I remember sitting in a boardroom in Singapore back in early 2024, advising a group of logistics directors on their first “Copilot” implementation. At the time, the room was buzzing with the novelty of it all. We were all marveling at how a chatbot could summarise a meeting or draft an email. But even then, I could see the cracks forming. One director leaned over and whispered, “Vijay, this is great, but I still have to tell it exactly what to do every five minutes. It’s like having a very talented intern who lacks any initiative.”
Fast forward to today, February 2024, and that “talented intern” has finally graduated. We are standing at the precipice of the most significant shift in enterprise technology since the move to the cloud: the transition from generative AI as an assistant to agentic AI as an autonomous orchestrator.
The “chatbot era”—defined by human-in-the-loop interactions where we spent half our day “prompt engineering”—is effectively over. As we look ahead to the major launches scheduled for next month from giants like EY and Huawei, it is clear that 2026 is the year the enterprise stops talking to AI and starts letting AI work.
The 40% Inflection Point: Why 2026 is Different
For the past year, we have been hearing whispers about “agentic workflows.” But the data is now catching up to the hype. Gartner recently predicted that by the end of this year, a staggering 40% of enterprise applications will be agentic. To put that in perspective, at the start of 2025, that number was less than 5%.
This isn’t just a marginal improvement in software. It is a fundamental change in the architecture of how business gets done. In the old model—what I call the “Prompt-Response” cycle—AI was reactive. You gave it a task, it gave you an output, and you decided what to do next. It was a linear, human-dependent process.
The new “Agentic” model is different. These systems don’t just respond; they reason, they plan, and they execute. They have “agency.” If you tell an agentic system to “optimise the Q3 supply chain for the APAC region,” it doesn’t just give you a list of suggestions. It logs into your ERP, analyses real-time shipping delays, checks current inventory levels in your Jurong warehouse, negotiates with secondary suppliers, and presents you with a set of completed actions for approval—or in some cases, it simply executes them within pre-defined guardrails.
Frankly, if your organisation is still focusing on how to write better prompts for a chatbot, you are already behind. The competitive moat in 2026 isn’t about who has the best “AI whisperers”; it’s about who has the most robust agentic orchestration.
The Horizon: EY, Huawei, and the Infrastructure of Autonomy
Next month’s anticipated launches from EY and Huawei are set to provide the definitive blueprint for this shift. Having spent decades advising C-level executives on technology transitions, I’ve seen many “game-changers,” but these feel different because they address the two biggest hurdles to autonomous AI: governance and infrastructure.
EY is expected to unveil its “EY.ai Agentic for Sales” platform, powered by a sophisticated partnership with Snowflake and Canva. This isn’t just another CRM plugin. From what we understand of the upcoming release, it aims to unify real-time data intelligence with autonomous content creation. Imagine an AI agent that doesn’t just tell a sales VP that a lead is “warm,” but autonomously researches the prospect’s latest financial filings, generates a bespoke, brand-compliant presentation in Canva, and drafts a personalised outreach strategy—all while ensuring every step complies with the company’s internal governance policies stored in Snowflake.
On the other side of the pond, Huawei is preparing to showcase its “AUTINOps” (AI-Native intelligent operations) solution alongside the “Atlas 950 SuperPoD” at MWC. This is the “heavy lifting” side of the agentic revolution. For years, we’ve struggled with the “velocity paradox”—AI agents are fast, but our underlying infrastructure is often too slow and fragmented to support them. Huawei’s approach suggests a move toward “Level-4 autonomous networks,” where AI agents handle network faults and risk identification without human intervention.
The bottom line is this: when the world’s largest consultancy and one of the world’s largest infrastructure providers both pivot toward autonomous orchestration in the same month, the “experimentation” phase of AI is officially dead.
From “Human-in-the-Loop” to “Human-on-the-Loop”
This shift necessitates a radical rethinking of governance. For the past two years, “Human-in-the-Loop” (HITL) has been the gold standard. The idea was simple: AI does the work, but a human must check every single output before it goes live.
In a world of single chatbots, HITL worked. But in 2026, when you have dozens or even hundreds of agents interacting in a Multi-Agent System (MAS), HITL becomes the ultimate bottleneck. If a human has to approve every micro-decision made by a fleet of agents, you lose the very speed and scale that makes the technology valuable.
