---
title: "Why Is Silicon Valley Bragging About Agents, While the Most Profitable Enterprises Are Obsessed with Workflows?"
author: "Unique Research"
sourcePublication: "Unique Research Substack"
originalPublishedAt: "2026-06-17T12:02:31+00:00"
canonical: "https://ffcap.cn/en/research/src-20260617-01html"
source: "https://uniqueresearch.substack.com/p/src-20260617-01html"
language: "en"
---

# Why Is Silicon Valley Bragging About Agents, While the Most Profitable Enterprises Are Obsessed with Workflows?

_Original · Unique Research · 2026-06-17_

_Editor's note: The first-person report and its judgments belong to the original Chinese author (hosted by Delta Wu). This English rendition retains the panel narrative across hardware, dual-track SaaS, enterprise Agent OS, Infra-to-finance, and the counterintuitions of enterprise landing, plus the full verbatim Q&A. All panelists, companies, and figures are preserved. Founder statements and cited numbers are source attributions, not independently verified findings._

AI Industry Observation

AI commercialization has no standard answer, but some people have already made money

The real commercialization debrief of five female founders at the Singapore Agent Summit

A $200 voice recorder, within months, made over 100 million US dollars in revenue across the Asia-Pacific region. 50% of buyers are CEOs and Founders worldwide — the busiest people on earth, the ones with the most money and the least shortage of efficiency tools.

This is a set of numbers I heard at the Singapore Agent Summit. The person saying it was Megumi, with four other female founders seated behind her. An all-women panel is rare in the AI industry; rarer still is what they talked about — not who raised more or whose model is bigger, but how to survive in the real business world, and survive well.

Megumi introduced herself in sixteen characters: rooted in the Northern Yangtze River, educated in the Eastern Ocean, served in the Western Ocean, settled in the Southern Ocean.

On the same stage were Guan Dian of PatSnap, Lusha of Dify, Jennifer of Tezign, and Jolie of Emerging AI.

Five companies, five completely different paths, but all making money. Their own data prove one thing: AI commercialization has no standard answer. Those who truly make money are all doing the same thing: finding a real scenario, using the right product form, and solving a concrete problem.

Hardware as the Entry Point: Plaud's $200 Trust Business

Megumi said one thing that stuck with me: most human intelligence comes from conversation, not documents. Plaud isn't just making an AI voice recorder; it's helping AI find the entry point back into human life.

The core problem this entry point solves isn't technology, but trust.

Plaud's product is "always ready but not always on" — the device can be worn all day, but isn't taking notes all day. The user decides when to start recording, when to highlight, when to share the note with whom, and so on. This hardware's physical form, embedded into life naturally in a very simple way, has won word of mouth and trust from global users.

More interesting is the user profile. 50% of global users are CEOs or Founders, the busiest people in the world. A $200 device saves them hundreds to over a thousand hours.

In plain terms, users aren't buying AI technology; they're buying back saved time. Over 100 million US dollars in APAC revenue in months, a Singapore office from 0 to nearly 100 people, and 100+ openings globally — behind these numbers is a simple fact: when AI finds the right entry point, commercial returns follow on their own.

Building a New Building on a Familiar Foundation: Patsnap's Dual-Track Model

Guan Dian represented Patsnap. Eighteen years ago this company started a B2B SaaS business in the intellectual-property intelligence platform, accumulating years of data capability in the industry. After AI arrived, Patsnap chose to use its existing IP data capability as the base layer, overlay AI capability, and build its own model-layer and orchestration-layer capabilities. On the business side, AI transformation has already brought striking growth: AI-empowered and AI-native products are growing very fast, even lifting the traditional SaaS business.

Guan Dian shared a very pragmatic judgment: not every step needs a large model; traditional small models already suffice to solve many node problems. The current billing model runs two tracks in parallel — traditional annual SaaS fee plus AI Credit, with clients buying AI capability as needed.

They're also trying a new model called FDE — full-time digital employee. First they "dispatch" a "digital employee" to the client for free, work alongside them for a while, map the real pain points, then charge by outcome. Outcome-based, not selling a piece of software and letting the client figure it out.

This model essentially lowers the client's decision threshold. You don't pay first for software; I send someone in to work for you first, and we talk money after results.

