An Afternoon with Vertex Ventures SE Asia and India and Google Cloud: When Your Customer Isn't Human: Founders on the Agentic Shift

The conversation around AI has moved on. Generative AI captured attention. Agents are capturing workflows.

At Vertex Ventures SEA & India, in partnership with Google Cloud, we brought together a room of founders and operators to get into what the agentic shift actually looks like on the ground. Not the theory. The real decisions being made right now inside companies that are rebuilding around agents.

Moderated by our General Partner Genping Liu, the panel featured Jeffrey Tiong, Founder and CEO of Patsnap, Sam Lin, COO of dtcpay, and Yuan Liu, Solution Lead for GenAI at Google Cloud.

Here is what came out of the conversation.

The productivity shift is real, but the framing is wrong

Most discussions about AI productivity focus on doing the same things faster. What the panellists described was different. Teams are now taking on projects they would never have attempted before.

One example from the room: building a fully functional educational app for a seven-year-old, covering math and English, in thirty minutes. The same project would have taken months of traditional development. The point is not the time saved. It is that the project would simply not have existed without agents.

This is the distinction that matters. Efficiency gains are incremental. The agentic shift is enabling work that was previously out of reach.

The cultural change is harder than the technical one

The biggest organisational shift is not the tooling. It is the mindset.

The companies moving fastest are building a culture where the default is to ask AI first, and only escalate to humans when AI cannot answer. That sounds simple. In practice, it requires actively dismantling years of knowledge hoarding, where expertise was a form of job security.

Some companies in the room have gone as far as giving every employee AI credits, up to $2,000, specifically to force adoption. The goal is not to make people use AI. It is to make not using AI feel like leaving money on the table.

A two-tier talent strategy is also emerging. Domain experts are being asked to distill their know-how into AI systems. Younger talent, comfortable with AI-native workflows, is being brought in to build on top of that foundation. The two need each other.

Products are being rebuilt for agents, not just humans

The question is no longer how do you build for your user. It is how do you build for the agent acting on behalf of your user.

Enterprise customers are already moving from single LLM calls to full agent workflows. Button-based UIs are giving way to conversational interfaces. And the panellists were aligned on one projection: 90% of apps will eventually disappear as agents call APIs directly. What remains will be the final 5 to 10% that requires genuine domain expertise. That is where the defensibility lives.

API-first architecture is no longer optional. It is the foundation on which agent-ready products are built.

Business models are shifting underneath everyone

Seat-based pricing was built for a world where humans do the work. That world is changing.

Revenue is moving from human usage to AI consumption. Usage-based and outcome-based models are becoming the standard. The panellists were clear: companies that have not started building agent-friendly APIs are already behind.

For companies operating across geographies, the complexity deepens. Running separate AI stacks for Western and Chinese markets is increasingly a necessity. Chinese models may be 20 to 30% less capable than their Western counterparts, but compliance requirements make dual infrastructure investment unavoidable.

In payments specifically, the shift is even more fundamental. Agent-to-agent authentication and authorisation systems are being built to handle a world where AI agents are increasingly initiating, authenticating, and settling transactions on behalf of consumers and businesses.

The memory problem is the one nobody talks about enough

Ask any team building seriously with agents and they will tell you the same thing. Memory is hard.

Getting agents to retain context reliably across sessions, teams, and markets is where a significant amount of real engineering work is happening. The approach that is emerging involves external storage systems with session summarisation: extracting key topics, preferences, and interaction history after each session, and retrieving them proactively when a new session begins.

A multi-model approach is also becoming standard for reliability. Running Claude, Gemini, and DeepSeek in parallel to avoid single-point failures is not overkill. It is infrastructure.

The one piece of advice for founders

Do not build a chatbot and call it an agent. Start with a real workflow. Own the outcome. Go deep before you go wide.

The companies that will win in the agentic era are not the ones with the most sophisticated models. They are the ones that have embedded their agents so deeply into enterprise workflows that replacing them means re-extracting years of institutional knowledge from scratch.

That is the moat. And it is being built right now.

For the latest news on Vertex Ventures SE Asia and India and our portfolio companies, follow us on Linkedin or subscribe to our monthly newsletter.