Today, I’m tearing it all down. After two decades in marketing, I’m not just changing my title; I’m changing my DNA.

For years, I felt I had made it. I navigated startups from Brunei to Singapore, believing my deep understanding of strategy was a permanent safety net. I was wrong. Standing in Singapore in 2026, looking at a job market where digital marketing manager roles are being hollowed out by automation, that safety net feels like wet paper.

The Cerulean moment

There is a scene in The Devil Wears Prada where Miranda Priestly schools her assistant, Andy, on her lumpy blue sweater. Andy thinks she’s above the stuff in the room, but Miranda explains that her sweater wasn’t a random choice; it was the result of massive technical decisions and million-dollar investments. It was Cerulean.

Most marketers today are playing the role of Andy. Just like others, we use AI to write a caption or optimise a budget, and think we’re keeping up with the fashion of the times. We aren’t. We’re just consumers of a trend someone else engineered. In Singapore, if you’re just a user, you are replaceable.

The shift from sketch to supply chain

In 2026, the golden age of the creative has been replaced by the era of the operator. Think of it like fashion: the creative dreams up the sketch, but the gold rush—the real money—is in the supply chain. It’s in the manufacturing and logistics that get that Cerulean sweater onto the rack.

To stay relevant, I hope to be the Marketing Engineer who builds the engine room. I am moving away from the soft side of creative briefs and into the hard side of Agentic Orchestration—building systems that don’t just chat, but actually execute the entire runway show.

Also Read: How AI agents are quietly rewriting the growth marketing playbook

Three tips to own the runway in 2026

• Shift from content to context: Generic AI copy is fast fashion, where it is just cheap and disposable. I must learn how to use RAG (Retrieval-Augmented Generation) to feed AI with my 20 years of proprietary strategy, so it knows my personal branding instead of guessing.
  
• Be the plumber of legacy sprawl: Singapore businesses are terrified of the 2026 Model AI Governance Framework. Don’t be a prompt engineer; be the logic builder who can use platforms like n8n or Make.com to connect the messy data safely.
• Solve the double-data entry trap: The real gold is in regulated industries, which can be fintech and logistics. Start with Flowise to build and understand private agentic workflows that automate manual verification. Let us make sure the Cerulean fabric actually makes it across the border.

So, just like everyone out there, are you ready to put in the work? Share with me your thoughts. I need to know where the best gyms are. I’m looking for recommendations from the tech community, primarily in Singapore.

• Where are the engine rooms? Which labs (like those at One-North or the Sea AI CoE) are building infrastructure rather than just AI wrappers?
• Where do I learn the hard stuff? Who is teaching agentic logic and workflow automation for non-CS veterans?
• To the founders: Are you building a platform that moves beyond writing and into doing?

The runway for digital marketers has ended. The runway for marketing engineers is just beginning. Let’s stop wearing the lumpy blue sweater and start designing the Cerulean.

—

The post Why my 20-year marketing career is going under the knife appeared first on e27.

Featherless AI, the open-source AI infrastructure platform that hosts more than 3,000 models, has announced a new round of investment led by Kickstart Ventures, signalling a broader shift in how enterprises think about AI ownership and infrastructure independence.

The funding is intended to accelerate the company’s core mission: making open-source AI practical and reliable at scale. Unlike subscription-based access to proprietary large language models, Featherless AI allows enterprises to build on infrastructure they own rather than one they merely rent. This is a distinction the company says is increasingly critical as AI becomes central to business operations.

“The first wave of adoption was defined by proprietary, closed-door ecosystems. We provide a neutral ground for a second phase,” the company said in a statement.

The company framed the investment as marking a turning point in the AI market. “While the first wave of adoption was defined by proprietary, closed-door ecosystems, we provide a neutral ground for a second phase where companies can own and run their own models without being tethered to a single cloud provider or a restricted tech stack.”

A central pillar of the Featherless AI strategy is hardware diversity. Through a strategic partnership with AMD, the platform ensures its catalogue of open-source models runs natively on AMD’s ROCm.