Enter the “Human-on-the-Loop” (HOTL) model.
In HOTL, the human role shifts from “executor” to “governor.” Instead of checking individual outputs, we are now responsible for defining the “Constitution” of the agentic ecosystem. We set the high-level goals, the risk thresholds, and the ethical boundaries. We monitor the system’s performance via high-level dashboards, stepping in only when the agents encounter an edge case or a conflict they cannot resolve.
I once advised a regional bank in Singapore that was terrified of letting AI anywhere near their customer-facing processes. Their breakthrough didn’t come from a better model; it came when they realised they could treat AI agents like employees with limited signing authority. Just as you wouldn’t let a junior analyst approve a $10 million loan, you don’t let a first-generation agent reroute your entire treasury. But you can let them handle the 90% of routine tasks that fall within clear policy boundaries.
The Rise of Multi-Agent Systems (MAS)
The true magic of the 2026 shift lies in orchestration—specifically, how multiple agents talk to one another. We are moving away from the “One Big Model” approach toward a “Microservices for AI” architecture.
In a modern enterprise workflow, you don’t have one agent that does everything. You have a “Supervisor Agent” that acts as the conductor of an orchestra. When a complex request comes in, the Supervisor deconstructs it into smaller tasks and assigns them to specialized agents:
- The Researcher Agent (fetching data from internal and external sources).
- The Analyst Agent (running the numbers and identifying patterns).
- The Compliance Agent (cross-referencing the plan with local regulations).
- The Execution Agent (interfacing with the API to make the changes).
This hierarchical orchestration is how we solve the “hallucination” problem. By forcing agents to cross-check each other—where the Compliance Agent can “veto” the Analyst Agent—we build a system of digital checks and balances that far exceeds what a single human could manage.
The Reality Check: The Scaling Gap
Despite the excitement, I must be blunt: not everyone is ready for this. While 40% of apps may be “agentic” by year-end, only a fraction of organisations will be getting real value from them.
The biggest obstacle I see across APAC today isn’t the AI models—it’s the data. You cannot have an autonomous agent if your data is trapped in 1990s-era silos or if your “real-time” inventory updates only happen every six hours. An agent is only as good as its senses. If it’s blind to the reality of your operations, it will just make the wrong decisions faster.
Furthermore, we are seeing a “Governance Gap.” An EY survey recently found that while nearly every tech leader is prioritising autonomous AI, more than half of these initiatives lack a formal oversight framework. This is the “velocity paradox” in action: the faster we go, the more likely we are to miss the turn.
What C-Suite Leaders Must Do Now
So, if you are a CTO, a CIO, or a VP of Operations sitting in Singapore, Sydney, or Hong Kong today, what should your roadmap look like?
First, stop buying “features” and start building “capabilities.” Don’t just look for the next “AI-powered” button in your existing software. Ask your vendors about their Agent-to-Agent (A2A) protocols. Can their agents talk to your other systems autonomously? If the answer is “no,” you are buying a dead-end tool.
Second, focus on “Outcome Responsibility.” Shift your KPIs from “how many prompts did we run?” to “what business outcomes did the agentic system achieve?” Start with low-risk, high-volume workflows—like merchant onboarding, internal IT tickets, or basic procurement—and build your governance muscle there.
Third, invest in your “Data Fabric.” As I’ve said many times before, brains need senses. Your agents need a live, unified view of your operations. If you haven’t solved your data integration problem, your agentic strategy is just a house of cards.
Conclusion: The End of the Beginning
The shift from prompts to agents marks the end of the beginning for AI in the enterprise. We are moving past the “wow” factor of generative text and into the “how” factor of autonomous execution.
Next month’s launches from EY and Huawei aren’t just product announcements; they are the starting gun for a new era of enterprise productivity. Those who embrace the shift from being “prompters” to “orchestrators” will find themselves leading organisations that are faster, leaner, and more resilient than ever before. Those who wait for the “perfectly safe” chatbot will find themselves relegated to the history books of the pre-agentic age.
The question for 2026 isn’t whether AI can do the work. The question is: are you ready to let it?