Most of the industries are putting Artificial Intelligence (AI) into their products. However, in most cases, firms are merely layering the AI tools over their current systems, without revising how their products will work on the ground level. The integrations are usually at the level of basic automation and not much depth or transformation. AI-first products are built differently from those that simply include AI features. Instead of layering models onto existing systems, these tools are designed from the ground up to use machine learning and automation in their core functions. This influences the way data will be handled by the system, the way things will be executed, and the way it will react to real-time entries. Meanwhile, to understand what building truly AI-native products looks like, we spoke with Yose Rizal, a technology builder with over 15 years of experience turning data into tools for business, government, and media. He is the founder of MWX, a decentralized platform helping small and mid-sized businesses gain access to ready-made AI solutions through a marketplace model. Nonetheless, the construction of this nature has new problems with it as well. The tools should also be adjustable with time, consistent in their functionality, and self-explanatory to the individuals using them. Without a wise design, products are likely to be unstable, confusing, or Unreliable. In this interview, we explore what that process really takes. 1. What makes a product truly AI-native, beyond just connecting to a model or API? 1. What makes a product truly AI-native, beyond just connecting to a model or API? A lot of what we see today is what I’d call AI-enhanced, not AI-native, meaning companies are plugging in APIs like ChatGPT or image classifiers into existing tools, but the overall architecture still assumes a human is driving the system end-to-end. AI-enhanced To be AI-native, the product needs to be designed from the ground up with machine intelligence as the core operating layer, not just an add-on. That means three things: machine intelligence as the core operating layer Decision-making and execution flows are led by AI agents, not just assisted by them. In the case of SME, for instance, founders don't have to tell the system to "generate an ad" then "schedule the post"; our orchestrator AI can chain agents together automatically based on goals. That’s AI-native: outcome-driven automation.The product architecture anticipates learning, not static rules. In an AI-native tool, you don’t just push updates; you plan for model retraining, behavioral drift, and human-in-the-loop corrections. Agents are built to adapt, and our infrastructure supports that cycle.The UX is designed around collaboration with AI, not control panels for task-based commands. That’s a big shift. In an AI-native product, the user doesn’t just "click to deploy"; they interact with suggestions, review outputs, and build trust through guided transparency. Decision-making and execution flows are led by AI agents, not just assisted by them. In the case of SME, for instance, founders don't have to tell the system to "generate an ad" then "schedule the post"; our orchestrator AI can chain agents together automatically based on goals. That’s AI-native: outcome-driven automation. The product architecture anticipates learning, not static rules. In an AI-native tool, you don’t just push updates; you plan for model retraining, behavioral drift, and human-in-the-loop corrections. Agents are built to adapt, and our infrastructure supports that cycle. The UX is designed around collaboration with AI, not control panels for task-based commands. That’s a big shift. In an AI-native product, the user doesn’t just "click to deploy"; they interact with suggestions, review outputs, and build trust through guided transparency. So for me, AI-native means: it learns, it adapts, and it does real work for the user, not just with them for with 2. MWX offers AI tools as plug-and-play solutions for small businesses. What’s the process of turning complex AI systems into simple, usable tools? At MWX, we know small business owners aren’t trying to “use AI”; they’re trying to grow their customer base, save time on admin, or get a report out faster. So, we reverse-engineer the interface from their goal, not the model. The first step is identifying atomic business tasks: write a product description, test an ad, summarize financials. From there, we wrap each task with the minimal necessary configuration, usually just 2 to 5 inputs, and hide the complexity behind default workflows. Then comes orchestration. MWX isn’t just a bunch of isolated tools; it’s a marketplace where agents can link together. So, we’ve built a smart facilitator that can chain multiple agents to accomplish a bigger goal with zero coding required from the user. And last, we invest heavily in clarity: every output is transparent, editable, and traceable, so users feel in control, even when an agent is doing the heavy lifting. It’s not about dumbing AI down; it’s about translating it into something people can trust, tweak, and use without a manual. 