9 key AI trends

Джимшер Челидзе
16 hours ago
7 min read

Introduction

In 2026, AI is no longer an experiment or a question of “which model to choose.” It is a question of managerial maturity: how systematically a company builds the loop data → model → quality control → security → accountability → people development.

There is a paradox in the market: GenAI usage is growing, but not everyone gets measurable results. According to Wharton WHAIR, 82% of executives use GenAI weekly, yet only those who turn AI into a managed product – not a set of disconnected pilots – actually win.

Below are 9 trends grouped into three blocks: technology, business applications, and management. At the end there is a practical checklist and a diagnostic format.

Key points

Process wins, not model size.
Recommender and decision systems create more valuable and faster impact than pure text/image generation.
Chatbots are fading, products and agents are coming.
Decision-cycle speed is your competitive advantage.
Without governance you get incidents, fines, and stress.

For owners

If profit is not growing – look at discounts, margins, deal approval time, and cost of goods sold. Fix current indicators: this is your baseline for any pilot.
If controllability is falling – build a corporate knowledge base with an AI assistant for policies and FAQs, and embed AI into processes and controls. Measure the number of approvals and errors weekly.
If routine is killing you – start with data extraction from documents: PDFs, scans, emails. Track processing time and the share of rework/returns.

Block I. Technology foundation

Trend 1. Lowering the entry barrier

Previously, AI model development was accessible only to large corporations. Today, building custom AI models is still expensive, but now there are ready-to-use services (ChatGPT and analogues) and self‑service AI tools: pre‑trained models, low‑code/no‑code, built‑in assistants in enterprise products, and infrastructure for local deployment.

Now the winners are not those with “the biggest model” or budget, but those who integrate AI into processes faster and more carefully.

However, the “bottleneck” has shifted: it is no longer technology, but implementation management – use‑case selection, procedures, data quality, and outcome control. Without rules, employees create a shadow AI landscape, outside the control of IT and InfoSec.

Trend 2. Models need fewer data to train: lower costs, higher risks

Technology is progressing rapidly. It takes fewer and fewer data to build models from scratch, and the way LLMs are used changes the approach. For example, LLMs can be trained and configured in few‑shot and zero‑shot modes, when they see just 2–3 examples.

Along with benefits, risks also emerge. A telling example: Microsoft Research VALL-E demonstrated voice synthesis from a three‑second audio fragment. Microsoft did not release the code because of abuse risks: impersonation, voice spoofing, and bypassing voice authentication.

For business the conclusion is simple: the easier the “magic” becomes, the more important anti‑fraud, security, and legal constraints are – not after a pilot, but before scaling.

Trend 3. Local deployment

Models are becoming less demanding in terms of IT resources. Algorithms themselves are getting more efficient. Model optimization and compression make local deployment a practical alternative to cloud – especially where confidentiality, latency, or regulation are critical.

In parallel, the “model orchestration” architecture is getting stronger: instead of a single “universal” LLM, a set of specialized models is used (extraction, classification, generation, verification), connected by routing, security policies, and monitoring.

At the same time, more and more tools appear that support exactly these scenarios.

Trend 4. Multimodality as a competitive standard

Modern solutions process text, tables, documents, images, audio, and sensor signals simultaneously. For business, this is a shift from “answers in chat” to end‑to‑end scenarios.

It is multimodality that makes AI applicable to real‑world processes, where data are rarely purely textual.

Block II. Business applications

Trend 5. Digital advisors (Decision Support Systems)

DSS are the main “money‑making” class of solutions that address executives’ needs and create tangible value.

Examples of DSS use cases

Pricing and demand/revenue alignment
Supply‑chain optimization
Credit approval decisions
Maintenance planning and predictive analytics for equipment failures
Insurance and risk scoring
Configuration of IT systems and placement of telecom towers
Revenue and profitability forecasting
Natural‑disaster prediction and recommendations on prevention/response
Optimization of product portfolios, loyalty programs, and churn reduction
Procurement and inventory planning, etc.
Project and portfolio management
Recommendations on how to work with contacts and clients

Case: BlackRock and the Aladdin platform

BlackRock is the largest asset‑management company in the world. At the end of 2025, its assets under management exceeded $14 trillion for the first time – a record for the industry. Q4 2025 revenue grew by 23% to $7.01 billion.

The point of this case is not size: BlackRock has spent many years building a platform‑based management contour. Aladdin is a single system for portfolio, risk, and operations management that turns analytics and control into the backbone of the operating model. This system became one of the key tools that helped the firm navigate the 2008 crisis.

Business takeaway: competitive advantage is created not by “a single model”, but by a platform of solutions + data + governance.

Trend 6. Synergy of AI with other digital technologies

Isolated AI often produces limited effect. Strong economics appear when there are sensors, communication channels, data storage and processing infrastructure, an anomaly‑detection model and, most importantly, a response process.

Case: China’s high‑speed railways

SCMP, citing a peer‑reviewed paper in the journal China Railway, reports that an AI system in Beijing processes data from the entire network (45,000 km) and flags anomalies in about 40 minutes with 95% accuracy. The number of minor track defects fell by 80%, and there were no speed‑reduction warnings due to serious faults over the year.

