Last week (November 2–4), three heavyweight reports landed at once: Gartner’s Top 10 Strategic Technology Trends for 2027, Gartner’s 3Q26 Quarterly Emerging Risks Report, and IDC’s FutureScape 2027 global predictions. In the same week, the Artificial Intelligence Parallel Forum of the 2026 World Laureates Forum was held in Lingang, Shanghai, under the theme “The Next Decade: The Development and Governance of Artificial Intelligence”.
To summarize this week's shift in one sentence, the most accurate is what Arnold Gao, Gartner research vice president, said to Yicai: "The focus in 2026 will be how to build and apply AI; the focus in 2027 will shift to how to manage AI-driven autonomous systems and redesign enterprises' technology architecture."
And what makes finance sit up straight the most is the number IDC puts on governance: By 2028, Global 2000 companies that have not installed an "emergency stop switch" for AI agents in production environments will pay 25% more in liability insurance premiums, and the number of material audit findings will double. Governance is starting to show up on the bill.
I. Three Reports, One Turning Point: From "Building AI" to "Governing AI"
First, look at Gartner's top ten trends. It groups the ten technologies into three main themes—the non-text AI frontier, governing autonomous agent platforms, and durable agile capabilities—and devotes most of the space to "how to govern them":
| Main storyline | Includes trend | Key predictions (Gartner, all are forecast values) |
|---|---|---|
| Non-text AI Frontier | Large Quantitative Model LQM, speech reasoning model, world model | By 2030, about 25% of strategic forecasting and operational modeling will be replaced by LQM instead of LLM; enterprise autonomous voice processing will be less than 10% in 2026 → 70% in 2030 |
| Autonomous Agent Governance Platform | AI context platform, AI coding agent control framework, AI agent identity, autonomous security exposure remediation | Enterprises that complete data semantic governance by 2027 can improve agentic AI accuracy by up to 80% and reduce costs by up to 60%; by 2028, 25% of enterprise security incidents will stem from lapses in agent permission management |
| Enduring Agile Capability | Sovereign technology, disposable apps, wetware as a service | By 2030, more than 75% of non-U.S. enterprises will implement digital sovereignty strategies; 80% of new applications at leading enterprises will adopt a disposable model |
The four terms in the second main line correspond to four plain questions: Does the agent understand the business, who manages permissions, who verifies the code, and who fixes vulnerabilities. The speed of machine output has exceeded the upper limit of manual review, and governance must be pushed down to the platform layer.
Turning again to Gartner's November 2 3Q26 Quarterly Emerging Risk Report, based on a sample of 308 risk management leaders, auditors, and executives. Two new risks entered the list for the first time and shot straight into the top five:
3rd
"AI strategic miscalculation" enters the list for the first time — overly aggressive investment, choosing the wrong use cases, and pushing ahead before the organization is ready
4th place
"AI-driven competitive substitution"—competitors use AI more effectively, leaving you behind
The top two are "AI discovering network vulnerabilities" and "geopolitical energy supply shocks" (for two consecutive quarters), while the fifth is "AI workforce readiness gap"—four of the top five are AI-related. Kevin Mercado of Gartner said: "AI risks are moving from technical and governance concerns into the core of enterprise strategy."
These two new risks point in opposite directions: on one side, the fear of investing too aggressively and returns not materializing; on the other, the fear of investing too slowly and being left behind by competitors. Caught in the middle, CIOs need neither more aggressive nor more conservative action, but "verifiable readiness"—first prove that the organization can absorb it, then talk about scale.
II. Governance Is Explicitly Priced for the First Time: Kill Switch, Liability Clauses, and Hidden Bills
IDC's "FutureScape 2027" spans 25 research areas and includes a total of 250 predictions (reported by domestic media on November 2); several related to agent governance can almost be copied directly into next year's procurement list:
IDC: By 2028, G2000 enterprises that have not configured an "emergency stop switch" for AI agents in production environments will have liability insurance premiums 25% higher, and the number of major audit findings will double
- By 2028, 75% of G2000 enterprises will include "liability clauses for actions autonomously initiated by AI agents" in contracts with technology vendors; the absence of such provisions in contracts will become a risk warning signal during vendor selection.
- By 2028, 70% of G2000 enterprises will have the ability to continuously validate AI agent decisions; "supervised autonomy" with clear escalation thresholds will become the standard operating model.
- Suppliers have not publicly adjusted prices, but the proportion of enterprises actually bearing higher agent-related costs will rise from the current 20% to 55% by the end of 2027.
- By 2029, 60% of enterprises will run most business processes in an "AI-first" model, compared with only 13% in 2026; during the same period, CIOs at G2000 will shift 25% of service budgets from "pricing by labor hours" to "outcome-based pricing for agent clusters."
The deployment side is even more telling. IDC's 'Future Enterprise Resilience and Spending Survey': As of July 2026, 95.4% of enterprises have deployed AI agents into an average of 11 production workflows, with monthly inference budgets nearing $118,000; yet most enterprises still rely on manual approvals, periodic audits, and post-hoc corrections, which are largely ineffective for real-time AI agents.
