Published by Digigen | Google Cloud Partner for Thailand and Singapore
A deep dive into Google Cloud's most consequential AI platform launch, with dedicated guidance for Singapore and Thailand enterprises.
If you build with Google Cloud, today's announcement of Gemini Enterprise Agent Platform is the most significant product change in two years. It is not just another product launch. It is the formal sunset of Vertex AI as a standalone service.
Buried in the official announcement is one sentence that should reshape every Google Cloud roadmap in your organization:
"Moving forward, all Vertex AI services and roadmap evolutions will be delivered exclusively through the Agent Platform, rather than as a standalone service."
Vertex AI is now a chapter of Agent Platform. Everything Google Cloud builds for AI from this point forward will be agent-first. If your team is still thinking about "deploying a model" rather than "deploying an agent," your mental model is officially out of date.
This post is a deep dive into what was announced, what it actually means in practice, and what we are advising our enterprise clients in Singapore and Thailand to do over the next 90 days.
For the last three years, the enterprise AI conversation has been organized around models. Which one is fastest? Which one is cheapest? Which one is best at reasoning? Google Cloud just declared that conversation finished.
The Agent Platform pitch is about delegating outcomes, not managing tasks. The product framing of "build, scale, govern, optimize" is borrowed directly from the DevOps lifecycle, and that is intentional. Google is saying agents are now production software, and they need a production platform with the same rigor as everything else you ship.
For our clients in Bangkok and Singapore, this matches what we have been seeing for months. Most enterprises are not stuck on "which model to pick." They are stuck on "how do we deploy this safely, observe it in production, prove it to the audit committee, and improve it over time." Agent Platform is Google's structured answer to all four.
The build pillar has the most concrete developer-facing improvements. Here is what matters.
ADK is now processing more than six trillion tokens per month on Gemini models. The big upgrade is a graph-based framework for organizing agents into networks of sub-agents. This is a meaningful architectural shift. Instead of writing one monolithic agent that tries to do everything, you compose specialized sub-agents into reliable, inspectable workflows.
For teams building anything beyond a simple chatbot, this is going to be the standard pattern. We expect to see most production agent architectures move to this graph model within 12 months.
Agent Studio is the low-code visual interface, and the killer feature is the export-to-ADK path. You can prototype visually with business stakeholders, then export the logic into a code-first ADK environment when you need deeper customization.
This solves a real problem we see constantly. Business teams build something useful in a low-code tool, then engineering has to rebuild it from scratch when it needs to scale. The Studio-to-ADK pipeline eliminates that throwaway work.
Agent Garden ships templates for code modernization, financial analysis, economic research, invoice processing, and more. These are not toy demos. They are designed to be composed into multi-agent systems as starting points.
For Thai and Singapore enterprises that want to move fast, the invoice processing and financial analysis templates are particularly relevant. Most CFO offices in our region run on a mix of SAP, Oracle, custom Excel workflows, and PDFs from suppliers. Agent Garden gives you a credible starting point.
This is where the platform earns its enterprise stripes. Most agent prototypes die between proof-of-concept and production because the runtime cannot handle real load. Google has rebuilt the runtime to address that head-on.
The revamped Agent Runtime delivers sub-second cold starts and lets you provision new agents in seconds. For anyone who has tried to deploy serverless agent infrastructure at scale, this is a significant improvement. Cold start latency was one of the main reasons agentic UX felt clunky in production.
The runtime now supports agents that run autonomously for days at a time. This unlocks workflows that were previously not viable: multi-day sales prospecting sequences, complex reconciliation tasks, deep research projects that span multiple business hours.
For our enterprise clients, the obvious applications are in finance (reconciliation, audit support), procurement (multi-step vendor onboarding), and sales operations (prospect nurture sequences with stateful context).
Memory Bank replaces temporary session state with persistent, long-term memory. Memory Profiles let agents recall high-accuracy details with low latency.
Payhawk's quote is illustrative. Their Financial Controller Agent now remembers user habits to auto-submit expenses, reducing submission time by over 50%. Gurunavi's UMAME restaurant discovery app uses Memory Bank to remember user preferences, reportedly improving user satisfaction by 30%.
For Thai and Singapore enterprises building customer-facing agents in banking, insurance, retail, and hospitality, Memory Bank is what makes agents feel like genuine concierges rather than transactional bots.
Most coverage of Agent Platform will focus on build and scale. We think govern is actually the most strategically important pillar, especially for our regional clients. Here is why.