Not tearing it down and rebuilding, but adding new floors on a familiar foundation. This "overlay" approach may suit most existing SaaS companies better than "disruption."

Agent Operating System: Tezign's GEA, an Agent System Built for Enterprises

Jennifer represented Muse AI (the overseas entity of Tezign). Their product is called GEA (Generative Enterprise Agent), an enterprise-grade Agent operating system — clients can call on demand from 450+ preset Agent Skills, build their own in the system, and integrate external Agents to form an agent network running inside the enterprise's own business processes.

Tezign's core advantage comes from years of DAM (digital asset management) accumulation, with deep understanding of clients' structured and unstructured data. On this basis they also self-developed a Creative Reasoning Model which, together with the Context System built from years of DAM accumulation, forms GEA's technical foundation — the Creative Reasoning Model drives the entire system's orchestration layer, dynamically coordinating 30+ base models.

But Jennifer also frankly spoke of the challenge. ASEAN isn't one market, but many markets — multiple language families, multiple cultures, huge regulatory differences. She gave the example of the healthcare industry, where each country's regulations are completely different; an Agent that works in Thailand may need its compliance logic redone for Indonesia.

Building a "platform" is far harder than building a "point tool," but the moat is deeper too. This is what Jennifer didn't say outright, but the whole room understood.

From Infra to Finance's Deep Waters: Technology Is Just the Entry Ticket

Jolie represented Emerging AI. The company is headquartered in Singapore and was global by design from Day 1. They started from AI Infra — GPU management, model serving — and now go deep into vertical industries.

Jolie shared an interesting turning point. Only two or three days after DeepSeek came out, a Middle East client proactively reached out: "can you make something for us?" The market was quickly educated, but after the education, the real challenge began — how to land AI into concrete business scenarios?

They ultimately chose the finance industry. The logic is direct: data-intensive, labor-intensive, high repetition, reliant on expert experience — where AI can bring the most value.

But finance is also one of the industries with the lowest error tolerance. Jolie says their solution is triple assurance: confidence in technical precision, airtight compliance and data confidentiality, and Human-in-the-loop deep in the decision process.

She also mentioned an easy-to-overlook detail — globalization isn't just a matter of technology and language, but of culture. To go deep into an industry, respect for local culture is the true barrier.

The "Counterintuitions" of Enterprise AI Landing

What enterprises buy isn't technology, but "controllability"

Lusha said one line I still remember: "RBAC and Compliance are an enterprise's lifeline. Controllable and visible are the test values for how long an Agent survives in the enterprise."

In plain terms: enterprises don't care whether you use GPT-4o or Claude 3.5; what they care about is — will employees using this leak data? Can it be audited? If something goes wrong, who's responsible?

Token cost is another underrated bomb. Lusha said something many people didn't understand: "every time you start an Agent, it's running content from scratch again." What does that mean? You don't know how many steps it will run this time, how many tools it calls, how many Tokens it consumes. To a consumer user that doesn't matter, but to an Enterprise, who pays for tens of thousands of extra dollars a month?

Consumer and Enterprise are completely different worlds.

Someone doing consumer products can ignore RBAC (role-based access control); someone doing enterprise AI who doesn't know this can't even get in the door. This isn't a technical threshold; it's a survival threshold.

The Real Craft of Globalization

The rules of enterprise AI landing are clear, but when you take these rules to different countries, you find another dimension of problems. Of the five panelists, Patsnap and Emerging AI both do globalization; after one round, everyone's lessons are more valuable than their experience.

Globalization's first step isn't translation, but becoming a "local"

Jolie told me a case that sent chills down my spine. They ran a Christmas sale in the Middle East and South America, using snowflake elements in the PPT. The local team urgently pulled the plug — they celebrate Christmas in summer; what do you mean by putting snowflakes? Cultural offense.

Jennifer's story cut deeper. Their team pushed healthcare marketing in ASEAN and was directly challenged by a client: "how well do you understand Singapore's regulations?" One sentence left the team speechless.

In plain terms, globalization's first step isn't translating the product, but first becoming a "local" in that culture.

