worker can force the organisation to confront a failure before leadership is ready to admit it. It asks whether managers will protect that challenge or quietly discourage it. It asks whether product and engineering teams are willing to hear that a system which looks strong on aggregate is creating real harm at the edges where people live.

The best ethical systems treat disagreement as intelligence

One of the clearest signs of maturity is how a company interprets human disagreement with AI. Weak organisations treat disagreement as resistance. Strong ones treat it as intelligence.

When frontline employees contest outputs, they are often surfacing something the system cannot see properly. Missing context. Policy ambiguity. A rare case type. A hidden operational cost. A human signal that does not fit the data structure neatly. If the organisation is wise, it treats those moments as valuable evidence about the limits of the model and the design of the workflow.

That requires a change in posture. Instead of asking, “Why did the employee not trust the system?” leaders should also ask, “What did the employee see that our system or process could not absorb?”

That is a much more serious question. It turns challenge into a source of institutional learning rather than a local inconvenience.

The frontline is where ethical claims become real

Every company can produce a responsible AI statement. Every company can describe review processes and approval structures. But the real meaning of those claims is tested in much simpler moments.

• A customer says the decision makes no sense.
• A patient’s situation does not fit the score.
• A claimant has evidence the workflow did not capture.
• A support case carries a kind of vulnerability the model was never trained to interpret well.
• A fraud flag looks statistically plausible but humanly wrong.

In those moments, ethics is no longer a principle. It becomes a live question of whether the employee in front of the issue can act on their judgement and be backed when they do.

That is why frontline challenge rights matter so much. They are where all the lofty language either survives contact with reality or quietly collapses.

—

The post The real test of ethical AI is whether a frontline employee can challenge it appeared first on e27.

Bitcoin climbed 5.08 per cent over 24 hours to US$81,106.29, outpacing the broader crypto market’s 4.83 per cent advance to a US$2.72 trillion total capitalisation. The move did not happen in a vacuum. Bitcoin now trades with a 97 per cent correlation to the S&P 500 and an 86 per cent correlation to gold. Those figures reveal a macro-driven repricing rather than a crypto-specific breakout. When digital assets, equities, and bullion rise together, traders are responding to shifting expectations about Federal Reserve policy and global liquidity.

The main catalyst arrived from a dovish turn in rate expectations. Federal Reserve Governor Christopher Waller signalled he would be inclined to support holding interest rates steady if inflation data improved. Before those remarks, markets priced in a 63 per cent chance of a September rate hike. Afterward, that probability fell to around 50 per cent. A pause in US-Iran tensions added further relief by easing oil price fears and removing a recent inflation headwind. The White House also contributed dovish comments, reinforcing the message that the Fed would not tighten aggressively.

With Treasury yields no longer threatening to surge, capital flowed back into equities and high-beta assets like Bitcoin. This repricing pushed Bitcoin into near lockstep with the S&P 500, confirming its current role as a macro-sensitive asset. The immediate trigger for the next move comes from the August Non-Farm Payrolls report due on September 4. Soft jobs data would strengthen the dovish case and could extend gains. A strong report would revive fears of a rate hike and put pressure on the entire complex.

The advance also received a powerful boost from a short squeeze in the derivatives market. Traders liquidated over US$415 million in Bitcoin short positions over 24 hours, with US$164 million of that total vanishing in just four hours. A separate measure showed Bitcoin liquidations hitting US$271 million in 24 hours, with shorts accounting for 95 per cent of the total, a 425 per cent surge.

Forced buying by traders covering bearish bets created a feedback loop that accelerated the price climb. This dynamic explains the speed and intensity of the move. The macro news lit the fuse, but leverage did the heavy lifting afterward. Crowded bearish positions unwound in a cascade, and each wave of forced buying pushed prices higher. That mechanical accelerator is not the same as organic spot demand, and it can reverse quickly if momentum stalls.

Also Read: Bitcoin slipped below US$80,000, so why are traders still betting on US$82,000?

