Home

Why It Is So Hard to Sell Core Banking Systems

Convincing a bank to replace its core banking system should be easy in theory. After all, the promise is huge: a safer, more reliable, and compliant engine that could save banks and their customers millions. Yet, in practice, it is one of the hardest sales in technology. The reasons go far beyond technology—they lie deep within the human, political, and regulatory fabric of the banking industry.

At the heart of the challenge are the people running the banks. Broadly speaking, there are two types. The first are stewards—or what I call “babysitters.” These bankers are conservative, risk-averse, and focused on not rocking the boat. They’re not sabotaging the bank’s future, but they aren’t championing innovation either. Their mindset is survival, not transformation. The second type are mavericks—rare individuals who look beyond their own tenure. They want to future-proof the bank, save customers money, and avoid being the “Kodak” or “Blockbuster” of finance. They embrace technology as the only way to adapt to a rapidly changing world. Unfortunately, the majority of decision-makers lean toward the steward side, while true mavericks are few and far between.

Replacing a core banking system is never a decision made by one person. Banks are regulated entities with complex governance structures, and every move requires the blessing—or at least the “non-objection”—of multiple stakeholders. Statutory board members are ultimately the only people who can sign off, but they face personal liability, which makes them cautious. CEOs may be the public face of the bank but are rarely the driving force. CFOs scrutinize the business case, sometimes comparing apples to oranges. COOs, responsible for both operations and IT, are often the most critical allies. Risk and compliance officers demand assurance that the new system meets regulations and often view change as more dangerous than sticking with the old. Security teams believe their standards are the best in the industry and impose their own requirements, while legal and procurement pore over contracts and policies, often dragging the process out. Ironically, the IT and operations staff who must actually use the new system often resist the most, fearing job loss or exposure of past failures.

Beyond the bank itself, external stakeholders shape the process as well. Regulators, auditors, consultants, and even investors all influence decision-making. Each has its own agenda, and alignment is rare. Regulators add yet another layer of complexity. Prudential authorities assess outsourcing risk, conduct regulators scrutinize consumer protections, privacy authorities demand adherence to strict data laws, and fiscal authorities determine tax implications that can reshape the business case. At best, regulators don’t object; rarely do they explicitly endorse. This culture of caution trickles down into every decision.

Selling a core banking system isn’t only about logic and cost savings—it’s about emotions and politics. Employees worry about their jobs. Executives worry about their reputations. Consultants may profit from legacy system failures and thus resist change. Sometimes resistance is downright irrational. One operations head once disliked automation because it replaced eighty people with eight. These personal and political dynamics often outweigh even the most compelling financial case.

For those selling core banking systems, the process feels less like sales and more like a long game of chess. Each stakeholder is a piece with unique powers, blind spots, and motivations. Winning requires patience, strategy, and the ability to anticipate moves many steps ahead. Brute force doesn’t work. Nor does a single champion. Deals are closed only when every piece is aligned—legal, regulatory, political, operational, and emotional.

The great irony is that the very systems banks cling to—patched, outdated, and inefficient—are the greatest risks to their survival. But change only happens when the right stakeholders see beyond the short-term and choose to embrace the future. Until more mavericks rise into leadership positions, the default stance of most banks will remain caution, not transformation.

為什麼核心銀行系統這麼難賣

理論上，要說服一家銀行更換其核心銀行系統應該很容易。畢竟，承諾是巨大的：更安全、更可靠、更符合監管要求的引擎，可以為銀行和客戶節省數百萬美元。然而，在實際情況中，這卻是科技領域最難推銷的產品之一。原因早已超越了技術本身，而是深深根植於銀行業的人性、政治與監管結構之中。

挑戰的核心是經營銀行的人。大致而言，可以分為兩類。第一類是「看守者」——我稱他們為「保姆」。這些銀行家保守、害怕風險，專注於維持現狀，不想讓局面動盪。他們並沒有蓄意破壞銀行的未來，但也不會推動創新。他們的心態是「生存」，而不是「轉型」。第二類是「先行者」——少數能跳脫個人任期限制的領導者。他們希望讓銀行能長遠經營下去，為客戶節省資金，並避免成為金融界的「柯達」或「百視達」。他們擁抱科技，因為這是唯一能應對快速變化世界的方法。不幸的是，大多數決策者偏向前者，而真正的先行者少之又少。