Another financial fact easily mistaken for a "technical problem": IDC predicts that by 2028, 60% of enterprises will be unable to effectively convert their own data into measurable business value; by 2029, 35% of G2000 companies will establish an AI model registration management system by jurisdiction. Governance is not a cost center; it determines whether data can become an asset.
Tianxunda Technology cites Cloudera's survey this month: 98% of surveyed enterprises had scaled back or withdrawn deployments due to AI operational failures, hallucinations, or unexpected behavior (vendor survey, respondent self-report, not independently audited, for directional reference only). A chatbot answering incorrectly is just a wrong sentence; an agent making a mistake could be a payment sent in error.
3. A counterintuitive conclusion: ERP will not be replaced, but the architecture must be redone
What ERP practitioners should actually mark as the key point this week is instead a "bucket of cold water" from Gao Ting on November 3: AI is lowering software development costs, and tool software with simple functions, limited differentiation, and short cycles, as well as some traditional SaaS, may be replaced by AI-generated alternatives; but core software such as ERP and CRM will not be replaced quickly, because what they truly provide is enterprise business logic and industry knowledge, not just the software itself.
This is two sides of the same coin as “disposable applications”: by 2030, 80% of new applications at leading enterprises will adopt a “use-and-discard” model, with deactivation trigger conditions, data archiving, and permission revocation defined from the outset. Enterprise software will thus split into two layers—core systems that carry rules and ledgers, and a large number of short-lifecycle peripheral applications rapidly generated by AI—and their governance approaches and budget accounting should not be mixed together.
But "not being replaced" does not mean "no need to change." Gao Ting also mentioned that AI agents will drive enterprises to view AI as core productivity and even the "enterprise brain," redesigning permissions, processes, and infrastructure around agents; voice reasoning models, AI context platforms, and AI coding agent frameworks are in a 1-to-3-year window, and the investment principle is "create business value in the near term, build core capabilities in the medium term, and retain technological options in the long term."
In CIO language: there is no need to rush to dismantle ERP next year, nor to chase every new model. Three things should change—organize business knowledge into context that agents can understand, give every agent an auditable identity and least privilege, and leave a stop switch on the core ledger pathway that can be pressed at any time.
IV. Practical Recommendations for CIOs / Digital Leaders: Where Should the 2027 Budget Be Reallocated?
1. First establish "brakes and a ledger," then scale up. For every agent put into production, equip it with an independent identity, least privilege, pause/disable buttons, decision logs, and a named responsible person; for high-risk actions (payments, posting, bulk changes to master data, external sending), retain human confirmation. This is also exactly the implementation form of the three hard constraints—"controllability / full-process traceability / digital passport"—in China's mandatory national standard for agent security in October.
2. Treat contracts as governance tools. When renewing this year, check item by item: attribution of responsibility for autonomous agent behavior, billing units (whether retries/failures/manual handoffs are billed), spending caps and alerts, price adjustment notice period, log export and exit clauses. IDC says 75% of G2000 will write liability clauses into contracts before 2028; those who write them early won't lose out.
3. Manage AI from a portfolio perspective, not a project perspective.Gartner recommends evaluating AI as "a set of business decisions": tie investments to strategic priorities, verify readiness before scaling, and measure value by "impact on decision quality" rather than by how many work hours are saved. The evidence is sobering—in a 2026 Gartner survey, only 17% of CFOs said that more than half of AI projects delivered measurable revenue growth or cost savings.
4. Layered governance for core systems and disposable applications. Keep the core ERP clean, stable, and rollback-capable, and consolidate business logic and industry knowledge into versionable assets; peripheral short-cycle tools may be rapidly generated by AI, but must have preset deactivation dates, data archiving, and permission revocation paths.
One timing pitfall to flag: the vast majority of the above figures are projections (2027–2031), not established facts; the 95.4% deployment rate, $118,000 monthly budget, and 98% withdrawal rate come from different surveys with inconsistent definitions, so they should not be combined into a conclusion. The only sequence executable today is this—readiness before scale.
To conclude: what is worth remembering this week is not that some model topped the leaderboard again, but that governance, for the first time, has a price, contract terms, and a board meeting agenda. In the past, governance discussions were about 'what should be done'; starting in 2027, they will be about 'how much more you pay if you don't do it.' This is not necessarily bad news for CIOs—only when something can be clearly accounted for can it finally make it into the budget, be written into responsibilities, and truly be implemented.
Sources: Gartner's 'Top 10 Strategic Technology Trends for 2027' and the Gao Ting media briefing (Zhidongxi, Yicai / Tencent News, 2026-11-03/04; Cyber Research Institute interpretation); Gartner's '3Q26 Quarterly Emerging Risk Report' press release (2026-11-02) and Help Net Security report; IDC's 'FutureScape 2027' (reported on 2026-11-02 by Qianlong.com, Rizhao News Network, Tencent News / IT Home, with 250 predictions across 25 domains), IDC's 'Future Enterprise Resilience and Spending Survey' (as of 2026-07), China Business Journal Network's IDC assessment (2026-11); Cloudera's survey this month (cited via Tianxunda Technology, not independently audited). Percentages containing years are institutional forecast values; self-reported data and third-party research are listed separately without extrapolation. As of 2026-11-07.
Zhongdi Information · Dedicated to Odoo and digital solutions for enterprise management