Sentiment indicators reflected the sudden shift. The Fear and Greed Index jumped to 78, signalling greed and overheated conditions. Influential market voices added fuel to the fire by declaring a crypto supercycle, which fed retail fear of missing out. Intense bullish social media narratives reinforced the move. This kind of euphoria often precedes volatility because latecomers chase the upswing after the initial institutional move.

The derivatives market will show whether leveraged speculation is stabilising or building toward another flush, as reflected in funding rates and open interest changes. The squeeze pushed prices higher, but the market now needs sustained spot buying and exchange-traded fund inflows to confirm the move as more than a one-day event.

From a technical standpoint, Bitcoin broke above the US$78,000 resistance with high volume, a bullish signal. The next major hurdle sits in the US$83,000 zone. That level aligns with long-term holder supply and previous rejection points. If Bitcoin holds above US$80,000, especially above the US$81,000 breakout level, the path could extend toward US$83,000 to US$85,000.

A clear break above US$83,000 would open the way toward the Fibonacci extension near US$86,500. If Bitcoin fails to hold above US$78,000, the surge likely came primarily from short covering and could fade. In that scenario, a pullback toward the US$78,000 to US$76,000 support zone would be the next test. The market needs a weekly close above US$81,000 to confirm the breakout and attract longer-term buyers.

Exchange-traded fund flows remain the key institutional signal. On September 2, spot ETFs recorded a net inflow of US$101 million. That is a positive sign, but one day does not establish a trend. If spot ETF inflows re-accelerate after the squeeze, they would provide fundamental support above US$80,000 and reduce the risk of a sharp reversal. Without that follow-through, the rally could lose steam once the forced buying from liquidations ends.

The market has seen this pattern before. Macro news triggers a spike, shorts get squeezed, and then the price drifts without new organic demand. The difference this time is the broader macro backdrop. If the Federal Reserve truly pivots to a dovish stance, the liquidity environment could support a longer rally. If the dovish signals prove temporary, the squeeze alone will not hold Bitcoin above US$80,000.

Also Read: Can Bitcoin defend the critical US$76,500 foundation zone before the September 11 inflation data triggers another massive liquidation cascade?

The near-term outlook is bullish but fragile. Bitcoin’s surge rests on two pillars: a macro reprieve and a violent short squeeze. The macro reprieve is real but depends on upcoming data. The August jobs report on September 4 will be the first test. Inflation data and Fed communications will follow.

Any hawkish surprise could unwind the rate cut expectations that underpinned this move. The second pillar, the short squeeze, has already spent much of its force. More than US$415 million in bearish positions are gone, and the Fear and Greed Index at 78 suggests the easy gains from sentiment reversal are behind us. The next leg higher requires spot buyers to step in with conviction.

From my perspective, Bitcoin’s 5.08 per cent move is a legitimate macro-driven breakout, but it is not yet a confirmed trend change. The 97 per cent correlation with the S&P 500 suggests this is a liquidity trade rather than a crypto-specific narrative. The US$101 million ETF inflow on September 2 is encouraging but insufficient on its own.

The real test is the US$83,000 resistance. If Bitcoin clears that level with sustained volume and ETF inflows accelerate, the path to US$86,500 becomes realistic. If Bitcoin rejects US$83,000 and falls back below US$78,000, the rally will look like a short-covering episode that ran out of fuel.

As I said, the next few days will reveal whether this is the start of a new leg or just a powerful bounce within a range. Watch the August jobs report, watch ETF flows, and watch how Bitcoin behaves around US$80,000 to US$81,000. Those signals will tell you more than any single price print.

—

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 Bitcoin just broke US$81,000: The real reason is not what you think appeared first on e27.

In the world of AI-powered content creation, ChatGPT Canvas is a game-changer. Unlike traditional ChatGPT interactions, where multiple prompts and re-prompts are necessary to fine-tune content, ChatGPT Canvas introduces a new, interactive interface that allows direct text manipulation, structured editing, and enhanced formatting options. This guide explores the benefits, key features, and practical applications of ChatGPT Canvas for content creators, marketers, and business professionals.

