Expanding Managed Agents in Gemini API: Background Tasks, Remote MCP and More

Google adds production-grade agent infrastructure to Gemini API — bringing autonomous AI agents closer to everyday business deployment in Malaysia and beyond.

---

AI Summary

Google has announced new capabilities for Managed Agents in the Gemini API, designed to help developers build reliable, production-ready AI agents. The update introduces background task execution and remote Model Context Protocol (MCP) support, among other features. For Malaysian businesses, this signals a shift from AI as a conversational tool to AI as an autonomous worker that can handle multi-step operations, connect to external systems, and run tasks over extended periods without human supervision. The practical implication: companies that previously found agent development too complex or brittle may now have a more stable foundation to build on.

---

Key Takeaways

• Google's Gemini API now supports Managed Agents with background task execution, meaning AI agents can work on long-running jobs without requiring constant user interaction or an active session.
• Remote MCP support allows agents to connect to external tools, databases, and services over a network — a critical requirement for real business workflows where data lives across multiple systems.
• The phrase "production-ready" is the key signal: Google is positioning these agents for live business use, not just prototyping or experimentation.
• Malaysian developers and IT teams building on Gemini now have infrastructure-level support for agent orchestration, reducing the need to build custom scaffolding from scratch.
• The announcement accelerates the competitive race among AI platform providers (Google, OpenAI, Anthropic, Microsoft) to own the agent infrastructure layer — a space that directly affects what Malaysian businesses can build and at what cost.

---

What Happened

Google published an announcement detailing new capabilities in its Managed Agents framework within the Gemini API. The core message is straightforward: the company wants developers to build AI agents that are reliable enough for production environments — meaning live business systems where failures cost money, time, or customer trust.

Two specific features were highlighted. The first is background tasks. In traditional AI interactions, a user sends a prompt, the model responds, and the session ends. Background tasks change this model. An agent can accept a instruction, then work on it over minutes, hours, or longer without the user waiting or staying connected. Think of it as assigning a task to a staff member who goes off, does the research, compiles findings, and reports back when done — rather than standing in front of you generating a response in real time.

The second feature is remote MCP support. MCP — Model Context Protocol — is an open standard that lets AI models connect to external systems: databases, APIs, file stores, SaaS platforms, internal tools. "Remote" means these connections happen over a network rather than requiring everything to run on the same machine or in the same environment. For a business, this matters because your data does not live in one place. Customer records sit in your CRM. Financial data sits in your accounting software. Inventory sits in your warehouse management system. An agent with remote MCP can potentially reach across all of these to complete a task.

Google framed these additions as part of a broader effort to make agent development more dependable. The language around "reliable" and "production-ready" suggests Google is responding to a real pain point: many companies have built AI agent prototypes that work in demos but break under real-world conditions — inconsistent outputs, failed tool connections, timeouts, and unclear error handling.

---

Why It Matters

The distance between an AI agent that impresses in a demo and one that actually works in a business environment is significant. Most Malaysian companies that have experimented with AI chatbots or simple automation tools have encountered this gap. A customer service bot handles easy questions well but collapses when a query requires pulling order history, checking inventory, and coordinating with a logistics partner — all in one interaction.

Background task support addresses a specific failure mode. Current AI interactions are synchronous: you ask, it answers. But many business processes are asynchronous. Processing a loan application, reviewing a supplier portfolio, generating a quarterly compliance report — these are not instant-response tasks. They require gathering information, cross-referencing, running calculations, and producing structured output. Without background task infrastructure, developers had to build complex workarounds: queue systems, status polling, retry logic, timeout handling. Google is now providing some of this plumbing at the platform level.

Remote MCP addresses a different but equally important gap. An AI agent is only useful if it can access the data and systems it needs to do its job. If an agent can talk but cannot act — cannot query a database, cannot call an API, cannot write to a file — it is just a chatbot dressed up with a fancier name. MCP standardises how agents connect to external resources. Remote support means the agent and the tools it uses do not need to be co-located, which is essential in cloud-based business environments.

The competitive context matters here. OpenAI has been investing heavily in agent capabilities with its Assistants API and broader agent framework. Anthropic has pushed MCP as a standard and is building agent tooling around its Claude models. Microsoft is integrating agents across its Copilot platform. Google's move with Managed Agents is an assertion that the Gemini ecosystem should be a serious contender for businesses choosing where to build agent-based applications.

For decision-makers, the signal is this: the infrastructure layer for AI agents is maturing rapidly. What required a team of specialist engineers six months ago is becoming platform-native functionality. This does not mean building agents is suddenly easy — but the floor is rising.

---

What This Means for Malaysia

Malaysian businesses sit in an interesting position relative to this announcement. The country's AI adoption is accelerating — driven by MDEC's digital economy initiatives, the MyDIGITAL blueprint, and growing awareness among SME owners that AI is not just a corporate luxury. But the gap between ambition and execution remains wide. Many Malaysian companies know they should be doing something with AI but are unsure where to start, particularly when it comes to agents that go beyond chat.

The Gemini API changes are relevant in several specific ways. First, for Malaysian software houses and digital agencies — particularly those in the Klang Valley and Penang tech corridors — Managed Agents with production-grade features create a viable platform for building client solutions. A digital agency in Kuala Lumpur serving retail clients could build an agent that monitors inventory levels across stores, automatically generates restocking recommendations, and sends purchase orders for approval — all running as background tasks on Gemini's infrastructure.

Second, for Malaysian enterprises with existing Google Cloud relationships, the Gemini API is already within their technology stack. Adding agent capabilities does not require introducing a new vendor or navigating a separate procurement process. This reduces friction — an important factor in a market where IT budgets are scrutinised carefully and vendor consolidation is preferred.

