Dcard GNTC Decoded: Taiwan's First Agent-Native Company and the Playbook You Can Copy

In May 2026, Dcard announced GNTC, an enterprise AI agent business. CEO Yu-Chin Lin spent a year visiting every department — finance, advertising, product, marketing — and cut his advertising team's workflow time by more than 80%. Then he packaged that methodology into a service he could sell to other companies. The real significance isn't "another AI startup." It's that this is the first time a Taiwanese tech company has publicly deconstructed its internal agent-native transformation playbook. This article skips the news recap and asks the questions that matter to practitioners: Can SMEs replicate this? Can one person do it? Where does it most often go wrong?

TL;DR

• Dcard's advertising department cut its overall workflow processing time by over 80% after building an AI Center with EntryDesk — but this is one department's number, not a company-wide average
• What GNTC is really selling is the FDE (Forward Deployed Engineer) methodology; EntryDesk is the vessel, but the consulting capability is the IP
• The Discover-Build-Scale framework scales down to a single person: start with one time-consuming repetitive process, no consultant needed
• According to xtract.io industry analysis, 40% of companies attempted agentic AI in 2025, but only 11% reached production; the biggest trap is agent washing
• EntryDesk has a free tier for self-service onboarding, but expect 2 to 4 hours of setup time to get the first scenario running

Why Did Dcard Build GNTC? The Strategy Behind the B2B SaaS Pivot

Many people's first reaction to GNTC is: "Dcard hopping on the AI trend." But looking at Yu-Chin Lin's interviews with bnext and meet.bnext, the underlying logic is far more grounded.

Dcard's advertising platform is heavily tied to Taiwan's young-adult demographic, and its traffic monetization has a natural ceiling. Starting in March 2025, Lin decided to personally take on the role of company-wide FDE, visiting each department to find workflows that could be transformed with AI agents. This wasn't a management directive handed down from above — it was the CEO doing his own fieldwork.

A year later, the advertising, finance, and product departments each had concrete results. The know-how that was validated and refined internally became the service that could be sold externally. GNTC is the productization of that process, not a pivot built from scratch.

One important caveat: GNTC only went public in May 2026, and Lin himself is serving as its first salesperson. Based on publicly available and verified information, Dcard itself is the most complete client case study. External enterprise results are still accumulating. That's a reasonable thing to keep in mind when evaluating the service.

As for whether GNTC divides Lin's attention away from Dcard's core business: according to the bnext interview, he uses a "75%+75%" framework. GNTC's tools continue to be used actively inside Dcard, making the two complementary rather than competing.

Advertising Workflows Cut by Over 80% — What Actually Changed?

"Over 80%" is the most cited figure from GNTC's media coverage. But its scope is narrower than most readers assume.

According to INSIDE and bnext's reporting, the number refers to the reduction in overall advertising workflow processing time after Dcard's ad team built an AI Center using EntryDesk. The transformation happened across three layers:

The first layer was data query automation. Ad team members used to manually pull reports from multiple platforms. They can now query integrated data with natural language, bypassing the BI team and manual spreadsheet work.

The second layer was proposal acceleration. Data specification documents that used to take days of multi-person collaboration now produce a reviewable draft in under 10 minutes. Overall proposal time dropped from about a week to a few days.

The third layer was creative asset automation. Repetitive ad upload operations were handed off to agents, reducing manual clicking.

The finance department's results represent a different dimension: after their month-end workflow was restructured, overtime decreased and error detection improved. That's a quality and hours-worked metric, not a speed metric. You can't apply "80% reduction" as a universal benchmark across departments.

For readers in other industries, the most transferable logic is: if your workflow involves "pulling from multiple data sources and consolidating into a single report," that type of task is closest to what the ad team automated. Copywriting, financial reporting, and marketing analytics all share a similar structure.

Agent-Native vs. Buying a Stack of AI Tools — What's the Real Difference?

Understanding GNTC's positioning requires separating three distinct levels of AI adoption.

Tool adoption is the most basic: you add ChatGPT or Notion AI to your work, but the underlying workflow steps stay the same. Individual tasks get AI assistance, but the process structure doesn't change.