The challenges of traditional ChatGPT editing

One of the biggest challenges with using standard ChatGPT or other AI tools for content writing is the difficulty in editing specific sections of generated text. Users often find themselves repeatedly prompting and refining responses to achieve a satisfactory output. This process can be time-consuming and inefficient, particularly when working on long-form content such as articles, blog posts, or marketing copy. Since the AI generates text as a complete block, adjusting a single section often requires rewriting the entire response or editing the entire block part by part manually or via prompts. That may however introduce issues such as with keeping a consistent tone, style, or format throughout the content.

What is ChatGPT Canvas?

ChatGPT Canvas is an advanced interface that simplifies content generation, editing, and refinement. It allows users to:

• Edit text directly within the AI-generated response
• Adjust content length for different formats
• Modify reading levels to suit various audiences
• Format content using markdown for better readability
• Suggest edits and improvements with built-in recommendations

Also Read: The scarcity mindset is killing creativity, not AI

Key features of ChatGPT Canvas

• Interactive editing

Users can click directly into the generated text to make edits, eliminating the need to regenerate the entire response. This feature is particularly useful when refining specific sections of an article or making minor adjustments to wording and tone. Unlike traditional ChatGPT, where users must copy the generated text, paste it into another document, and manually edit, ChatGPT Canvas allows seamless in-place modifications without disrupting workflow.

• Content length adjustment

ChatGPT Canvas allows users to modify the length of their content effortlessly. Whether expanding a script for a longer video or condensing a post for social media, users can fine-tune the content to meet their needs. This feature eliminates the frustration of starting over due to content being too long or too short, making it ideal for those who need flexibility in their writing.

See for example before and after:

• Reading level adaptation

Different audiences require different levels of complexity in writing. With ChatGPT Canvas, users can adjust the reading level to cater to:

• Middle school students
• General web readers
• Academic or professional audiences

This feature ensures that content is accessible and appropriate for its target readers. By simplifying or elevating the language, users can tailor their messaging to resonate with their intended audience without needing to rewrite the entire content manually.

• Emoji and formatting support

For users who want to enhance engagement, ChatGPT Canvas supports:

• Automatic emoji integration
• Bold and italic text formatting
• Headings and subheadings
• Markdown support for web publishing

These features are particularly useful for social media posts, blog formatting, and digital marketing content. Formatting is essential for readability, and with ChatGPT Canvas, users can quickly transform plain text into a visually appealing format with structured elements.

• Version control and undo feature

One of the standout features of ChatGPT Canvas is version control. Users can:

• Restore previous drafts
• Undo changes without affecting the entire document
• Modify individual sections instead of re-prompting the AI

This makes content revision more efficient and user-friendly. Having the ability to track changes and revert to earlier versions ensures that no valuable content is lost during the editing process.

• AI-generated images

ChatGPT Canvas allows users to insert AI-generated images into their content. While the initial image quality may vary, refining prompts can significantly improve the output. This is useful for enhancing blog posts, articles, and marketing materials.

Pro tips: At times, instead of asking it to generate an image, I ask it to propose a few image and explain why.

Also Read: How creativity, commerce and AI collide in mid-2026 marketing mix

Areas and applications of ChatGPT Canvas

• Content marketing and social media

ChatGPT Canvas is ideal for creating structured marketing copy, including:

• Facebook and Instagram ads
• Email campaign templates
• Social media captions and descriptions
• Product descriptions and landing page copy

The ability to quickly generate multiple versions of ad copy allows marketers to test different messaging approaches efficiently.

• Blog and article writing

For writers and bloggers, ChatGPT Canvas streamlines the content creation process by:

• Generating structured outlines
• Allowing easy edits within the interface
• Improving formatting for web readability
• Supporting markdown for seamless publishing

These features make it an excellent tool for drafting, refining, and finalising long-form content.

