iManage forecasts AI reality check & data overhaul by 2026

iManage has set out a series of predictions for how enterprise use of artificial intelligence will change in 2026, with a sharper focus on limits in agentic tools, more formal approaches to hallucination risk, and a shift in data structure and user interfaces.

Jan Van Hoecke, VP AI Services at iManage, said companies will draw clearer lines between systems that act independently and products that package automation inside fixed workflows. He also said enterprises will adapt their governance as they accept that today's large language models still produce errors that cannot be eliminated through incremental improvements.

Agentic reality

Van Hoecke expects a market correction around "agentic AI" as organisations compare expectations with what they observe in production deployments. He said enterprise buyers will start to distinguish between autonomous agents and products that follow predetermined sequences.

"2026 is the year when agentic AI will get a reality check, as the gap between marketing promises made in 2025 and their actual competencies will become starkly visible," said Jan Van Hoecke, VP AI Services, iManage.

He described many current offerings promoted as agents as "rigidly programmed systems" that follow preset paths. He compared the stage of development with early autonomous driving systems that could maintain a lane under constrained conditions. He said true autonomy remains out of reach in the near term.

"Currently, many products promoted as AI agents are, in reality, rigidly programmed systems that simply follow predefined paths," said Van Hoecke.

Hallucination management

iManage also expects enterprises to move away from treating hallucinations as an exceptional crisis and towards standardised controls and risk mitigation. Van Hoecke said organisations will take more responsibility for reducing the impact of mistakes rather than waiting for vendors to solve the issue.

"In 2026, the AI hallucination crisis will reach a critical juncture as organisations realise they must learn to coexist with the current fundamentally imperfect technology - until a new technology comes into play that can effectively address the issue," said Van Hoecke.

He said businesses will adopt a range of tactics, including additional checking and human review for high-stakes decisions. He also raised the prospect of insurance policies focused on hallucination-related risk.

"The focus will shift from AI hallucination 'crisis' to management," said Van Hoecke.

The comments also point to a continuing debate on liability when AI tools produce inaccurate outputs. Van Hoecke said organisations will push for clearer documentation and statements of limitations from model providers and application vendors.

"As the industry deliberates who carries the liability for AI's mistakes and inaccuracies - the tool makers or the users - enterprises will stop waiting for vendors to solve the problem and take matters into their own hands," said Van Hoecke.

Data architecture

On information management, Van Hoecke said today's approaches to retrieval and generation perform well when users ask for facts but struggle when they ask for explanations. He highlighted Retrieval Augmented Generation as effective for "what" queries but limited for "why" and "how" questions.

He attributed that to how many systems structure knowledge, which can make it difficult to represent relationships across documents and data sets. He said specialised domains such as legal and professional services information make those limits more visible.

"There will be a fundamental shift in how data is structured for AI systems, driven by the limitations of current approaches in answering complex questions," said Van Hoecke.

He said organisations will increasingly rely on AI systems to structure data autonomously. He said machines will map relationships across data points at scale and reduce the human role in creating structure.

"Consequently, in 2026, with machines taking the lead, the method of structuring data will undergo a complete transformation, gradually eliminating the human role in creating structure, to reveal the business-critical interconnections across multiple data points," said Van Hoecke.

Middleware squeeze

Van Hoecke also forecast pressure on middleware providers that build custom applications on top of data platforms. He said platform owners that control both data and the interface layer will gain influence, as the value concentrates around access to data and insight.

"Given the essential link between data and AI, middleware companies that specialise in building custom applications layered on top of data platforms will begin to get pushed to the margins, forced to compete on niche features - while the core value of data and insight is captured by the platform owners," said Van Hoecke.

Interface shift

Another predicted change centres on how workers interact with software. Van Hoecke said standard chat interfaces will give way to user experiences generated dynamically for specific tasks. He framed this as a move away from fixed, vendor-designed interfaces towards temporary interfaces created for a single job.

"In 2026, the current traditional vendor-designed, standard AI chat-based user interfaces will transition to dynamically AI-generated task-specific user interfaces that adapt to users' immediate needs," said Van Hoecke.

He also said this approach responds to the complexity of feature-heavy applications. He used an example where a user states a goal and the system generates a short-lived "micro-app" for that task.

"Instead of searching through endless menus in an overstuffed application like Excel, the user will simply state their goal - "Compare the Q3 and Q4 sales figures for our top 5 products and show me a chart" - and the AI will instantly generate a temporary, purpose-built interface - a "micro-app" - solely designed for that one single task," said Van Hoecke.

He said the organisations that lead in these interface approaches will be those that control their data and manage storage and interface creation through AI systems.

"The AI organisations that will truly lead in providing such bespoke user interface-generating capability are those that possess and control their own data," said Van Hoecke.