更換核心銀行系統從來不是某個人單獨能做出的決定。銀行是受嚴格監管的機構，治理結構複雜，每一步都需要多方利害關係人的同意——或者至少「不反對」。最終能簽字批准的只有法定董事會成員，但他們必須承擔個人責任，因此非常謹慎。CEO 也許是銀行的公開代言人，但很少是推動這類交易的人。CFO 會檢視商業案例，有時卻出現「橘子比蘋果」的錯誤比較。COO 負責營運與 IT，通常是最關鍵的支持者。風險與合規主管需要確保新系統符合規範，並且往往認為變革比維持現狀更危險。資訊安全團隊認為自己的標準最好，會提出額外要求。法律與採購部門則反覆檢查合約與政策，拖慢進度。諷刺的是，真正要使用新系統的 IT 與營運人員，卻往往最強烈反對，因為他們擔心失去工作，或害怕舊錯誤被揭露。

除了銀行內部，外部利害關係人同樣影響決策。監管機構、審計師、顧問，甚至投資人都會左右結果。每個角色都有自己的議程，要達成一致非常困難。監管機構更是增加了層層複雜性。審慎監管機關會評估外包風險，行為監管機構會檢查銀行是否保障消費者利益，隱私機構要求嚴格遵守個資法規，財稅機構則制定稅務規則，直接影響成本結構。監管者通常最多只會「不反對」，很少明確支持。這種謹慎文化滲透到銀行的每一個決策中。

銷售核心銀行系統不僅僅是邏輯與成本的問題，更是情感與政治的角力。員工擔心失業，主管擔心名聲受損，顧問有時靠舊系統出錯來賺錢，因此對改變沒有興趣。有時反對甚至完全不合邏輯——某位營運主管就因為自動化「太有效」而反對，理由是它讓八十人的工作縮減到八人。這些個人與政治因素往往比再強而有力的財務理由還要來得重要。

對於那些試圖推銷核心銀行系統的人來說，這個過程更像是一場長期的西洋棋對局，而不是單純的銷售。每個利害關係人都是棋盤上的棋子，各自擁有不同的力量、盲點與動機。要贏得比賽，需要耐心、策略，並能預測數步之後的局勢。蠻力行不通，單一支持者也不夠。只有當法律、監管、政治、營運與情感等各方面全部達成一致時，交易才可能完成。

最大的諷刺是，銀行緊抓不放的舊系統——那些補丁累累、過時低效的東西——才是它們生存的最大風險。但真正的改變只有在關鍵決策者能超越短期顧慮，選擇擁抱未來時才會發生。在更多先行者走上領導位置之前，大多數銀行的默認姿態將依然是謹慎，而非轉型。

Why Most Businesses Don’t Qualify for Venture Capital Funding

Recently I had dinner with a friend, and she mentioned that she wants to start her own business. I shared my experience that there are really two types of startups: one is the SME, or small and medium enterprise, which focuses on building a profitable business; the other is the venture capital-backed startup, which is a completely different game. Many people don’t realize how stark the difference is, and this misunderstanding often leads entrepreneurs to chase the wrong kind of funding.

Venture capital is a highly specialized business. Great companies can still be terrible VC investments. The first big misunderstanding is about profitability. Entrepreneurs naturally want to build profitable companies. Venture capitalists, however, are not interested in profits, at least not in the short term. Profits usually mean slowing down growth, and growth is the only thing that matters to investors. They prefer companies that reinvest or even overspend every dollar to capture markets as quickly as possible. If a company ever finds itself generating more cash than it can reinvest effectively, investors will usually push for a sale, an IPO, or even a leadership change to keep the money flowing back into growth.

Margins are another critical issue. Reasonable margins might look appealing to most business owners, but venture capitalists want outrageous margins. Software that can be copied endlessly, drugs that can be sold globally, and technologies that scale without much additional cost are far more attractive than services. Businesses that require adding more people or more assets for every increment of growth simply don’t fit the VC model. Platforms or products that scale exponentially with minimal cost are what investors are after.

Another mismatch is return expectations. Doubling an investor’s money sounds impressive, but in venture capital it is not nearly enough. Because of the way funds are structured—with 10-year horizons, management fees, and the need to offset inevitable failures—VCs need to aim for tenfold returns just to stay competitive with the stock market. A steady business that can deliver consistent but limited gains will never satisfy those requirements.

Even if the economics made sense, the personal side often doesn’t align. Venture capitalists look for certain founder profiles: entrepreneurs they’ve already backed successfully, people with high-profile achievements in related industries, graduates of elite schools, or extremely charismatic personalities. Many solid business operators simply don’t fit the mold. Bias also plays a role, as women and minorities have historically been funded at lower rates. The truth is, who you are matters almost as much as what you’re building.