The PHEST Framework: one more H than PEST

Megumi shared a framework she built over 20 years of cross-border business — PHEST. It's PEST (Political, Economic, Social, Technological) plus one H, History.

She gave two examples. The Philippines is almost 100% import-dependent; what does that mean? The slightest geopolitical tremor sends consumption decisions swinging violently. If you do the Philippine market without understanding this, you're groping an elephant.

An even more elegant example: Thailand is the only country in Southeast Asia never colonized. Why? Megumi says, understand this history and you understand Thais' business logic and consumer psychology — the independence in their bones and their attitude toward outsiders are both written in that history.

Understand a country's history and you'll understand its present.

Guan Dian's trust theory: the Japanese market values stability even more; you need to operate locally long term and understand the details of how enterprises run well enough to knock on the door.

Guan Dian's pit is representative. Early on going global they assumed finding a salesperson would suffice, then found it's nowhere near enough — you need someone with an Entrepreneur spirit who can build trust, build a network, and solve problems from zero to one on the ground.

A deeper point: the way trust is built differs completely across markets. The US market buys innovation — have a new technology and others will try it. The EU market buys compliance — did you pass GDPR? Where is the data stored? The Japanese market buys relationships — the first meeting isn't about business; talk for half a year first.

Being able to build Trust is the key to succeeding in that market.

Lusha's "dual track" and Singapore's window

Lusha's globalization playbook is interesting; she calls it "ground forces + air forces." The ground forces are communities — individual developers, individual users, word of mouth. The air forces are cloud vendors — AWS, Azure, GCP Marketplaces, global coverage the moment you list.

Breakthrough and Elevation

After all this talk, one feeling grew stronger and stronger.

The essence of AI commercialization isn't a model race, or whose technology is more advanced, but three things: deeply understand the scenario, build trust, and let the product form follow the need.

Looking back at these five companies, I realized they're all doing the same thing: translating AI capability into a product form the client can understand.

Plaud makes a hardware entry — the client doesn't know what an LLM is, but knows "press record, AI organizes my meeting notes." Patsnap does data overlay — the client needn't understand RAG, only needs to know "type in an idea, the system tells you whether a similar patent exists globally."

So what exactly is the right posture for AI commercialization? Three feelings, the kind you share with friends after a heart-to-heart —

Don't chase the trend. Everyone's talking about Agents now, but many of those actually making money are Workflows that look "not cool enough." Go find a vertical scenario where you understand the client's pain better than they do; technology is only the threshold, scenario understanding is the moat. To be frank, spending three months soaking in that industry beats reading 100 papers.

For product managers, Workflow isn't a lowly Agent; it's a more pragmatic way to land. Don't chase 100% automation; chase 99% certainty and controllability. Human-in-the-loop isn't that you failed at automation; it's that you made the right product decision.

For teams wanting to globalize, don't rush to translate the product UI; study local history and culture first. Find someone with an Entrepreneur spirit to lead the new market — this person needn't know technology best, but must know best how to make people trust you in that environment.

At the end of the roundtable, Guan Dian said a very down-to-earth line: "We've got the PPT ready, but the high heels aren't ready yet." Five women on stage, behind them five completely different AI commercialization paths.

I was thinking: the best Agent isn't the smartest Agent, but the one that best knows when to let a human step in. Globalization isn't translating a product into many languages, but translating your way of thinking into many cultures.

Megumi said: understand a country's history and you'll understand its present. Doing AI commercialization is the same — understand a scenario's depth, and you truly understand its need.

What do you think?

More Conversation Detail

Panelists: Jolie Li / 李卓 (Emerging AI COO); Jennifer Chen (Muse AI CEO); Megumi Yoshinaga (Plaud Head of APAC Marketing); Dian Guan / 关典 (Patsnap Co-Founder); Lusha Chen (Dify GM of APAC)

Host: Delta Wu (Unique Research Founder & CEO)

Delta Wu: First, please each briefly introduce your company and your role.

Jolie: Hello everyone, I'm Jolie from Emerging AI. At the start we focused more on building the whole AI Infra system; now we're in the third year. We've also moved more into vertical industries, like finance and smart manufacturing, which we now value more, going deep into their workflows to land scenarios.