• Scriptwriting and video content

YouTubers and video content creators can use ChatGPT Canvas to:

• Generate structured video scripts
• Adjust script length for different formats
• Enhance clarity and readability with formatting tools

This feature simplifies the scriptwriting process and allows for easy collaboration with editors and co-creators.

The takeaway

ChatGPT Canvas is a powerful tool that enhances AI-generated content creation. By allowing direct editing, structured formatting, and content customisation, it significantly improves efficiency and quality. However, to maximise its potential, users should combine AI-generated content with personal insights, real-world examples, and authentic storytelling, seeing it as an aid rather than a replacement and engage in the mass generating of content that lacks depth, originality, and personal touch, ensuring that the final output resonates with audiences and maintains credibility. This ensures the content remains engaging, credible, and unique.

—

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 and human creativity: How ChatGPT Canvas bridges the gap appeared first on e27.

Claude dashboards have become a kind of “AI flex” on Instagram recently, especially among founders, marketers, finance people and operators.

The biggest reason is that dashboards suddenly became very easy to make and very easy to share. Claude Code can now turn data or work from a session into an interactive Artifact: charts, tables, filters, KPI cards, timelines, and so on, and publish it as a webpage. Anthropic specifically promotes dashboards as one of the uses for Claude Code Artifacts.

That’s a meaningful shift. A year ago, a founder wanting something like this had two options: pay a developer to build it, or live with a spreadsheet nobody opens. Now the barrier is closer to zero, which is exactly why it’s spreading so fast. It’s not that the underlying analysis got smarter. It’s that the packaging got trivially easy.

And dashboards are perfect social-media content. Compare these two:

“I asked AI to analyse my business,”

versus a screenshot showing:

• REVENUE RM183,420 (US$45,368) up 18.4 per cent
• CONVERSION 3.8 per cent up 0.7 per cent
• TOP PRODUCT: Necklace

charts, graphs, customer segments, forecasts.

The second one looks like someone built a sophisticated internal software system. That’s highly shareable even when the underlying data analysis isn’t particularly complicated.

Dashboard does not equal business intelligence. A beautiful dashboard made from bad data is just a beautiful way of making a bad decision. If the revenue figure is wrong, or the “top product” metric is counting returns as sales, none of that shows up on the screenshot. It just looks confident. That’s the trap: the format signals rigour whether or not the underlying numbers earned it.

Also Read: In Southeast Asia, going global used to mean picking the biggest market, that logic is already dead

For a business, I’d rank the value like this:

• Reliable underlying data
• Correct KPIs
• Useful business questions
• Analysis and decision rules
• Dashboard design

Instagram naturally makes dashboard design look like the important part because that’s what photographs well. Nobody posts a screenshot of “we finally reconciled our SKU-level cost data,” even though that’s usually the unglamorous work that makes everything above it trustworthy.

For something like a jewellery business, for example, I’d much rather have a relatively boring dashboard that tells you sales, gross profit, SKU sell-through, inventory ageing, booth ROI, customer basket size, channel profitability, and flags where money is being lost, than a gorgeous 20-chart dashboard nobody actually uses. One of those tells you to reorder before you run out of your best-selling piece. The other just looks good in a founder’s story highlights.

So the trend itself is useful, but the opportunity isn’t “I should make a Claude dashboard too.” It’s “I can now build a lightweight management system without paying someone to develop one from scratch.” That’s much more interesting, and it’s the part that doesn’t screenshot nearly as 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.

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

The post The AI dashboard gold rush: Beyond the pretty charts appeared first on e27.

Here’s a test you can run on your own organisation. Think of the last genuinely important decision your team made. Now ask: did the data shape that decision, or did someone make the call and then reach for the data to justify it?

If you’re honest, it’s usually the second one. And that tells you something most “data-driven” organisations don’t want to hear: they’re not data-driven at all. They’re report-driven. They’ve gotten extraordinarily good at producing and consuming data, and they’ve never actually learned to decide with it.

Seeing is not deciding

We’ve spent a decade conflating two completely different things. One is seeing data, building the dashboard, running the report, watching the metric. The other is deciding with data, letting what you see change what you do, and then committing to that change.