Fundraising itself is also a massive hurdle. For most first-time founders, raising a VC round takes six to twelve months of full-time effort, often ending with nothing. Meanwhile, the business they were supposed to be running suffers from lack of attention. For someone trying to manage operations day-to-day, that trade-off is impractical.

And then there’s the issue of control. Running your own business comes with freedom, but raising venture capital usually means adding investors to the board, and those investors have the power to replace you. Many founders don’t fully realize that taking VC money also means taking on a new boss.

So should you raise venture capital? Maybe. If you’re building software, biotech, or another kind of business that can deliver the explosive growth and outsized returns investors demand, venture capital can accelerate everything. It provides not only money, but also connections, credibility, and confidence. But for most businesses, there are better options. Revenue-based financing, crowdfunding, or simply reinvesting profits are often better aligned with the goals of the company. At the end of the day, venture capital isn’t about building good businesses. It’s about building VC-scale businesses. And that’s a very different thing.

Being in Singapore, I see a unique dynamic. Despite the government providing a lot of funding schemes and initiatives to cultivate startups, the local culture is not particularly entrepreneurial in the VC sense. At best, it leans toward SME-style ventures—focused on steady profits and sustainability, rather than aiming to grow fast, burn money, and get rich quickly through an IPO.

為什麼大多數企業不適合尋求創投資金

最近我和一位朋友共進晚餐，她提到自己想要創業。我分享了自己的經驗，其實創業有兩種類型：一種是中小企業，專注於建立一個能夠獲利的生意；另一種則是透過創投（Venture Capital, VC）融資的創業，這是一個完全不同的遊戲。許多人沒有意識到這兩者之間的巨大差異，因此常常誤走方向，追逐不適合自己的資金來源。

創投是一個高度專業化的行業。即使是很好的公司，也可能是很糟糕的創投投資標的。第一個常見的誤解是關於「獲利」。企業家自然會希望建立一個有利潤的公司，但創投短期內對獲利並不感興趣。獲利往往意味著放慢增長，而投資人只在乎成長。他們希望公司把每一分錢都再投入，甚至超支，以便盡快擴張市場。如果一家公司現金流超過了它能有效再投資的能力，投資人通常會推動出售、IPO，甚至換掉管理層，以確保資金繼續被用來加速成長。

毛利率也是關鍵問題之一。對於一般企業主來說，合理的毛利率已經很不錯，但對創投而言，他們追求的是極高的毛利率。軟體可以無限複製、藥品能夠全球銷售、科技產品能在成本不大幅增加的情況下擴張，這些才是創投眼中的理想標的。而需要增加更多人手或資產才能增加收入的生意，通常都不符合創投的模式。能以極低成本快速擴張的產品或平台，才是他們所追逐的方向。

投資回報的期待也是一大差距。把投資人的資金翻倍，聽起來很厲害，但在創投的世界裡，遠遠不夠。由於基金結構的關係——十年的週期、高額的管理費，以及必須抵消必然失敗的投資案——創投需要追求十倍的回報，才能保持與股市相當的競爭力。那些能帶來穩定但有限回報的企業，根本無法滿足這樣的要求。

即使在財務上勉強說得過去，個人條件也常常不對盤。創投偏好某些特定的創業者類型：曾經替他們賺過大錢的企業家、在相關產業有高知名度成功經歷的人、來自頂尖學府並已經創造出驚人成果的畢業生，或是極具魅力的 A 型人格。許多優秀的經營者根本不符合這些特質。而且，偏見也會影響結果，女性和少數族裔的融資成功率歷來都較低。現實是，你是誰，往往和你要做什麼一樣重要。

募資本身也是巨大的挑戰。對多數第一次創業的人來說，完成一輪融資通常需要六到十二個月的全職努力，最後還很可能一無所獲。與此同時，他們原本該經營的事業就被擱置了。對於需要每天親自打理運營的人來說，這樣的取捨幾乎不可能。

最後還有控制權的問題。自己經營生意最好的地方在於自由，但一旦引入創投，通常就要讓投資人進入董事會，而這些人擁有更換你的權力。許多創業者沒有充分意識到，拿了創投的錢，意味著你重新有了「老闆」。

所以，你該不該尋求創投？也許吧。如果你正在打造的是軟體、生物科技或其他能帶來爆炸性增長和巨大回報的業務，創投能加速一切。它不僅帶來資金，還提供人脈、聲譽與信心。但對於大部分的企業而言，其他方式可能更合適。基於營收的融資、群眾募資，或是單純地將利潤再投資，往往更符合公司的目標。歸根結底，創投不是用來打造「好公司」，而是用來打造「適合創投規模的公司」。這是完全不同的概念。