In Singapore, MAS Project MindForge requires demonstrable GenAI risk management for financial institutions. In Thailand, the Bank of Thailand and the Securities and Exchange Commission have signaled tighter scrutiny on AI deployments. Across both markets, PDPA-style data protection regulations create governance requirements that ad-hoc agent deployments simply cannot meet.
Agent Identity gives every agent a unique cryptographic ID with defined authorization policies. Every action an agent takes is auditable and traceable back to its policy. For your audit team, this is the difference between "we have an AI doing things" and "we have a documented system with verifiable controls."
The new Agent Registry indexes every agent, tool, and skill in your enterprise. This solves the "shadow AI" problem. Right now, in most enterprises we work with, individual teams are spinning up agents in n8n, Make, LangChain, or custom Python without any central oversight. Agent Registry gives you a way to bring that under control without killing innovation.
Agent Gateway is the central control point for connectivity between agents and tools. It enforces consistent security policies and includes Model Armor protection against prompt injection and data leakage. If you remember one thing from this post, remember Agent Gateway. This is the single most important component for any enterprise that wants to deploy agents at scale without creating an unmanageable security surface.
The fourth pillar closes the loop between deployment and improvement. Agent Simulation lets you stress-test agents against synthetic interactions before shipping. Agent Evaluation continuously scores live traffic with multi-turn autoraters.
The feature that surprised us most is Agent Optimizer, which automatically clusters real-world failures and suggests refined system instructions. If this works as advertised, it materially shortens the agent improvement cycle.
Singapore enterprises are uniquely positioned to lead APAC adoption of Agent Platform. Three reasons make this the right moment to commit.
For MAS-regulated institutions, Agent Identity plus Agent Gateway plus Agent Registry plus the Security Command Center dashboard is essentially a turnkey answer to the technical control requirements in Project MindForge guidance. The cryptographic agent identity gives you the auditable trail. Model Armor gives you the prompt injection and data leakage controls. The Security dashboard gives your CISO the unified view that risk committees expect. We expect this combination to become the default reference architecture for DBS, OCBC, UOB, and the broader Singapore financial services sector within 18 months.
Singapore enterprises operate across AWS, Azure, and Google Cloud as a matter of course. Regional groups headquartered in Singapore have data sitting in Jakarta, Kuala Lumpur, Bangkok, and Manila for residency reasons. Agent Platform combined with Cross-Cloud Lakehouse gives you a way to deploy agents that reason across this distributed footprint without forcing data movement. For PDPA cross-border transfer compliance, this is genuinely useful.
Google highlighted FairPrice Group's "Store of Tomorrow" program as a flagship retail case study globally. AI agents in Smart Carts, multimodal AI for store operations, generative media for thematic ads. This is Singapore enterprise AI shipping at scale today, not next year. For other Singapore retailers, F&B groups, and consumer businesses, the FairPrice playbook is worth studying closely.
For the full Singapore market recap of Google Cloud Next '26 with TPU, Agentic Data Cloud, and Workspace Intelligence coverage:
Read the Singapore Recap →Thailand has been slightly behind Singapore on enterprise AI adoption, but Agent Platform actually lands at the perfect moment for the Thai market. Here is why.
Most Thai organizations we work with are running early agent prototypes on Make.com, n8n, or custom Python. These work for proof-of-concept, but they have no governance story. The audit committee will not approve them, the security team will not let them touch production data, and they cannot scale beyond a single team. Agent Platform with Agent Identity, Agent Gateway, and Agent Registry is the bridge from prototype to production. Thai enterprises that have been waiting for "the right time" to commit to enterprise AI now have it.
The Bank of Thailand has signaled tighter expectations on AI risk management for Thai financial institutions, and the SEC has been asking listed companies harder questions about AI governance. PDPA enforcement has become more active. Agent Platform's governance architecture maps directly to these emerging Thai regulatory expectations. For Thai banks, insurers, and SET-listed corporates, this matters in the next 12 months, not three years.
For Thai consumer-facing businesses, the Omnichannel Gateway in Gemini Enterprise for Customer Experience does not include native LINE support. This is a real gap. LINE remains the dominant messaging channel in Thailand, and you cannot deploy a serious customer-facing agent strategy without it. The good news is that Agent Platform's MCP and native ecosystem integration patterns make a LINE OA bridge much cleaner to build than it would have been on Vertex AI. Digigen has built LINE integrations for Thai clients before and we expect to see strong demand here over the next two quarters.