Everything in the modern analytics stack optimises for the first. Prettier dashboards, faster pipelines, more metrics, real-time everything. And it’s all beside the point if the thing the data implies never actually gets done. A dashboard that changes no decision is just expensive decoration.

The appetite to close this gap is enormous, by the way. Salesforce’s 2026 research found that 93 per cent of business leaders say they’d perform better if they could just ask questions of their data in plain language. Read that as what it actually is: a near-universal admission that the data is there and people still can’t easily turn it into action. The bottleneck was never seeing. It’s the leap from seeing to deciding, and no amount of dashboard polish closes it.

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

Why we hide in reports

Report-driven cultures aren’t stupid. They’re rational responses to incentives, and the incentives are backwards.

Producing a report is safe. Making a decision is exposed. If I build you a dashboard and the business goes sideways, that’s not on me, I delivered the data. If I make the call the dashboard implied and I’m wrong, that’s very much on me. So smart, self-preserving people accumulate reports as a way of looking rigorous while avoiding the risky act of commitment. The dashboard becomes a place to hide.

Then there’s the certainty trap. People tell themselves they can’t decide yet because the data isn’t complete, so they ask for more. But more data rarely produces more clarity; it usually just produces more delay and a false sense that certainty is coming. It isn’t. Most real decisions have to be made on a strong-enough signal, not a perfect one, and a culture that waits for perfect is a culture that decides late, every time.

And underneath both: nobody’s accountable for the decision. We measure whether the report shipped on time. We almost never measure whether a clear signal in that report actually changed what anyone did. We’ve instrumented the production of insight and left the use of insight completely dark.

The fix is human, not technical

Here’s the part that should be encouraging. Because this is a behavioural problem, not a technology one, it’s fixable, and fixable without buying anything.

Start every analysis from the decision, not the question. “What will I do differently depending on the answer?” If nothing, don’t build the report. That one discipline kills half the dashboards nobody uses and forces every remaining insight to have somewhere to go.

Also Read: How AI and blockchain could make commerce decisions more accountable

Give people explicit permission to act on strong-enough signals. Name the difference between reversible and irreversible decisions, you can move fast and loose on the reversible ones, and most decisions are more reversible than people treat them.

And close the loop, relentlessly. What did we predict? What happened? What did we learn? Organisations that revisit their decisions build judgement that compounds. Organisations that don’t repeat the same hesitation forever, decision after decision, learning nothing.

The reframe

So stop measuring your dashboards and start measuring your decisions. Not how much data you have, every competitor has data. Not how many reports you produce, reports are cheap. Measure whether decisions actually change because of what the data showed, whether they get made fast enough to matter, and whether your organisation is getting better at making them over time.

The companies that win the next decade won’t be the ones with the best dashboards. Plenty of people will have great dashboards. They’ll be the ones whose people can stand in front of the data and decide, quickly, under uncertainty, and honest enough to check whether they were right.

Your dashboards are fine. That was never the problem. The problem is that you’re all looking at them, nodding, and then not deciding anything.

—

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 decision discipline: How to turn insights into action appeared first on e27.

Podcast production has exploded. The number of episodes published annually grew from roughly 3.9 million in 2015 to around 29 million in 2023.

Hours of valuable information are shared every day through this long-form audio. Yet despite how much useful information is buried inside podcasts, there still isn’t a comprehensive way to index them.

We have searchable indexes for news and databases for financial filings and academic papers. So what makes podcasts so much harder?

Having worked on indexing podcast content, I think the difficulty comes down to four problems: transcription, fragmented sources, content quality, and speaker identification.

Podcasts are audio first

Before you can make sense of a podcast, you first have to turn it into text. That sounds straightforward, especially given how much speech-to-text models have improved. But transcription is still far from a solved problem when you care about accuracy.

Proper nouns are a particularly difficult problem. Speakers constantly mention company names, product names, people, tickers, and industry-specific terminology that transcription models may not recognise. A company like Lyft, for example, can easily become “lift,” or Claude can become “cloud.”