Business takeaway: impact is created not by text generation, but by the chain sensors → analytics → action → quality control.

Trend 7. From chatbots to products and agents

The industry is moving away from “bots that answer” towards solutions embedded into processes, with permissions and constraints, that perform actions and are measured by process KPIs, not “dialogue quality.”

Gartner forecasts that by the end of 2026, 40% of enterprise applications will be integrated with task‑specific AI agents – up from less than 5% in 2025. By 2035, agentic AI may generate around $450 billion, or 30% of enterprise software revenue.

But autonomy cannot be turned “to max” immediately. To move beyond the “which model is better” debate, use an autonomy‑level framework.

Autonomy framework

Type	What it does	Advantages	Risks	How to implement
Assistant	Searches, summarizes, suggests options	Fast, cheap, low risk	Hallucinations, limited impact	FAQ / knowledge base, logging, curator
Co‑pilot	Handles part of the task under human supervision	Balance of speed and control	People may stop double‑checking	Checklists, mandatory validation
Agent	Autonomously performs operations	Reduces manual effort	High cost of errors	Non‑critical processes, minimal rights, monitoring
Multi‑agent	End‑to‑end processes with several agents	Maximum potential	High complexity and cost	Only with mature processes, data, and governance

Rule: do not skip levels. Assistant → Co‑pilot → Agent → Multi‑agent.

Block III. Management and people

Trend 8. From experiments to managed deployment

In 2026 the key shift is from pilots to discipline.

ETR notes that enterprise AI has “shifted from experimentation to execution”: companies are narrowing access (fewer “seats”) but increasing spend on targeted capabilities, strengthening governance and ROI focus.

As one CISO on the ETR panel puts it: “Companies are buying less access to AI and more capabilities.”

ROI is not a slogan, but an obligation

A multi‑year Wharton WHAIR study (800 executives, companies with $50M+ revenue) shows:
82% of leaders use GenAI weekly (up from 37% in 2023).
72% of companies track ROI metrics: productivity, profitability, throughput.
3 out of 4 already see positive returns on early investments.
88% plan to increase GenAI spending in the next 12 months

Importantly, this does not mean “guaranteed ROI for any process.” Winners are those who know how to measure impact, shut down “empty” use cases, and scale proven ones.

Minimal AI‑governance loop for 2026

To prevent AI from turning into “shadow IT”, you need:

Roles: process owner, data owner, model owner, quality curator, InfoSec/compliance.
Autonomy policies: what AI can do without a human, and what requires explicit approval.
Logging and audit of all AI actions.
Monitoring of quality, drift, and incidents.

Trend 9. Human–AI synergy (70/30)

AI augments people, but does not remove responsibility. A practical rule of thumb: AI takes ~70% of the routine (collection, structuring, drafts), humans own ~30% – context, constraints, and final decision.

But Wharton highlights a new risk: skill atrophy. 43% of leaders warn about skill degradation, even though 89% believe GenAI overall enhances work.

Management takeaway: winners are companies that formalize the roles “AI helps – human is accountable,” set clear guardrails, quality metrics, and control loops.

What to do

Define boundaries for 2–3 processes: what AI does without approval, what is always checked by a human.
Add a rollback mechanism: a “stop” button and an error log.
Build AI‑literacy into leadership development programs.

Three quick pilots by function

Sales: call analysis, reasons for churn, prompts for sales reps.
Finance: document classification, variance checks, AR dashboard.
HR: assistant for policies, onboarding automation, answers to FAQs.

Typical mistakes

No process owner – AI “hangs” between IT and the business.
No quality metrics – impossible to prove value and defend budgets.
InfoSec brought in at the end – blocks, rework, and lost time.
No change management – resistance from people eats up all the benefit.
“AI for AI’s sake” – no link to economics and specific processes.
Sources
Wharton Human‑AI Research (WHAIR), GBK Collective. “82% of Enterprise Leaders Now Use Generative AI Weekly, Multi‑Year Wharton Study Finds, as Investment and ROI Continue to Build.” BusinessWire, October 28, 2025.

https://www.businesswire.com/news/home/20251028556241/en/82-of-Enterprise-Leaders-Now-Use-Generative-AI-Weekly-Multi-Year-Wharton-Study-Finds-as-Investment-and-ROI-Continue-to-Build

ETR (Enterprise Technology Research). “Enterprise AI Trends 2026: How Leaders Measure ROI and Risk.” ETR Research, February 4, 2026.

https://research.etr.ai/etr-data-drop/enterprise-ai-trends-2026-how-leaders-measure-roi-and-risk?hsLang=en

Gartner (Anushree Verma, Sr Director Analyst). “Gartner Predicts 40% of Enterprise Apps Will Feature AI Agents by 2026.” Gartner Press Release, August 26, 2025 (updated September 5, 2025).

https://www.gartner.com/en/newsroom/press-releases/2025-08-26-gartner-predicts-40-percent-of-enterprise-applications-will-be-integrated-with-task-specific-ai-agents-by-2026