Workflow redesign is the middle layer: you restructure the steps themselves, using AI to handle certain portions, with humans doing final review and connecting the pieces.

Agent-native is the top layer: agents become the primary executors of the workflow. Humans supervise and handle exceptions. The design logic shifts from "what does the person do" to "what does the agent do, and when does the person step in."

This is also why GNTC is different from traditional RPA implementations. RPA (robotic process automation) runs fixed scripts and handles structured, predictable tasks, but breaks down when exceptions occur. AI agents can reason, handle unstructured data, and adapt steps during execution. According to xtract.io industry analysis, roughly 80% of enterprise data is unstructured — that's exactly where RPA struggles. The 2026 best practice is a hybrid architecture: RPA handles fixed processes, AI agents handle the exception-reasoning layer.

On the common comparison question: EntryDesk vs. n8n/Make, or vs. Claude Projects plus Notion AI.

EntryDesk's key differentiator is its enterprise governance layer: built-in permissions management (who can access which agent), audit trails (a complete log of every execution), ISO 27001 certification, and private deployment support. These features matter for regulated industries like finance, healthcare, and government. For individual users, they may be over-engineered.

n8n is more flexible and highly customizable, but requires technical ability. Claude Projects works well for personal knowledge management and conversational workflows, but lacks cross-tool execution governance and a complete audit log.

If you're working solo, Claude Projects plus Zapier or Make is likely sufficient. If you're an enterprise needing multiple departments to share agents with traceable execution records, that's where EntryDesk's positioning actually makes sense.

For a deeper look at how to choose the right AI agent tooling, see the AI Agent Beginner's Guide.

What Is FDE Thinking? Can One Person Use It?

FDE (Forward Deployed Engineer) is a common role in Silicon Valley startups — an engineer embedded inside a client organization who uses engineering thinking to solve business problems. Lin Yu-Chin transplanted this role internally at Dcard: he personally acted as the company's FDE, visiting each department to find workflows that could be converted into agents.

One quote from the bnext interview captures the core logic well: "Capturing data is just asking a question in natural language and turning it into a piece of code." Expand that principle into a framework and you get Discover-Build-Scale:

Discover: Describe your work requirements in natural language. Find the fixed pattern of "what do I need as input, what do I want as output." No engineering background needed — just the ability to clearly describe what you do.

Build: Convert that fixed pattern into a repeatable agent, connecting the tools and data sources you need.

Scale: Let more people (or more processes) reuse the agent, accumulate usage data, and continuously improve.

This framework doesn't depend on team size. The difference between the enterprise version and the individual version is only at the execution layer:

Enterprise path: There's a dedicated FDE role (or a designated person serving that function), with Discover happening across multiple departments simultaneously, followed by FDE-led Build and governance architecture.

Individual path: You are simultaneously the FDE and the executor. Start with one time-consuming repetitive process. You don't need a 20-person team or a consultant.

When I worked through the FDE thinking myself, the most useful exercise was forcing myself to write down the inputs and outputs of one workflow. Take writing a weekly summary report: inputs are "Slack message history + completed GitHub PRs + calendar events," output is "a 200-word weekly update draft." Once you write it out that clearly, you can judge whether it's agent-able.

Three-step individual FDE starter:

1. List tasks you do more than twice per week
2. Write one sentence describing the inputs and output format for each
3. Build the first agent in EntryDesk's free tier or Claude Projects to test

For more on how Claude fits into agent architectures, see the Claude Managed Agents Taiwan Guide.

Starter Guide: What Does EntryDesk Free Tier Actually Cover? What's the Real Setup Cost?

EntryDesk's official site confirms a free tier with 30+ tool integrations (Slack, Gmail, Salesforce, BigQuery, Notion, Jira, and others), no-code agent building, ISO 27001 certification, and support for Claude, GPT-4, and Gemini.

That sounds comprehensive. But a few real costs deserve honest disclosure.