This is a huge issue, especially when you are trying to build a searchable index. If the company name itself is transcribed incorrectly, the episode may never appear when someone searches for it. You need another layer that understands the context and corrects these transcription errors. The problem becomes even more challenging with different accents, speaking styles, recording environments, overlapping speakers, and poor audio quality.

The news on the other hand doesn’t have this problem. No form of processing is required as it’s already in the text form.

With podcasts, transcription is an additional computational step before indexing can even begin. At a large scale, that becomes a meaningful cost.

The source landscape is fragmented

The second problem is fragmentation. The barrier to creating a podcast is extremely low. Unlike traditional media, where a relatively small group of publications accounts for much of the trusted coverage, valuable podcast content can come from almost anywhere.

This is partly what makes podcasts interesting. You get perspectives, conversations, and expertise that would never appear in traditional media. But it also makes indexing them much harder. With news, there are a relatively limited number of publications that people consistently trust. You can get by with just indexing the major outlets.

Also Read: Asia’s AI trust gap: strong transparency, weak security and unclear data practices

Podcasts work very differently. A valuable piece of information might come from a huge show, a niche industry podcast, an independent expert, or a founder appearing on a tiny podcast with only a few thousand listeners. That means you cannot simply identify a few hundred trusted sources and call the job done. To build a useful podcast index, the coverage has to be dramatically wider. The long tail is not optional. It is often where the most interesting information lives.

That creates a scale problem that is easy to underestimate until you actually try to build it.

Content provenance and noise

The lack of editorial boundaries creates another problem: noise. AI-generated podcasts have become increasingly common. In our own work with financial podcasts, we have seen roughly 15 per cent of the content we encounter appear to be AI generated. Identifying and filtering this content is becoming surprisingly difficult.

I have worked around voice AI since 2023, and I used to think I had a good ear for identifying synthetic voices. I am far less confident today. As voice models improve, identifying AI voices has become a challenge.

There are other forms of duplication too. Podcasters frequently publish clips from other podcasts (reactions). An indexing system has to understand the difference. Is the person speaking actually a guest on this podcast? Is this a clip from another show? These problems can be solved using LLMs these days.

Not to forget, podcast advertising introduces complications. Podcasts increasingly use dynamically inserted ads, meaning the audio file itself can change depending on when or where it is played. Unlike an article sitting at a fixed URL with pretty much fixed content, podcast content is not always static.

Knowing what was said isn’t enough

The final problem may be the most important: speaker identification. Knowing that a statement was made is useful. Knowing who made it is far more useful.

Imagine someone saying that a particular company has an enormous competitive advantage. The meaning of that statement changes depending on whether the speaker is the company’s CEO, a competitor, an investor, a customer, or an independent industry expert.

The words might be identical but the perception will change accordingly. A useful podcast index needs to understand who is speaking and what their relationship is to the subject.

This is one area where modern AI models come in handy. Given enough context, they can often identify speakers, infer roles and resolve ambiguous references.

Also Read: Why Singapore’s AI finance race is now about data, not models

AI changes what is possible

A few years ago, building a comprehensive podcast index would have been theoretically unfeasible. You would need to transcribe millions of hours of audio, fix errors, identify speakers, and continuously process a huge stream of new episodes.

The only probable way would have been to hire troves of human annotators. LLMs and modern speech models change the economics of that problem. For the first time, it is becoming practical to turn podcasts from an audio format people have to manually consume into a structured data source that machines can understand. And that matters because the information inside podcasts is unusually valuable.

Executives explain how they think. Investors discuss their theses. Researchers describe work that may never appear in a paper. Founders talk about their companies in more detail than they would in a press release. Industry experts casually reveal insights buried inside hour-long conversations.

Until now, most of that information disappeared into the podcast feed after it was published. We are finally reaching the point where it doesn’t have to. Now that podcasts can be indexed, the more interesting question is what we can build once all of that information becomes searchable.

—

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 Why podcasts are the next big data revolution appeared first on e27.