3. As AI tools become more common, how should teams think about user experience and transparency? AI is powerful, but when users don’t understand how a decision was made or how to fix it, trust breaks down fast. In my preference, ideally, every agent is built around three UX pillars: explainability, reversibility, and observability. Explainability means showing the “why” behind a result, whether that’s highlighting which part of a prompt drove the output or visualizing which data points informed a recommendation. Explainability means showing the “why” behind a result, whether that’s highlighting which part of a prompt drove the output or visualizing which data points informed a recommendation. Explainability Reversibility gives users the confidence to try. They know they can undo, tweak, or roll back any AI action without breaking things. Reversibility gives users the confidence to try. They know they can undo, tweak, or roll back any AI action without breaking things. Reversibility Observability means surfacing cost, impact, and performance. We show users how much they spent, what the result was, and where improvements can happen, in human terms, not dev logs. Observability means surfacing cost, impact, and performance. We show users how much they spent, what the result was, and where improvements can happen, in human terms, not dev logs. Observability The goal is to make AI feel less like a black box and more like a smart teammate you can learn with. 4. How do feedback loops, retraining, and version control change when the product itself is AI-native? They stop being optional and start being part of the core product lifecycle. In an AI-native system, you're not just pushing feature updates; you're managing behavioral evolution. That means: behavioral evolution Every agent needs built-in feedback capture, explicit (like ratings), and implicit (like task abandonment). Every agent needs built-in feedback capture, explicit (like ratings), and implicit (like task abandonment). You need version-aware orchestration. If one model improves, it shouldn’t break the other agents connected to it. You need version-aware orchestration. If one model improves, it shouldn’t break the other agents connected to it. And you need retraining logic that’s modular, so you can fine-tune agents for different sectors (retail vs. F&B, for example) without starting from scratch. And you need retraining logic that’s modular, so you can fine-tune agents for different sectors (retail vs. F&B, for example) without starting from scratch. There is also what we call “soft retraining”, using anonymized behavioral patterns to optimize templates, prompts, and agent sequencing before touching the models themselves. So, AI-native products require not just DevOps, but ModelOps + UXOps as an integrated discipline. 5. In regions with limited access to data or technical infrastructure, what are the practical steps to make AI tools usable at scale? You don’t scale by waiting for perfect infrastructure; you scale by designing around constraints. In Southeast Asia, we see a few key principles play out: Edge-light architecture: Cloud-based agents that don’t demand local compute. Edge-light architecture: Cloud-based agents that don’t demand local compute. Edge-light architecture: Data-lean defaults: Agents that start useful even without personalized training. A small warung in Jakarta shouldn’t need a CRM database to run its first campaign. Data-lean defaults: Agents that start useful even without personalized training. A small warung in Jakarta shouldn’t need a CRM database to run its first campaign. Data-lean defaults: Offline-compatible onboarding: You’d be surprised how far WhatsApp flows, SMS, and simple mobile interfaces can go. Offline-compatible onboarding: You’d be surprised how far WhatsApp flows, SMS, and simple mobile interfaces can go. Offline-compatible onboarding: Localized templates and language support: AI that speaks English is nice. AI that speaks Indonesian, Vietnamese, and Taglish is what gets adopted. Localized templates and language support: AI that speaks English is nice. AI that speaks Indonesian, Vietnamese, and Taglish is what gets adopted. Localized templates and language support: And most of all, you need trust bridges, partners, ambassadors, or community leaders who can help introduce the tools with cultural context and confidence. AI scale isn’t just a data challenge. It’s a design challenge, and a distribution one. 6. MWX offers AI solutions to small businesses through a marketplace model. What lessons have you learned about packaging complex tools into accessible, no-code products? The biggest lesson? No-code doesn’t mean no context. When building a marketplace, we learned that surfacing 100 AI agents isn’t useful unless you help users figure out which one solves their problem. which one solves their problem So, we don’t just list agents. We build: Goal-based discovery ("I want more sales", "I need weekly reports")Bundled workflows, like “AI Starter Kit for Online Sellers”A concierge layer, a plain-English assistant that recommends agents based on business needs Goal-based discovery ("I want more sales", "I need weekly reports") Goal-based discovery Bundled workflows, like “AI Starter Kit for Online Sellers” Bundled workflows A concierge layer, a plain-English assistant that recommends agents based on business needs A concierge layer And on the backend, we make sure every agent has a common interface and shared schema, so you can chain tools together without any technical integration. It’s not enough to build powerful tools. You have to package them in a way that makes people feel ready to use them, even if they’ve never touched AI before. ready Conclusion: Building AI-native products is more than connecting to a powerful model or API. It requires a new mindset: one that reimagines how products are designed, how users interact with AI, and how systems evolve. Meanwhile, from architecture and UX to trust-building and deployment, every layer must be rethought. The future of software is not just AI-enhanced, it is AI-driven, and it begins with building from the ground up.