First, setup time is not zero. According to bnext's 365-day implementation record, one of Dcard's actual early adoption challenges was that "a 20-minute installation process frustrated employees." That's at a tech company with engineering support. For users without any technical background, getting the first scenario running will likely take 2 to 4 hours, including understanding the tool logic, configuring integrations, and testing the workflow.

Second, the CEO-as-salesperson signal means enterprise consulting still requires manual contact. GNTC is open for business, but it isn't a self-serve purchase flow — you need to reach their sales team. For SMEs with limited budgets, the recommendation is to get one concrete scenario working on the free tier first, then engage consulting with specific questions. The conversation will be significantly more productive.

Third, vendor lock-in risk needs to be tested yourself. The verified public information available doesn't include a clear statement of EntryDesk's data export policies. That's a real open question. During your trial period, actively test: can you export the agent logic, integration configurations, and execution history to another platform?

Recommended first scenario: Route key Gmail subject-line summaries to a designated Slack channel automatically. This is verifiable within 30 minutes and lets your team see a concrete outcome while testing whether integrations work smoothly.

Risk Disclosure: Three Traps That Cause Agentic Transformations to Fail

According to xtract.io industry analysis (2026 Q1, Tier 3 industry observation), 40% of companies attempted agentic AI in 2025, but only 11% reached production. A failure rate approaching 90% comes from three traps that appear repeatedly.

Trap 1: Legacy Friction

Most existing systems and tools were designed for humans clicking through interfaces, not for agents calling APIs. When you try to hand a workflow off to an AI agent and discover that the core tool has no API, or requires VPN access, or needs manual login — that's legacy friction. Part of the reason Dcard could transition relatively smoothly is that, as a tech company, its tooling stack is already relatively modern. Legacy enterprises face a much larger version of this obstacle.

Fix: During the Discover phase, confirm that every tool in the workflow you want to automate has a callable API or webhook before you commit.

Trap 2: Agent Washing

This is the cognitive trap most worth watching for in 2026: rebranding a chatbot that accepts prompts as an "AI agent." The name sounds impressive, but every step still requires human confirmation, and there's no genuine cross-tool autonomous execution. When evaluating any tool, ask three questions: (1) Can it call external tools? (2) Can it complete a task of 3 or more steps without human intervention? (3) Does it have an audit trail of every execution? All three need to be "yes" to qualify as a real agent.

Trap 3: Data Bottleneck

An agent's capability ceiling is determined by data quality. If your data is scattered across multiple systems, inconsistently formatted, and not searchable, there's very little an agent can do. A significant part of why Dcard's advertising team achieved over 80% reduction is that they already had structured advertising data and clearly defined KPIs. The data was agent-ready before the agent was built.

Fix: Before adopting any agentic tool, audit your data situation. Where does your core business data live? Is there an API to query it? Is the format consistent?

Finally, to address a fair question: does GNTC have documented results from external Taiwanese companies? Based on verified public information available as of this writing, Dcard itself is the most complete and well-documented case study. GNTC only went public in May 2026, and external client results are still being accumulated. That's not a reason to dismiss it — it's information to hold while you evaluate, rather than using Dcard's internal metrics as a proxy for what external clients will achieve.

Conclusion: Agent-Native Isn't a Destination — It's a Reset of How You Work

GNTC's genuine contribution isn't another AI platform. It's the first time a Taiwanese company has publicly deconstructed a complete agent-native transformation path, giving SMEs and individual workers a concrete reference point.

The most valuable thing to take from this isn't "use EntryDesk." It's the FDE thinking itself: start with Discover (describe your workflow clearly in terms of inputs and outputs), move to Build (start with the smallest possible agent), and eventually Scale (let more processes reuse the same logic). This framework doesn't require corporate resources. A single person can start today.

Whether GNTC and EntryDesk are the right tools for you depends on your workflow characteristics, how modern your existing tooling is, and your compliance requirements. You don't need to decide right now.

Spend 15 minutes today listing tasks you repeat more than twice per week. That's your Discover list. Start with the top item. Describe what it needs as input and what it produces as output. Then ask: could an agent do this?

If your answer is "probably," you've already done the hardest part.