Delta Wu: Wow, it perfectly fits the "Agent entering the workflow" I just mentioned.

Jennifer: Hello everyone, I'm Jennifer, CEO of Muse AI. Muse AI is Shanghai Tezign's overseas entity. Tezign and Muse AI are committed to helping multinational enterprises build multimodal-model workflows. Since we've always started from content, we now focus on end-user product innovation and marketing content operations.

Delta Wu: So it's essentially Tezign's overseas branding, but the business is fairly similar to domestic?

Jennifer: Yes, because we've always served MNCs, so now we've just gone from domestic MNCs to Regional MNCs.

Delta Wu: OK, looks like Singapore is a very important first stop going global. If you went to SuperAI, you'd find the sponsor WiFi password on the badge wristband is Muse AI.

Jennifer: Yes, the WiFi is ours.

Megumi: Good morning everyone, my name is Megumi, a Japanese-born Chinese. First, in one sentence: rooted in the Northern Yangtze River, educated in the Eastern Ocean, served in the Western Ocean, settled in the Southern Ocean. But the point is ultimately choosing Singapore — we can talk about that later.

The company I'm at now is Plaud. If you attended SuperAI over the past few days, you may know we named the whole Main Stage this year, one of the biggest sponsors. Many people said on social circles that Plaud actually competed past so many large-model companies to win the naming right. We're very proud to say we're one of the most profitable companies in AI hardware, so far. Based on data released yesterday morning by The Straits Times and Lianhe Zaobao, the Asia-Pacific region is expected to contribute about 100 million US dollars in revenue in 2026. This time last year we had no Singapore office and no employees, and this year we're already at nearly 100.

Delta Wu: Oh, so there are still open jobs?

Megumi: Yes, about 100+ opening roles, welcome to apply.

Guan Dian: Hello everyone, I'm Guan Dian, co-founder of Patsnap. Patsnap is headquartered right here in Singapore; from Day 1 it was founded by a few NUS alumni. We've always run global business from Singapore, a fairly typical B2B enterprise-service company. Clients are enterprise customers across about forty to fifty countries and different industries, mainly serving corporate R&D and IP departments. Typical ones like Huawei, DJI, or America's 3M, Dyson — any company investing heavily in R&D uses it. The product is a system called Innovation Intelligence, helping enterprises do R&D more efficiently.

Delta Wu: Right, and in Singapore these IP things can all be unlocked through your database.

Guan Dian: Yes, it's a very good position.

Delta Wu: Then let's talk about commercialization. Patsnap has lots of data and headquarters in Singapore. Over these AI-boom years, how have your monetization and product value changed?

Guan Dian: I think it's an exploratory process. Patsnap now has 18,000+ enterprise clients globally, many of them Fortune 500. Organizationally we're transforming into an AI-native organization, but commercially it's not a 180-degree shift. The traditional annual SaaS model still exists; for clients it's a steady, certain thing funded from R&D or IP budget.

And when we deliver results, not everything needs a large model. Actually many things are solved well enough by traditional small AI models we've used for years to handle node problems; no need to burn so many Tokens, so the annual-fee way persists.

On top of that, we've now fully introduced Credit-based pricing. Whether it's traditional products with AI features added that users call, they can pay extra Credit; and there's also a wave of purely AI-native products, where many detail tasks across the whole R&D workflow can be completed end-to-end by an Agent, also charged by credit. We're still exploring and about to try the now-popular FDE (full-time digital employee) model, going outcome-based directly into the enterprise. Because enterprise clients' R&D processes may have formed over decades, they can't fully adapt to our way, so we have to go in.

Delta Wu: After going in, first understand where the client's pain is and find the best step, because not every step needs AI. Now how do you distinguish which income increase is from AI, versus Token consumption that was already there? What changes do you see?

Guan Dian: Commercially we keep them separate. Traditional growth is relatively steady. I bracket the pure AI-Agent revenue separately; and the AI part added on traditional products can also be seen through Token consumption.

Delta Wu: So not every business has to rebuild its model around a large model. Overlaying some AI features on existing data and foundations is actually a good attempt, bringing 20%+ growth to the existing business. And enterprise clients now also become "tell and prove my ROI," so you still have to send people in.

