Introduction — When AI Shifts from “Assistant” to “Worker”
In 2024, when people talked about AI in the enterprise, most imagined chatbots answering questions or tools that helped draft emails. But by 2026, we are entering a genuinely new era — the era of Agentic AI, where AI does not just “suggest” but actually gets the work done end to end, makes decisions independently, and coordinates with other systems without requiring humans to guide every step.
Gartner’s numbers make this shift clear: enterprise applications with task-specific AI agents are expected to rise from less than 5% in 2025 to 40% by the end of 2026 — an eightfold increase in just one year.
But there is another side to the story. Gartner also predicts that more than 40% of Agentic AI projects will be canceled by the end of 2027 due to escalating costs, unclear business value, or inadequate risk controls.
This article looks at both the upside and the downside, along with practical guidance for Thai business leaders.
The Current Landscape — The Leap from 5% to 40%
To understand the scale of this shift, we first need to look at how Agentic AI differs from earlier forms of AI.
Traditional AI (Copilot): A human tells AI what to do one step at a time, such as “summarize this document” or “draft a reply to this customer email.” The AI completes the task and then waits for the next instruction.
Agentic AI: The AI is given a goal, such as “resolve this customer complaint completely,” and then plans the steps on its own — finding the customer’s history, analyzing the issue, drafting a response, sending the email, following up, and preparing a summary report, all without a human pressing a button at each stage.
Data from the first half of 2026 shows that organizations worldwide have increased their use of AI agents by around 18% compared with the previous period, an acceleration rate that outpaces most other technologies in recent years.
Three main factors are driving this growth:
- Large language models are now capable enough — AI can genuinely understand context, plan across multiple steps, and solve complex problems.
- Processing costs are falling rapidly — the cost per AI call is dropping significantly each year, making production use economically viable.
- Business pressure is mounting — shortages of skilled labor, rising labor costs, and competitors that are already moving first with AI agents.
The Agentic AI Market Is Exploding — Numbers Executives Need to Know
The Agentic AI market is expanding at a pace rarely seen in the tech industry:
| Indicator |
Data |
| Agentic AI market value in 2026 |
Approximately $9–10 billion |
| Projected market value in 2034 |
$139–199 billion |
| Annual growth rate (CAGR) |
~40–44% |
| Share of enterprise IT budgets spent on Agentic AI |
10–15% in 2026 (IDC estimate) |
| Enterprise software revenue from Agentic AI in 2035 |
More than $450 billion (~30% of total revenue) |
What do these numbers tell us? Agentic AI is not a short-lived trend — it is becoming part of core business infrastructure, much like cloud computing evolved from an “option” into the “standard” over the past decade.
Gartner predicts that by 2035, Agentic AI will drive 30% of enterprise software revenue, or more than $450 billion. This is not just a trend line — it is an industry-wide restructuring.
For organizations making decisions now, one point stands out: IDC estimates that 10–15% of enterprise IT budgets in 2026 will be allocated to Agentic AI. If your company has not budgeted for this yet, you may already be falling behind competitors.
Why 40% of Projects May Fail
This is the side of the story that gets less attention, but it matters just as much. Gartner forecasts that more than 40% of Agentic AI projects will be canceled by the end of 2027. The three main reasons are:
1. Costs Escalate Beyond Expectations
Many organizations begin Agentic AI initiatives with a small, promising PoC (Proof of Concept). But once they scale into production, costs can rise sharply — including compute, maintenance, team training, and data management.
The most overlooked cost is often the “cost of mistakes.” When an AI agent makes the wrong decision in a critical workflow, the cost of fixing the damage can be higher than having a human do the work in the first place.
2. Business Value Is Not Clearly Measurable
“AI works 50% faster” sounds impressive. But if that process is not actually a bottleneck, greater speed does not automatically translate into revenue or profit. Many companies deploy AI agents into processes that “look innovative” instead of choosing ones that create real business impact.