身在新加坡，我看到的是另一種現象。儘管政府提供了許多資金計劃和創新倡議來培養創業環境，但本地文化並不是特別創投導向。最多只能說比較偏向中小企業風格——追求穩定的獲利與可持續性，而不是快速成長、燒錢，然後透過 IPO 一夜致富。

The Personal Qualities That Define Great Leaders

Leadership is never just about knowledge or strategy. It is about character, conviction, and the ability to inspire. Skills may help you solve problems, but it is your personal qualities that determine how far you can go, how resilient you will be, and how deeply you will influence those around you.

The greatest leaders are united by one thing: a relentless belief in their mission. They don’t just talk about goals; they embody them. Anchored in deeply held values, they push forward even when the odds are stacked against them. Every failure becomes fuel for growth, every success a stepping stone to the next horizon. Conviction, resilience, and courage are not optional traits. They are the foundation of enduring leadership.

Today’s leaders face a world of contradictions. They must drive vision from the top while empowering voices from the ground. They must serve both customers and employees. They must balance long-term investments with short-term pressures. They must dream big with divergent thinking while executing with laser-focused discipline. Great leaders are not paralyzed by these tensions; they embrace them. They act with courage, adapt quickly, and never stop learning. While no contradiction can be perfectly solved, dialogue, agility, and vision allow leaders to chart a path forward.

Look at Jeff Bezos guiding Amazon through the chaos of the dot-com crash. Many companies collapsed, but Bezos’s unwavering belief in Amazon’s mission to be Earth’s most customer-centric company carried the organization through near-death moments and laid the foundation for its global dominance. Consider Steve Jobs, who was once ousted from Apple but returned years later to rebuild the company into a symbol of innovation and creativity. His passion, his obsession with design, and his ability to inspire people with a vision of changing the world made Apple one of the most admired companies on earth. And beyond business, think of Lee Kuan Yew. With foresight and conviction, he transformed Singapore from a small, struggling nation into a thriving global hub. His clarity of purpose, pragmatism, and values-driven leadership carried a people through uncertainty and built a lasting legacy.

Another defining trait of great leaders is their relentless growth. Compare them today to who they were just a few years earlier, and the transformation is striking. Bezos evolved from a founder with an online bookstore idea to a global strategist reshaping industries. Jobs turned personal setbacks into fuel for creative reinvention, building Pixar and later revolutionizing consumer technology. Lee Kuan Yew continually adapted his policies to match the world’s shifts, learning and evolving while staying true to his vision of Singapore’s survival and prosperity. Great leaders never remain the same; they reinvent themselves, and through that reinvention, they reshape the organizations and societies they lead.

Your knowledge matters, but knowledge alone is not enough. What defines you as a leader is your resilience when tested, your values when pressured, your ability to balance contradictions when pulled in opposite directions, and your hunger to keep growing. Leadership is not static; it is a living journey of courage, learning, and inspiration.

So when challenges come, as they always do, remember this: leadership is not just about what you know. It is about how you stand tall in adversity, how you inspire trust and hope, and how you choose to lead when the world is watching. Great leadership is born not from certainty, but from conviction and the courage to keep moving forward.

定義偉大領袖的個人素質

領導力從來不僅僅是知識或策略，它真正關乎的是人格、信念，以及激勵他人的能力。技能或許能幫助你解決問題，但決定你能走多遠、能承受多少壓力、能影響多少人的，是你的個人素質。

最偉大的領袖有一個共同點：他們對使命的堅定信念。他們不只是談論目標，而是身體力行，將目標化為信念。憑藉深植內心的價值觀，他們即使在逆境中也能繼續前行。每一次失敗都成為成長的燃料，每一次成功都是邁向更高境界的踏板。信念、韌性與勇氣不是可有可無的特質，而是持久領導力的基石。

當今的世界充滿矛盾，領袖必須在拉扯之間找到平衡。他們既要推動自上而下的願景，又要賦能基層的聲音；既要服務顧客，又要關注員工；既要兼顧長期投資，又要面對短期壓力；既要用發散思維大膽創新，又要用專注思維精準執行。偉大的領袖不會被這些矛盾束縛，他們懂得擁抱矛盾，勇敢行動，快速調整，不斷學習。雖然矛盾無法徹底消解，但透過對話、靈活與遠見，他們能為未來找到前進的方向。

