You Won’t Lose Your Job to AI. You’ll Lose It to Someone Who Uses AI.

Insights from the Scaling Business Summit 2026, Ho Chi Minh City.

Dru Nguyen, Managing Partner of Tictag, opened with a show of hands. More than half the room was already using AI in their businesses. His session did not debate whether AI matters. It addressed something more useful: why AI tools fail in local markets, and what that means for companies building or deploying AI in Southeast Asia.

1. AI Is Not the Threat. Inaction Is.

At the World Economic Forum, Jensen Huang used a radiologist as his example. Scanning, flagging, documenting, AI handles the repetitive layer. The human focuses on what cannot be automated: judgment, experience, patient care.

Dru's version was blunter. "You are not going to lose your job to AI, but you are going to lose your job to someone who uses AI." This is not a future warning. It is a current hiring and training decision.

Lesson 1: AI adoption is no longer a competitive advantage. It is a baseline. The question is how fast your team builds fluency with it.

2. Every AI Model Runs on a Data Pipeline Most People Never See

Before any AI model can identify, classify, or predict, it needs raw data, annotation, and quality control. Raw data is unprocessed video, images, or text. Annotation gives it context and labels. Quality control verifies accuracy before it trains the model.

Dru illustrated this with a smart trash can that sorts waste into compost, disposables, and cans. The hardware is simple. The intelligence comes from thousands of verified, labeled data points. Every user correction improves the model.

"Without accuracy, the data is trash. You can't use it."

Lesson 2: Before choosing any AI tool, ask what data it was trained on and how that data was verified. Accuracy in the annotation layer determines everything downstream.

3. Most AI Was Not Built for Southeast Asia

The major LLMs and speech recognition tools most businesses use are predominantly trained on Western datasets. They perform well in English and internationally standardized contexts. They perform significantly less well for local languages, regional accents, and cultural nuance.

Dru's specific example: a model that cannot distinguish between a Vietnamese northern and southern accent has a fundamental accuracy problem for any speech-dependent application. "Even though it's Vietnamese, there's still different vocabulary words used." Western-trained models treat them as identical. They are not. This is what sovereign AI means in practice, countries developing training data that reflects their own language, culture, and context rather than borrowing someone else's.

Lesson 3: Western-trained AI has a measurable accuracy gap in Southeast Asian contexts. Local language and cultural applications need locally sourced training data.

4. The Education Gap Is the Workforce Gap

A Singapore business school student Dru described was required to use LLMs for every assignment and document each step. The output: a complete restaurant concept, menu, branding, content, positioning built and documented with AI. One student. One project.

Vietnam's school system is not there yet. The desire for AI education exists. The infrastructure does not. "If you're not allowed to use AI in school, how are you ever going to compete in the real world?" Graduates entering the market now were trained without it. Companies that build AI fluency internally are not waiting for the system to catch up, they are building a compounding advantage now.

Lesson 4: Vietnam's AI education gap is a workforce gap. Build internal AI fluency. Do not wait for the school system to deliver it.

The CEO Execution Playbook: What to Do Tomorrow

1. 1. Test your AI tools on local inputs. Run your speech or language tools against northern and southern Vietnamese accents. If accuracy drops, you have a training data problem.
2. 2. Ask vendors about training data. Before deploying any AI model locally, ask what it was trained on. Western-centric answers should inform your accuracy expectations.
3. 3. Map repetitive tasks in your team's week. Identify the five most time-consuming repeatable tasks. Test whether an LLM can handle any of them. Document what works and spread it.
4. 4. Build internal AI fluency now. Basic prompting, workflow integration, tool documentation. Do not wait for the education system.

→ Watch the full session on YouTube