Also Read: Without governance, AI agents risk becoming enterprise chaos engines

The company says this provides a “competitive, auditable alternative to proprietary hardware systems”, giving businesses a structural cost advantage over those locked into single-vendor GPU ecosystems.

Joan Yao, a General Partner at Kickstart Ventures, highlighted the platform’s relevance to emerging markets. “Featherless is making frontier AI accessible at a fraction of the cost — and that matters enormously in markets like Southeast Asia, where the next wave of AI-native builders shouldn’t have to pay hyperscaler prices to compete,” she said. “That’s exactly the kind of infrastructure bet we want to be behind.”

Countering AI monopolies with open architecture

Beyond commercial viability, Featherless AI has positioned itself as a structural counterweight to what it describes as the danger of AI monopolies. By ensuring that state-of-the-art models remain accessible outside proprietary “walled gardens”, the platform aims to preserve developer flexibility and prevent any single company from gatekeeping the tools used to build the next generation of applications.

The company’s technical credibility rests on original research. The founding team is responsible for RWKV, a breakthrough open-source architecture developed as a challenger to the transformer models that have dominated the field since the publication of the “Attention Is All You Need” paper in 2017. RWKV offers an alternative design that the team argues is more efficient and equally capable, a claim that has attracted significant attention from the research community.

For enterprises weighing the cost and strategic risk of reliance on closed AI systems, Featherless AI is presenting itself as a third path: the performance and breadth of a managed platform, without the dependency on a proprietary provider.

—

Image Credit: Featherless AI

The post Featherless AI secures backing to make open-source models viable at enterprise scale appeared first on e27.

Software engineers with AI skills now earn between 13 per cent and 25 per cent more than their peers, marking a significant shift from just a year ago when AI capabilities had little effect on compensation, according to a new industry report.

NodeFlair, a tech career platform based in Singapore, published its Tech Salary Report 2026 this week, drawing on more than 230,000 verified salary data points across roles and markets.

The findings point to a tech labour market in which AI fluency has moved from a desirable attribute to a measurable financial advantage.

“AI fluency is no longer a nice-to-have. It’s now a salary advantage,” said Ethan Ang, founder of NodeFlair. “Just a year ago, coding with large language models still felt more experimental than transformative for many teams. In 2025, that changed quickly.”

The report also reveals a widening salary gap between junior and senior engineers. Salaries for senior, lead and manager-level roles rose 10.8 per cent or more, compared with 5.3 per cent for junior positions and just 1.7 per cent for mid-level engineers.

Also Read: Philippines’s talent deficit is becoming an economic risk

The divergence reflects growing demand for experienced engineers who can make architectural decisions, manage ambiguity and deploy AI tools to greater effect.

Ang said engineering leaders described AI as increasingly capable of handling execution tasks traditionally assigned to entry-level staff, e.g writing routine code faster and at lower cost. However, he noted that higher-order skills such as system design, trade-off analysis and navigating complex requirements remain areas where experienced engineers hold a clear edge.

At the top of the market, the highest-earning 10 per cent of engineers saw salary increases of up to 19 per cent, further widening the gap between top performers and the broader workforce.

What changed in 2025

NodeFlair attributed the turnaround to two converging factors: the maturation of AI coding tools into production-grade workflows, and a shift in how employers assess technical talent.

In 2024, many companies were still running pilots, and the productivity case for AI remained unclear. By 2025, tools enabling agentic coding workflows had become widely adopted, making the return on AI investment more tangible and prompting companies to price AI skills accordingly.

For early-career engineers, Ang urged embracing AI rather than treating it as a competitive threat. He noted that, on the ground, younger engineers have been quicker to adopt AI tools than their senior counterparts, and that pairing those skills with strong fundamentals in problem-solving and system design remains the most durable path to career value.