Now over to Megumi. Plaud is distinctive in having both hardware sales and software subscriptions. Starting from such a smart-hardware product, tell us how you think about commercializing these AI features, including the relationship between subscription and hardware?

Megumi: Great question. Before talking this through, we need to understand the product: why are we so profitable? Many people's perception of Plaud is "the world's first and number one AI note taker." But actually we aren't making an AI voice recorder; we're helping AI find the entry point back into human life — that's also why we have a hardware.

What does that mean? Today's AI products mainly focus on working on documents. But have we considered that most human Intelligence actually comes from Conversation — information not captured in documents? So Plaud is like your personal chief of staff or AI assistant, helping you record, at any time, the easily forgotten intelligence of each day, then feeding it into a great workflow and optimizing it with AI. We're actually redefining the AI-Human interaction; that's why hardware is needed.

This product form leads to our business model: we have hardware revenue, about 200 dollars for 10,000 devices. But the company's strongest point is that our paid-user ratio is very high. Sharing a data point: 50% of Plaud's global users are CEOs or Founders, the busiest people in the world; to them 200 dollars, or each minute and hour, is worth more than that. Plaud on average saves CEOs hundreds or even over a thousand hours. A thousand hours times my time cost — totally worth it. This goes back to what panelists said: it's already an era that pursues ROI and requires you to prove your product's value; for something valuable, people are willing to pay.

Delta Wu: Save money or save time. So do you encourage users to wear it all day?

Megumi: This may be an easy misunderstanding. I wear it all day myself; it's always ready for you, but not always on. People may think wearing Plaud means it's recording in real time the whole time; actually it only turns on when needed.

Delta Wu: Understood; it can actually manage our lifetime context. Like today after a day of meetings, it can store all of today's context in memory and pull it out when we need it.

Now Jennifer. Tezign was a leading company from early AI to later Generative AI domestically; now doing this business overseas. The product value may be the same as domestic, but in commercializing overseas, is there anything different? For example, are the models different?

Jennifer: The models don't need to be different. At Tezign we've always been multi-agent, and we believe every future enterprise will always sit in a multi-agent workflow. Tezign's latest product is called GEA (General Enterprise Agents), an operating system; now there are 450+ Agents in this Marketplace that can work together very flexibly.

A very important foundation of this product is the context system underneath. Tezign has done Digital Asset Management for many years, so the understanding of clients' internal structured and unstructured data is very deep. This is to let Agents, through such a system, complete every task that needs executing in each department.

When we go overseas, building the system itself is meant to support this flexibility. I'm familiar with the Thailand-Singapore whole ASEAN system. Here there are multiple language families and cultures, and each company has huge teams working across so many different markets. So when helping clients build such workflows, the Agents we call may differ, and the way we source resources differs too.

On billing, Subscription is still something that makes clients feel secure. So module Subscription paired with Token usage is the way we'll now comprehensively establish.

Delta Wu: Based on the Asset Management / DAM system to manage clients' past digital assets, ad creatives, marketing materials, and building Agents on that basis is very convenient for many enterprises.

Jennifer: Yes. On the business side we also made a big shift: we accept that a client may not have only our DAM, but its own system, and Tezign's operating system can be built on various context systems. On this we also built our own model, called Creative Reasoning. It's essentially an Orchestration layer; we use it to build Agent-to-Agent workflows, making context calls more flexible.

Delta Wu: Understood; the future is multi-agent helping enterprises complete complex tasks. Jolie, can you introduce: you focus on several different industries to help clients; is the current business model and billing mainly FDE?

Jolie: Thank you, Mr. Wu. The other panelists' companies are fairly mature, through a period of development; we may be the younger startup in the room, but from day one we decided to be a global company.

For a 0-to-1 company doing B2B, the earliest stage must co-create with clients to get client cases. We were at Infra in AI's earliest days; actually we're an Infra operating system, from GPU management, Python, to model serving, plus Agent building, deployment, and inference management — an integrated Infra system. All AI capability must ultimately land in industries or scenarios. Back then in the Middle East, only two or three days after DeepSeek came out, gentlemen in robes asked me what DeepSeek could do for them. The market being educated is a very important trend.