3. Risk Controls Are Inadequate
What makes Agentic AI different from traditional AI is that it actually takes action — sending emails, approving documents, placing orders, or updating records in business systems. Without the right controls, the damage from AI that “acts incorrectly” can be far greater than AI that merely “suggests incorrectly.”
Common issues include:
- No human-in-the-loop approval mechanism for high-impact decisions
- No clear audit trail showing what the AI decided and why
- No clear guardrails defining what the AI is allowed and not allowed to do
- No fallback plan for when the AI fails or systems go down
What This Means for Thai Businesses
For Thai organizations, Agentic AI brings both opportunity and complexity across several dimensions.
Clear Opportunities
Reducing labor gaps: Thailand is entering an aging society, and many sectors already face shortages of skilled workers. AI agents can take over repetitive, expertise-heavy work — allowing people to focus on tasks that require judgment and creativity.
Strengthening competitiveness: Thai companies that adopt Agentic AI early can gain advantages in both cost and speed, especially in industries competing internationally, such as manufacturing, logistics, and financial services.
Upgrading customer service: AI agents with improving Thai-language capabilities can provide 24/7 customer support, reduce wait times, and handle more complex requests without always escalating to human staff.
Challenges That Require Preparation
Data readiness: AI agents depend on accessible, high-quality data. Many Thai organizations still have fragmented data across multiple systems, inconsistent standards, or formats that are difficult for AI to use effectively.
Regulation and privacy: Thailand’s Personal Data Protection Act (PDPA) sets clear compliance requirements. Allowing AI agents to access and process personal data requires thoughtful design and governance.
Workforce capabilities: Operating and supervising AI agents requires different skills from using traditional software. Teams must understand both the business and the technology in order to set boundaries and review AI performance effectively.
Strategy for Executives — 5 Steps to Start Today
Based on the data, here are five practical steps executives should consider.
1. Start with the Most Painful Problem, Not the Most Exciting Technology
Do not start with the question, “Where can we use Agentic AI?” Start with, “Which process is costing us the most money, time, or customers?” Then evaluate whether an AI agent is truly a good fit for that problem.
2. Define Measurable ROI Before Launching the Project
Every project should have clear target metrics — how much cycle time will be reduced, how much cost will be saved, and how revenue will increase. If you cannot define ROI clearly, do not start.
3. Design the “Braking System” Before the “Acceleration System”
Before allowing AI agents to operate in live environments, define clearly:
- What AI can do on its own without approval, such as answering general questions
- What AI must get human approval for before acting, such as approving budgets above X amount
- What AI is strictly forbidden from doing, such as deleting customer data
4. Start Small, but Think Big
Begin with pilots in 1–2 high-impact, low-risk processes. Measure results, learn quickly, and scale systematically. Do not attempt a “big bang transformation” — the statistics are clear that 40% of oversized projects will fail.
5. Invest in People as Much as Technology
Build teams that understand both business operations and AI so they can effectively supervise AI agents. In the age of Agentic AI, the most valuable skill is not coding — it is the ability to define goals, design processes, and verify outcomes.
Conclusion — Massive Opportunity, but Only with a Plan
Agentic AI is the biggest shift in enterprise technology since cloud computing. The numbers do not lie — the market is growing from $9 billion to nearly $200 billion within eight years, and 40% of enterprise applications are expected to include AI agents by the end of this year.
But the numbers also tell the other side of the story — more than 40% of projects will fail without sufficient planning.
The key to success is not adopting AI as fast as possible, but adopting it as intelligently as possible — choosing the right problems, defining clear ROI, building strong control systems, and preparing people to guide and supervise AI effectively.
Organizations that get this right will not just “keep up” — they will build an advantage that competitors struggle to match.
If you would like advice on building an Agentic AI strategy for your organization, or want to assess readiness before starting a project, contact the Enersys team. We would be happy to help analyze your situation and design a plan that fits your business context.
References