For Thai banks running Thai-language conversational AI at scale, Memory Bank changes the customer experience equation. Agents that remember previous conversations in Thai, retain user preferences across months, and recall account context with low latency move the experience from transactional bot to genuine relationship. Combined with the inference cost improvements from TPU 8i, the economics finally support production-scale Thai-language agent deployments.
For the full Thailand market recap of Google Cloud Next '26 covering TPU 8i, Agentic Data Cloud, security, and Workspace:
Read the Thailand Recap →Google included quotes from Burns & McDonnell, Color Health, Comcast, Geotab, Gurunavi, L'Oreal, Payhawk, and PayPal. Reading between the lines, three patterns emerge.
Geotab calls out the developer experience and faster build-test-deploy cycles. Comcast highlights multi-agent architecture deployment via Agent Runtime. The platform is winning on speed, not just capability.
Both Payhawk and Gurunavi anchor their stories on Memory Bank. Without persistent context, agents are smarter chatbots. With it, they become genuine assistants.
L'Oreal explicitly mentions multi-LLM flexibility and MCP integration. Enterprises do not want lock-in. Agent Platform supports Claude Opus, Sonnet, and Haiku alongside Gemini.
PayPal's quote also references Agent Payment Protocol (AP2), the trust foundation for agent-based commerce. As autonomous agents start transacting on behalf of users, AP2 becomes a major theme to track for Singapore's digital payments sector and Thailand's growing e-commerce stack.
The platform is genuinely comprehensive. Most competing offerings (LangChain plus self-hosted infrastructure, Microsoft Copilot Studio, AWS Bedrock Agents) cover one or two of build, scale, govern, optimize. Agent Platform covers all four with first-party integration. For enterprises that want a single coherent platform, this is the strongest offering in the market today.
The governance story is the real differentiator. Agent Identity, Agent Registry, Agent Gateway, and the Security dashboard are exactly what regulated enterprises in Singapore and Thailand need.
The Vertex AI consolidation is bold and probably correct. Forcing everyone onto a single platform reduces fragmentation and accelerates the agent-first paradigm.
Migration from existing Vertex AI deployments will require planning. If you have production workloads on Vertex AI today, map your migration path now, not in six months.
Agent Platform pricing has not been fully detailed publicly. Expect the model to evolve, build in flexibility.
The maturity of some features (Agent Optimizer in particular) needs to be validated in production. Pilot the entire stack but commit to battle-tested components first (ADK, Agent Runtime, Agent Identity, Agent Gateway).
By consolidating onto Agent Platform, you are betting that Google maintains its current velocity. Based on the last 18 months, that is a defensible bet, but it is still a bet.
Catalog every existing agent, model deployment, and shadow AI workflow. You cannot govern what you have not inventoried.
Pick one high-value use case. Build it end-to-end on ADK plus Agent Runtime plus Agent Identity. Document everything.
Stand up Agent Registry, define Agent Identity policies, route everything through Agent Gateway with Model Armor enabled.
Map every existing Vertex AI workload to its Agent Platform equivalent. Build a phased migration plan with timelines.
Throughout: invest in team enablement. Agent Platform is a different skillset from traditional ML engineering.
Gemini Enterprise Agent Platform is the most consequential platform launch from Google Cloud since Vertex AI itself. It signals that Google is all-in on the agentic enterprise model, and it gives serious enterprises a credible, governed path to deploy AI agents at scale.
For Thai and Singapore enterprises, this is good news. The governance features specifically address the regulatory environment we operate in. The platform consolidation reduces complexity. The customer proof points show that production deployments at scale are happening today.
The hard part is no longer the technology. The hard part is organizational. Getting your build, security, compliance, and operations teams to operate as a single function around agent deployments. Platforms like this only succeed when the underlying organization is ready for them.
The Vertex AI era is over. The Agent Platform era is here.
Where will your organization be in 90 days?
Talk to Digigen →MAS-ready governance, Cross-Cloud Lakehouse for multi-cloud realities, FairPrice case study, and 90-day actions for Singapore enterprises.
Read full recap →TPU 8i economics for Thai-language inference, LINE integration considerations, Workspace migration, and 90-day actions for Thai CIOs.
Read full recap →Digigen is a Google Cloud Partner and Microsoft Solutions Partner with operations in Thailand and Singapore. We help enterprises design, deploy, and govern AI, cloud, and productivity solutions.
Contact us at hello@digigen.io or visit digigen.io.
For our AI agent consulting and platform work, see The Agentiv.