想想傑夫·貝佐斯（Jeff Bezos）如何帶領亞馬遜走過網路泡沫的崩潰。無數公司倒下，但貝佐斯對「成為全球最以顧客為中心的公司」的使命堅信不疑，使亞馬遜從瀕臨毀滅中走出，並奠定了全球霸主的基礎。再看看史蒂夫·賈伯斯（Steve Jobs），他曾被逐出蘋果，卻在多年後回歸，將公司重建為創新與設計的象徵。他的熱情、對設計的執著，以及以「改變世界」為核心的願景，使蘋果成為全球最受尊敬的企業之一。而在政治領域，新加坡的李光耀憑藉遠見與堅定，將一個資源有限、前途未卜的小國轉型為繁榮的全球樞紐。他清晰的使命感、務實的作風與價值驅動的領導力，不僅帶領人民度過不確定的時代，更奠定了持久的國家基業。

另一項定義偉大領袖的特質是持續成長。將他們現在與幾年前相比，你會驚訝於他們的蛻變。貝佐斯從一位創辦線上書店的創業者，成長為改變產業的全球戰略家。賈伯斯將個人挫折轉化為創意能量，打造出皮克斯，隨後更徹底革新了消費科技。李光耀則不斷根據世界局勢調整政策，在保持新加坡生存與繁榮願景的同時，不斷學習與進化。偉大的領袖從不原地踏步，他們在自我再造的過程中，也重塑了所帶領的組織與社會。

知識固然重要，但光靠知識不足以支撐你度過風暴。真正定義領袖的，是他在考驗中展現的韌性、在壓力下堅守的價值觀、在矛盾中找到平衡的能力，以及對成長的渴望。領導不是靜止的狀態，而是一場關於勇氣、學習與啟發的生命旅程。

所以，當挑戰來臨時——而挑戰永遠會來——請記住：領導不只是你知道什麼，而是你如何在逆境中挺身而出，如何激發信任與希望，以及你在眾人矚目之下選擇如何領導。偉大的領導不是誕生於確定，而是源於信念，以及不斷向前的勇氣。

Learning to Build Agentic Apps with Azure AI Foundry

Building agentic applications with Azure AI Foundry can feel like stepping into a new world for a solution architect. The promise is huge, an entire ecosystem for creating, deploying, and managing AI agents at enterprise scale, but it requires rethinking how we design architectures, plan adoption, and integrate security and governance. Coming from a background of traditional solution design, I quickly realized that approaching this space with the right framework makes all the difference.

I began with Microsoft’s Cloud Adoption Framework, which breaks down the journey into familiar stages. Defining the strategy helped me clarify why the business wanted to adopt agentic AI in the first place and what value we expected. Planning translated those motivations into actionable steps, and preparing the environment with Azure landing zones gave me confidence that the foundations were solid. Adoption meant actually building and deploying workloads, and the final piece, securing them, was a reminder that AI systems must follow the same rigorous governance standards as any enterprise platform.

The next learning curve was understanding AI landing zones. These act as the enterprise-scale foundation for AI adoption and can be deployed with or without a broader platform landing zone. With a platform landing zone, services like networking and identity are centralized, offering scalability and compliance. Without one, you can start faster, but consistency suffers. As I came to see it, landing zones are the equivalent of a data center for AI agents, and they form the baseline that everything else plugs into.

Once the infrastructure was clear, I had to choose how to actually build agents. Azure AI Foundry makes it possible to experiment in multiple ways: low code or no code tools for fast prototyping, and pro code environments with VS Code extensions, REST APIs, or Semantic Kernel SDKs for full customization. At first I leaned on low code tools to get hands-on experience, then gradually moved into pro code scenarios as integration needs and complexity grew. The key lesson was to start simple and deepen over time, carefully selecting models and balancing cost versus performance while deciding which tools the agent should integrate with, such as Azure AI Search, Bing grounding, or Logic Apps.

Another critical design decision was whether to rely on single agents or adopt multi-agent systems. Single agents are predictable and easier to debug, making them a good starting point. Multi-agent setups, however, shine in dynamic or decomposable workloads where specialized agents collaborate, such as combining HR, IT, and compliance agents for employee onboarding. Semantic Kernel provides the orchestration layer for this coordination, allowing workflows to scale as complexity grows. The approach that worked for me was to start with single agents and only move to multi-agent orchestration once the use cases demanded it.

One of the biggest mindset shifts was recognizing that observability and evaluation are not optional. Unlike traditional apps where metrics are straightforward, agents can feel like black boxes unless you design for visibility. Azure AI Foundry’s traceability features log tool calls and agent interactions, while its evaluation metrics check groundedness, fluency, and relevance. Combined with AI safety tooling, these capabilities help ensure outputs remain safe, reliable, and aligned with organizational goals. For me, it was the equivalent of application performance monitoring in conventional systems: without visibility, improvement is impossible.