Through constant exploration, we see finance as the easiest industry to land. Because finance is data-intensive with lots of human labor; AI more replaces high-repetition, high-error-rate, expert-experience-heavy positions.

Delta Wu: But finance has a very low error tolerance; if one number is off it's a huge risk. How do you solve AI hallucination and uncertainty?

Jolie: We indeed connect to finance across research, trading, and every step. At each step, from our data, to AI generation of factors and strategies, to the final trading decision: one, we're technically very confident in our precision and accuracy; two, we let compliance and data confidentiality handle more processing. At the same time, we let the "Human in the loop" step go deep into the whole decision process. We let clients truly go deep into the decision step early on, gradually building trust in the product.

Delta Wu: Next let's talk globalization. Please each discuss: how do we do globalization? How choose markets? Entering the local, what factors must we consider? If it was originally a Chinese version, how does it become a global version (including product experience, technical architecture, compliance culture, etc.)? Please share pitfalls too.

Jolie: I'll start. I have long software going-global experience doing global markets. Our HQ is now in Singapore.

Globalization's first step isn't just localizing the product or translating it to English or Japanese; second is, I think, full respect for local culture. A case: at another large going-global company I worked at, during annual Holiday Season or Black Friday sales, if in the Middle East or South America we ran a Christmas sale using lots of snowflakes or winter elements, local users would feel "you didn't respect my culture," because we're in completely opposite seasons. It's a very small point, so creatives must be personalized. Also crucial is whether software and hardware data compliance already runs locally. Now, our highest-revenue countries are Japan and the US.

Jennifer: Before Tezign I always did advertising creative. From that advertising-creative level, ASEAN is very challenging, because you face not only language but each country's customs. After joining Tezign I love bringing China's content-matrix and AI Avatar playbook to ASEAN.

But yesterday discussing a new marketing strategy with a client, I was challenged: how well do we understand Singapore's strict rules in healthcare regulation? I think this is what we must pay close attention to in multi-country go-to-market. Regulation spans many layers, and applied to different industries (healthcare, finance, FMCG) it's very different. This is what we'll watch especially in recent discussions.

Megumi: I can share my own framework every time I go global into a new country or region: on the common PEST (Political, Economical, Social, Technology) I add one word, H (History). We use PHEST to analyze risk, opportunity, and priority.

An economic and political example: the Philippines is a very important Plaud market in APAC. But affected by international conflict, the Philippines as a country almost 100% import-dependent takes a very big economic impact, directly affecting everyone's purchasing power, which lengthens everyone's decision-making cycle when buying new products.

Then History: globalization isn't the first time it's happened; hundreds of years ago everyone lived through the colonial era. If you carefully study Southeast Asian countries' history — for example Thailand is the only country never colonized by any country — why? These insights lead to a very good local understanding.

Guan Dian: On this question I think from two angles, one the pit, one the experience.

The pit: often entering a new country, we used to assume finding someone who can sell (since we're Enterprise) suffices. Actually it's not; you need an Entrepreneur-spirit person. Opening a new market isn't just translating and localizing a relatively mature product to sell to clients, but a second startup from scratch. For example in a market like Japan, the very top Japanese talent won't come work at a foreign startup. How do you attract good talent? You really rely on an Entrepreneur spirit to talk, to sell the big picture, and make them believe. Either find such a person to lead, or go yourself.

The experience: it's less an ability than the all-round ability to build Trust with clients in any new market. The key points for building trust differ across markets. For example in the US, Americans love innovative things; give them very new Technology or Visualization and they get excited and willing to try. But in the EU they watch Compliance and Regulation closely; GDPR came out earliest in the EU, so we were also the first outside China and the US to build a local server in the EU. As for Japan, the whole Enterprise Sale has a very long value chain, trust formed over decades. Being able to build Trust in a way that fits the local conditions is the key to succeeding in this market.

Delta Wu: Thank you very much to these five panelists! Although we all call it AI, we're actually doing different things, each with its own best practice — whether starting from Infra, combining software and hardware, overlaying Generative AI, or going deep into enterprise workflows.

---

Original publication: https://uniqueresearch.substack.com/p/src-20260617-01html
On-site reading page: https://ffcap.cn/en/research/src-20260617-01html