Also Read: What happens when AI starts talking to AI at work

The next wave

Looking ahead, NodeFlair expects the largest salary premiums to accrue not to those holding AI-specific job titles, but to professionals who combine domain expertise with practical AI execution — product managers who can prototype with AI, data professionals who can move AI models into production, and engineers who can work fluidly alongside AI agents.

“The biggest premiums will go to people who can combine domain expertise with AI execution,” Ang said. “Not just knowing the tools, but knowing how to apply them to create measurable business value.”

—

Image Credit: NodeFlair

The post AI skills now translate into real pay gains for software engineers, NodeFlair finds appeared first on e27.

The financial services industry does not have a shortage of fraud tools. However, most fraud and AML platforms were built for a world where attacks were opportunistic and patterns were predictable. That world is gone.

Today, fraud operations run with the discipline of a software company’s continuous iteration, AI-generated synthetic identities that pass KYC, and cross-channel coordination that exposes every silo in a legacy architecture. The platforms built to catch yesterday’s fraud cannot keep up with today’s.

Zoven is built for this moment.

Founded by Vivek Karna and Prashanth YV, both fintech veterans who spent years watching fraud and compliance teams fight a losing battle with the wrong tools, Zoven is the platform they wished had existed.

Beyond rules. Beyond dashboards.

At the heart of the Zoven platform is its Fraud Risk and AML product  and it is different from every other tool in this category in one fundamental way.

Traditional fraud and AML platforms offer three things: rule creation, case management, and regulatory report generation. These are the baselines but also backward looking. They do not help your team understand what it means, why it matters, or what to do about it. Zoven does all three and then adds an intelligence layer that no legacy platform can replicate.

Feed Zoven your Standard Operating Procedures. Our AI agents execute every step, run the investigation end to end, and produce detailed reports with crisp summaries. Your fraud and AML teams don’t start cold. They start informed, and they move fast.

Consider what this means operationally. A fraud analyst opening a case today spends the first hour on retrieval  pulling transaction history, running entity checks, reviewing related cases. None of that is analytical work. Zoven’s AI agents complete it in seconds and deliver a structured preliminary report the moment the case is opened. What the analyst does next is reasoning, not retrieval. It is the work that actually stops fraud.

Now open for onboarding

Zoven is now accepting early customers across India, the United States,  and Southeast Asia. The programme is designed for banks, fintechs, and payment facilitators that are ready to move beyond legacy fraud infrastructure.

Early customers receive direct access to the Zoven product team, SOP integration support, a custom pilot scoped to one product or fraud typology, and founding customer pricing. Institutions operating with Zoven’s early access programme have seen preliminary investigation time drop from hours to minutes within the first 30 days of deployment.

The best way to understand what Zoven does is to see it on your own data. Institutions interested in early access are invited to book a demonstration at zoven.ai.

About Zoven

Zoven is an AI-native risk intelligence Platform building infrastructure for the next generation of fraud prevention, AML compliance, and merchant risk management. Its platform manages the entire merchant risk lifecycle  from onboarding and transaction monitoring through to chargeback resolution and regulatory reporting  through a single intelligent system. Zoven is headquartered in Bengaluru, India, and is currently onboarding customers across India, the United States, and Southeast Asia.

—

Want updates like this delivered directly? Join our WhatsApp channel and stay in the loop.

This article was sponsored by Zoven

We can share your story at e27 too! Engage the Southeast Asian tech ecosystem by bringing your story to the world. You can reach out to us here to get started.

Featured Image Credit: Zoven

The post Zoven AI launches the AI-native platform that puts fraud and AML teams back in control appeared first on e27.

In the past three years, two of the most well-funded veterinary technology startups in the world have quietly shut down.

Fuzzy, a US telehealth platform for pets, raised US$80.5 million before closing in 2024. ZumVet, Singapore’s best-known vet-tech startup, raised US$3.7 million before winding down the same year. Neither failed because the founders were wrong about the problem. Pet healthcare is genuinely broken — fragmented records, inconsistent quality, and no data continuity when an animal changes clinics.