Of course, none of this matters if the system isn’t secure. Foundry layers governance and security controls across the stack. Managed identities and login protect users, while prompt and content filters ensure responsible AI practices. Virtual networks, NSGs, and VPN gateways provide network security, and Defender for Cloud adds threat protection. Purview further enhances data governance and compliance. I realized that while agents may feel like futuristic AI entities, architecturally they must be treated as microservices that adhere to the same enterprise-grade security principles as any other system.

Looking back on my early steps with Azure AI Foundry, several lessons stand out. Choose your building approach based on the maturity of your use case, whether no code, low code, or pro code. Pick models and tools carefully, weighing cost against performance. Start small with single agents, and scale into multi-agent orchestration when the complexity justifies it. Bake in observability, evaluation, and responsible AI practices from day one. And finally, leverage AI landing zones for enterprise-ready deployments that bring security, scalability, and governance to the forefront.

For me as a solution architect, Azure AI Foundry has become more than just a platform for deploying language models. It is a bridge between experimentation and enterprise readiness, providing the frameworks, tools, and safeguards needed to build agentic applications responsibly. The journey can feel daunting at first, but with a structured approach and focus on architectural principles, agentic AI quickly becomes less of a mystery and more of the next natural step in modern system design.

學習使用 Azure AI Foundry 建構 Agentic 應用程式

對一位解決方案架構師來說，使用 Azure AI Foundry 來建構 agentic 應用程式，就像踏入一個全新的世界。這是一個龐大的承諾——一整套可用於建立、部署與管理 AI 代理的生態系統，能在企業規模下運作，但同時也需要我們重新思考如何設計架構、規劃採用方式，以及整合安全與治理。來自傳統解決方案設計背景的我，很快就發現，用正確的框架來面對這個領域會帶來完全不同的效果。

我從微軟的雲端採用框架開始，它將這段旅程拆解為熟悉的階段。首先是定義策略，幫助我釐清企業為何要導入 agentic AI，以及我們期望獲得的價值。接著是規劃，將這些動機轉化為可執行的步驟；準備環境則是利用 Azure 登陸區，讓我對基礎建設有了信心。當進入採用階段時，代表真正開始建構與部署工作負載；最後的安全階段提醒我，AI 系統必須遵循與任何企業平台相同的嚴格治理標準。

接下來的學習曲線是理解 AI 登陸區。它們是 AI 採用在企業規模下的基礎，可以選擇有或沒有更廣泛的平台登陸區來部署。有了平台登陸區，像是網路與身分服務會集中化，能帶來可擴展性與合規性；沒有的話，雖然能更快開始，但一致性會受到影響。對我而言，登陸區就像是 AI 代理的資料中心，成為其他一切元件的基礎。

當基礎設施清楚之後，我需要決定實際如何建構代理。Azure AI Foundry 提供了多種嘗試方式：低程式碼或免程式碼工具可以快速原型設計，而專業程式碼環境則透過 VS Code 擴充套件、REST API 或 Semantic Kernel SDK 提供完全自訂的能力。一開始我依靠低程式碼工具來獲得實作經驗，隨著整合需求和複雜度提升，逐漸轉向專業程式碼的場景。最重要的教訓是從簡單開始，隨時間加深，並謹慎挑選模型、在成本與效能之間取得平衡，同時決定代理需要整合的工具，例如 Azure AI Search、Bing grounding 或 Logic Apps。

另一個關鍵的設計抉擇是要依賴單一代理，還是採用多代理系統。單一代理可預測且容易除錯，是很好的起點。然而在動態或可分解的工作負載中，多代理更能發揮優勢，不同專業的代理可以協作，例如在人員入職過程中結合 HR、IT 與合規代理。Semantic Kernel 提供了這種協作所需的協調層，讓工作流程能隨著複雜度增加而擴展。對我來說，最有效的方法是先從單一代理開始，只有在需求出現時才轉向多代理協調。

其中一個最大的思維轉變，是意識到可觀測性與評估並非選項，而是必須。不同於傳統應用程式的指標相對直觀，若沒有設計可視性，代理就像黑盒子。Azure AI Foundry 的追蹤功能能記錄工具呼叫與代理互動，其評估指標則檢查回應的依據性、流暢度與相關性。結合 AI 安全工具，這些能力能確保輸出安全、可靠，並符合組織目標。對我來說，這就像傳統系統的應用程式效能監控：缺乏可見性，就不可能改善。