Third, the regulatory dimension. Malaysia's PDPA (Personal Data Protection Act) governs how personal data is collected, processed, and stored. AI agents that connect to external systems via remote MCP will inevitably handle personal data — customer records, employee information, transaction histories. Malaysian businesses deploying such agents need to ensure that data flows comply with PDPA requirements, particularly around data residency, consent, and purpose limitation. The technical capability to connect agents to external systems does not automatically include legal clearance to do so. This is where Malaysian companies need to pair technical experimentation with legal review.

For the broader ASEAN region, Malaysia positions itself as a digital economy leader. Access to production-grade agent infrastructure from a major provider like Google strengthens the case for building AI-enabled services locally rather than relying on imported solutions. Malaysian startups building vertical AI tools — for Islamic finance, halal supply chain management, Bahasa Melayu content processing — now have a stronger foundation to build on.

---

How Your Business Can Use This

If you are a Malaysian business leader considering how to apply this, start with a specific workflow problem, not the technology. Identify a process in your organisation that is repetitive, multi-step, and requires pulling information from different sources. Examples: monthly GST reconciliation across accounting systems, supplier performance reviews that require data from procurement and quality systems, customer onboarding that involves verification across multiple databases.

Once you have identified a candidate process, evaluate whether the Gemini Managed Agents framework can support the workflow. The two features announced — background tasks and remote MCP — map directly to the requirements of such processes. Background tasks handle the asynchronous, long-running nature of the work. Remote MCP handles the need to connect to multiple external systems.

A practical starting point: assign one developer or technical team member to build a small prototype agent on the Gemini API over a two to four week sprint. Pick a single, contained workflow — not your most complex process. Document what works, what breaks, and where the gaps are between the platform's capabilities and your specific requirements. This is low-cost exploration with high learning value.

For companies without in-house development teams, this is a signal to begin conversations with Malaysian digital agencies and AI consultancies about agent-based solutions. Ask potential vendors specifically about their experience with Gemini's Managed Agents, how they handle PDPA compliance in agent workflows, and what their approach is to testing and quality assurance for autonomous systems.

---

The Agentic AI Angle

This announcement is fundamentally about agentic AI — not chatbots, not simple automation, but systems that can plan a sequence of actions, execute them across multiple systems, and adapt when things change. The background tasks feature is what makes this practical. An agent that must complete a ten-step supplier evaluation — pulling financial data, checking certification databases, reading news feeds, scoring results, and drafting a report — needs time and autonomy. It cannot function if every step requires a human to prompt it forward.

Remote MCP is what gives the agent reach. An agent without tool access is like an analyst locked in a room with no computer, no phone, and no files. It can think but cannot act. MCP connections are the agent's hands — reaching into databases, calling APIs, reading documents, writing outputs. Remote support means these hands can stretch across networks, which is how real business systems are organised.

Consider a concrete Malaysian example. A logistics company in Port Klang wants an agent that monitors shipment delays across multiple carriers, identifies shipments that are at risk of missing delivery windows, checks customer contracts for penalty clauses related to late delivery, and drafts proactive customer communications with revised ETAs. This workflow requires connecting to carrier tracking APIs, internal CRM data, contract databases, and communication platforms — all of which could be mediated through remote MCP connections, with the agent running the entire process as a background task that operates continuously.

---

Risks and Limitations

Google's announcement describes platform capabilities, not proven outcomes. The phrase "production-ready" is Google's positioning — whether Managed Agents actually deliver the reliability needed for live business use will depend on real-world deployment experience, which is still accumulating. Malaysian companies should treat this as early-stage infrastructure with strong potential but limited battle-testing.

Data privacy and regulatory compliance remain the responsibility of the deploying organisation. Remote MCP connections that transfer personal data to external systems may trigger PDPA obligations. Agent decisions that affect customers — credit assessments, service denials, pricing changes — may require human oversight mechanisms to satisfy both legal and ethical standards. Technical capability does not equal regulatory permission.

---

The Bottom Line

Google's expansion of Managed Agents in the Gemini API represents a meaningful step toward making autonomous AI agents practical for business deployment. Background tasks and remote MCP support address two of the most significant technical barriers: asynchronous execution and external system connectivity. For Malaysian businesses, this creates an opportunity to move beyond chatbot experiments toward agents that handle real operational workflows.

The action to take this quarter: identify one multi-step, cross-system process in your organisation, and begin a small-scale evaluation of whether agent-based automation can address it. The infrastructure is arriving. The companies that learn how to use it first will have a meaningful advantage.

---

FAQ

What is the Gemini API Managed Agents framework, and how is it different from a regular chatbot? Managed Agents is Google's infrastructure for building AI agents that can execute multi-step tasks, connect to external systems, and run background processes — going beyond the ask-and-respond pattern of a standard chatbot.

Do Malaysian businesses need a Google Cloud account to use these features? Access to the Gemini API typically requires a Google Cloud or Google AI Studio account. Companies already using Google Cloud can integrate Managed Agents within their existing setup.

How does PDPA affect AI agents that connect to external databases via remote MCP? Any agent handling personal data must comply with PDPA requirements around consent, purpose limitation, and data protection. Connecting an agent to an external system does not exempt the business from these obligations — if anything, it increases the need for clear data governance.

---

Sources / References

• Google Blog — "Expanding Managed Agents in Gemini API: background tasks, remote MCP and more" (https://blog.google/innovation-and-ai/technology/developers-tools/expanding-managed-agents-gemini-api/): Primary source for all factual claims about the announcement, including the specific features of background tasks, remote MCP support, and Google's production-ready positioning for Managed Agents.