They failed because they tried to solve it with software first.

I’ve spent the last two years building veterinary infrastructure across Thailand, Myanmar, and Laos, and I’ve come to believe that the sequence matters more than the product. In regulated, trust-based industries, software is the last layer you build, not the first. If you invert that order, your burn rate scales faster than your adoption, and no amount of funding will close the gap.

The software-first trap

The standard vet-tech playbook looks rational on paper. Build a clean, modern product. Sell it directly to clinics or consumers. Use growth capital to subsidise acquisition until network effects kick in.

The problem is that veterinary medicine is not a software adoption problem. It is a trust and workflow problem. Clinics don’t switch their practice management system because the new one has a better interface. They switch — or don’t switch — based on whether their technicians can run a full day of appointments without the system failing, whether patient histories survive migration, and whether the vendor will still exist in five years.

None of those is solved by a better onboarding flow. They are solved by being embedded in the infrastructure that clinics already depend on — regulatory registries, diagnostic equipment, referral networks, and supplier relationships. Companies that start with software have to manufacture this trust through sales and marketing spend. Companies that start with infrastructure inherit it.

What we did differently

When we started building in Southeast Asia, we deliberately reversed the order. We built the network before we built the SaaS.

The first product was a microchip registry — arguably the least exciting product in vet-tech. It is also one of the most load-bearing. A microchip registry is the one database a clinic cannot operate without once microchipping becomes mandatory. It touches government, breeders, importers, insurance, and every clinic that scans an animal.

Also Read: How SMEs can vet and choose AI partners that truly deliver

We spent three years on registry, partner onboarding, and regulatory relationships before launching a full practice management system. Today, the network covers more than 880 partner hospitals and over 110,000 registered animals across three countries, with roughly 30 per cent market share in Thailand. The SaaS layer, which we recently launched, sits on top of that — not in front of it.

The point is not that microchips are the answer. The point is that in any regulated vertical, there is usually one boring piece of infrastructure that every participant has to touch. If you build that piece first, every subsequent product you launch inherits distribution. If you skip it, every product launch is a cold start.

Three things I’d tell another founder

• First, identify the load-bearing layer in your industry before you decide what to build. In vet-tech, it is registries and diagnostic workflow. In fintech, it is KYC and settlement. In logistics, it is customs and warehousing. These are rarely the most fundable ideas, because they look operational rather than technological. That is exactly why they are defensible once you own them.
• Second, accept that the first three years will look slow by venture standards. We operated for most of our early history without institutional funding. That was partly constraint and partly choice — raising a large round early would have pushed us toward the software-first playbook, because capital needs to be deployed into things investors can measure quarterly. Infrastructure does not produce quarterly metrics. It produces compounding ones.
• Third, be sceptical of comparables. When a founder in your category raises US$50 million and gets written up as the category leader, the instinct is to copy the model. But the company that raises first is not always the company that wins. In veterinary software specifically, the correlation between funding raised and long-term survival has been negative. The companies still operating in Southeast Asia are almost entirely bootstrapped or lightly funded. That is not a coincidence.

Also Read: How telemedicine can revolutionise the veterinary world?

Timing matters, but only if the infrastructure is ready

Regulatory tailwinds are arriving across the region. Thailand made pet microchipping mandatory in January 2026. Malaysia announced a mandatory pilot the following month. More markets will follow.

Regulation is a powerful forcing function, but it rewards whoever is already operating the infrastructure. It does not reward whoever has the best-marketed software. A founder who starts building a registry the day the mandate is announced is already three years behind.

The broader lesson, I think, is that the most common failure mode in regulated verticals is not building the wrong product. It is building the right product in the wrong order. Software is easier to build than trust, which is why founders default to it. But in industries where the end user has to believe the system will still work in a decade, trust has to come first, and software has to serve it.

The vet-tech graveyard is not a story about bad products. It is a story about inverted sequences.

—

The post Why vet-tech keeps failing: The case for network-first infrastructure appeared first on e27.