當然，如果系統不安全，其他一切都沒有意義。Foundry 在整個堆疊中加入治理與安全控制。受控身分與登入保護使用者，提示與內容篩選確保負責任的 AI 實踐。虛擬網路、NSG 與 VPN 閘道提供網路安全，而 Defender for Cloud 則新增威脅防護。Purview 進一步強化資料治理與合規性。我意識到，雖然代理看似未來感十足的 AI 實體，但在架構上，它們必須被視為微服務，遵循任何其他企業系統相同的安全原則。

回顧我在 Azure AI Foundry 的早期探索，有幾個教訓特別清晰。根據使用案例的成熟度來選擇建構方式，無論是免程式碼、低程式碼，或專業程式碼。謹慎挑選模型與工具，仔細衡量成本與效能。從小規模的單一代理開始，當複雜度增加時再擴展到多代理協調。從第一天起就將可觀測性、評估與負責任 AI 實踐納入基礎。最後，善用 AI 登陸區來進行企業級部署，確保安全性、可擴展性與治理。

對我而言，Azure AI Foundry 已不僅僅是部署語言模型的平台，而是連結實驗與企業就緒之間的橋樑。它提供了必要的框架、工具與防護措施，讓我們能以負責任的方式建構 agentic 應用程式。這段旅程起初可能令人望而生畏，但只要採用有結構的方法並專注於架構原則，agentic AI 很快就不再神祕，而是邁向現代系統設計的下一個自然步驟。

How to Develop Breakthrough Core Banking Products and Services

Banks must develop major innovations to prosper, but they don’t know how to. Many still try to build them with an old producer model. In that model, vendors publish roadmaps, banks write long requirement documents, and system integrators deliver projects after months of work. Academic research and field practice point to a different path. Important innovation often come from users. These users share and improve ideas in communities. Breakthroughs scale when a platform gives them simple tools that turn designs into live products without friction. In banking, many “new” services begin as user workarounds before they become official. The next growth line is often being tested by customers and frontline teams already.

Core-banking innovation should shift from vendor-led requirements to bank- and user-led creation. The shift rests on three building blocks. First, find lead users whose needs are ahead of the market. Second, build communities so good solutions spread and improve. Third, embed toolkits inside the core banking platform so teams can design, test, and ship safely. Done well, idea to launch time drops from months to days, while first time right releases increase.

Lead users feel the pain first and have strong reasons to solve it. In core banking, these may be cross border SMEs juggling FX and instant refunds, collections teams that live in exception queues, and treasury teams reconciling real time flows. Sitting with these users reveals spreadsheets, excel macros, and policy rules that already express the product logic the bank will need. Treat these artifacts as a map of the real solution space. Rates, tiers, holiday calendars, schedules, tolerance windows, dispute timers, and end of day modes become a shared vocabulary for product design.

Communities turn isolated fixes into reusable patterns. Instead of long documents passed between teams, share recipes that others can copy and adapt. A clear path for dormancy and reactivation with notices, timers, and fee reversal rules. An instant refund path that handles partial reversals and time limits. An end of day mode with guardrails for delayed posting. In practice, once a strong recipe lands, local variants appear quickly across markets, and duplication drops.

Toolkits make the shift repeatable. Start with a user friendly product language and libraries. Teams configures rates, tiers, holiday calendars, schedules, posting order, and dispute timers in one place. Add a simulation ledger that replays real events from the past and compares balances, fees, and accounting between a baseline product and a new variant. Finish with one click translation from the product language to production artifacts, with strict checks on schemas, limits, liquidity, and audit. This turns intent into predictable behavior and gives risk and finance evidence they trust.

Governance becomes stronger by moving controls earlier and automating them. The core banking platform enforces accounting balance, limits, liquidity buffers, and rule checks at run time. The simulation verifies business results on real traffic before release. The pipeline signs and versions every change and links it to approvals and tests. Separation of duties is the default. Developers propose configurations, risk and finance approve, platform teams deploy, and a successful simulation replay is a hard gate. Controls run early and often with clear artifacts, which reduces surprises late in the process.

Measurement keeps the system honest. Track time to ledger from approved idea to ready configuration. Track iterations per week to show learning speed. Track first time right rate to capture clean releases without rollback. Track accounting differences between baseline and variant on real traffic. Add engineering signals such as build success rate and change failure rate. When these indicators improve together, the shift is working.

This paradigm shift changes the operating model. Less time goes into static product specifications. More effort goes into shaping the solution space, sharing patterns, and enforcing guardrails in code. The core banking becomes a programmable platform. The bank becomes a learning community that improves by building and testing, not by office politics. Once interest logic, holiday calendars, and other core behaviors are modular, and once the simulation catches edge cases before regulators do, teams stop fearing product change. That confidence is the real unlock. It turns innovation into steady, reliable breakthroughs driven by users and scaled by the bank.

如何開發突破性的核心銀行產品與服務

銀行必須持續推出重大創新才能成長，但許多機構並不清楚該如何著手。許多銀行仍停留在傳統「生產者（Producer）」模式：供應商發布產品路線圖，銀行撰寫冗長的需求文件，系統整合商歷經數月才交付專案。學術研究與實務經驗指出，真正有效的道路不同：重要的創新往往來自「使用者」。這些使用者在社群中分享並改進點子；當平台提供簡單且好用的工具組，便能把設計無阻力地轉成可上線的產品。在銀行業，許多所謂的「新」服務其實先源於使用者的權宜作法，之後才被正式產品化。也就是說，下一條成長曲線往往已在客戶與前線團隊之間悄悄驗證。

核心銀行的創新應從「供應商主導需求」轉向「銀行與使用者共同創造」。這個轉變建立在三個基石之上。第一，找到「領先使用者」，也就是需求走在市場前面的客群。第二，建立社群，讓優秀解法可以擴散並持續精進。第三，將工具組嵌入核心銀行平台，使各團隊得以安全地設計、測試與發佈。若執行得當，從構想到上線的時間可由數月縮短到數天，而且「一次到位」的發佈比例將明顯提高。

領先使用者最先承受痛點，也最有動機去解決問題。在核心銀行情境中，這類使用者可能是同時處理外匯與即時退款的跨境中小企業、長期處理例外案件的催收團隊，或需即時對帳的財資團隊。貼身觀察這些使用者的日常，往往能看到反映產品邏輯的試算表、Excel 巨集與政策規則。把這些產物視為「真實解決空間」的地圖：利率、分層、假期行事曆、排程、容差窗口、爭議計時器與日終模式，會成為產品設計的共同語彙。

社群能把零散的修補轉化為可重用的模式。與其在團隊間傳遞冗長文件，不如共享可複製、可調整的「配方（recipe）」。例如：包含通知、計時與費用回沖規則的休眠／再啟用流程；支援部分沖銷與時限控制的即時退款流程；具備延後入帳護欄的日終模式。實務上，只要有強韌的配方落地，各市場的在地化變體便會快速出現，重工也隨之下降。

工具組讓轉型可複製、可擴張。首先是「使用者友善」的產品語言與元件庫，讓團隊能在同一處定義利率、分層、假期行事曆、排程、入帳順序與爭議計時器。再來是「模擬帳本」，可重播歷史事件，將基準產品與新變體在餘額、費用與會計處理上逐一比對。最後是「一鍵轉譯」：把產品語言編譯為可上線的產物，並嚴格檢查資料結構、限額、流動性與稽核要求。這能把設計意圖轉成可預期的系統行為，並提供風險與財務部門可信賴的證據。

將控制前移並自動化，可讓治理更強健。核心銀行平台需在執行期強制檢查會計平衡、限額、流動性緩衝與規則；模擬環境在發佈前以真實流量驗證業務結果；變更管線對每一項調整進行簽署與版本管理，並串連核准與測試證據。權責分離成為預設：開發人員提出設定，風險與財務核准，平台團隊部署，而通過模擬重播則是必備關卡。控制因此能更早、更頻繁、且以清晰工件運作，減少後期意外。

度量指標能讓系統保持誠實。追蹤「上帳時間」——從核准的點子到可部署設定的前置期；追蹤每週迭代次數以衡量學習速度；追蹤「一次到位」發佈比率；以真實流量比對基準與變體的會計差異；並加入工程訊號，如建置成功率與變更失敗率。當這些指標同步改善，代表轉型正在奏效。

這場典範轉移也重塑了營運模式。與其把時間投入在靜態產品規格，不如把力氣放在界定解決空間、共享模式，並把護欄固化在程式碼中。核心銀行成為可程式化的平台；銀行成為以「構建與測試」驅動學習的社群，而非被辦公室政治牽動的組織。當利息邏輯、假期行事曆等核心行為模組化，且模擬能在監管機關之前攔截邊界情境，團隊便不再畏懼產品變更。這份信心正是關鍵啟動器——讓創新從口號變為由使用者驅動、由銀行擴大的穩健突破。