Does AI need more prompts? Nope, it needs orchestration.
By Bart van Wanroij
Why connected systems beat better prompts
For the past two years, the legal AI conversation has revolved around a single question: which model performs best? Firms compared outputs, experimented with prompts, and evaluated tools based on drafting quality or summarization accuracy. That phase was inevitable, but it is already becoming less decisive. Once language models reach a certain level of competence, marginal improvements matter far less than the environment in which they operate.
According to our specialist, Bart van Wanroij, the next leap in legal productivity will not come from better text generation alone. It will come from orchestration. As he says: “The ability of AI to coordinate workflows across systems, knowledge sources, and structured data.”
Prompts were an important starting point, but agents represent the next stage. Beyond that lies a world where systems interact with one another through AI-driven processes. Turning what is currently a document assistant into something closer to a workflow engine. This changes how legal leaders should think about AI adoption. The focus moves away from isolated tools toward the architecture that connects them.
From generating text to coordinating work
Most legal AI usage today still concentrates on documents. Think about: summarizing contracts, comparing versions, drafting clauses, or extracting information. These capabilities deliver measurable efficiency gains, yet they rarely alter the underlying workflow. Lawyers continue to navigate between repositories, knowledge systems, financial data, and matter files manually, stitching together context before producing an output.
But for Bart transformation begins when that navigation layer becomes automated. Consider a typical scenario inside a corporate legal team or law firm. A claim or dispute arrives. The team reviews the contract repository, checks similar historical matters, verifies payment or policy data, evaluates risk exposure, and prepares a response. Each step depends on different systems, different data structures, and human coordination.
An AI agent can orchestrate that sequence in advance. Rather than replacing legal judgment, it prepares the groundwork by gathering context, surfacing relevant knowledge, and assembling a structured starting point. The lawyer enters the process later, focusing on analysis and decision-making rather than information hunting. This is where AI begins to influence productivity at a structural level instead of a task level.
When systems expose what they can do
Behind the scenes, technology vendors are moving toward standardized ways for systems to communicate their capabilities to AI workflows. The underlying concept is straightforward even if the terminology sounds technical. Instead of building rigid integrations for every scenario, systems can describe what information they contain and what actions they support. An AI agent can then query those capabilities dynamically.
In practical terms, that means an agent might:
- consult a matter management repository;
- retrieve precedent documents from a knowledge base;
- verify structured financial data, and;
incorporate those elements into a coherent workflow without requiring custom development each time. The orchestration layer becomes adaptive rather than fragile.
For legal organizations, this flexibility introduces both opportunity and responsibility. Greater automation increases efficiency, but it also raises questions about reliability, traceability, and control.
Why traceability becomes non-negotiable
Legal work operates under strict accountability requirements. If an AI-generated response references prior cases, internal policies, or template guidance, the organization must understand how that output was produced. Without transparency, speed becomes a liability.
This is why document and knowledge platforms are evolving beyond simple storage functions. They increasingly act as provenance engines, capturing the relationship between inputs, processes, and outputs. Instead of saving only the final document, systems preserve the underlying query, the source materials used, and the context in which the result was generated.
Opening a generated response could therefore reveal not only the text itself but also the structured question that initiated it and the knowledge sources that informed it. That level of traceability strengthens defensibility, improves knowledge reuse, and reinforces trust in AI-assisted workflows.
Connecting orchestration
to the foundation
The orchestration concept connects directly to themes raised by other voices from our company. Bart Bogaerts emphasizes that AI effectiveness depends on a strong home base: the environment where data resides securely and context can be accessed. Marcel Lang highlights internal knowledge as the primary differentiator between organizations that merely use AI and those that gain strategic advantage.
Orchestration brings those ideas together. Once agents can move across systems, the quality of the underlying environment becomes decisive. Permissions must align, knowledge must be curated, templates must follow standards and governance controls must be embedded rather than added later. If those foundations are weak, orchestration amplifies inconsistency. When they are strong, orchestration multiplies value.
AI inside the document experience
Another evolution is emerging alongside orchestration. Until recently, AI largely operated around documents, generating or analyzing content externally before users returned to traditional editing tools. Increasingly, assistance will occur within the document environment itself.
Imagine drafting a response while an embedded assistant evaluates structure, checks alignment with templates, identifies omissions, and suggests improvements in real time. Rather than functioning as a separate application, the assistant becomes a contextual collaborator integrated into daily work.
As systems gain awareness of organizational knowledge and workflows, outputs become less generic and more tailored to institutional practices. The technology fades into the background, while the quality and consistency of work improve.
The strategic question for legal leaders
The most important decision facing legal organizations is therefore no longer whether to experiment with AI. The real question is whether their ecosystem is prepared for orchestration. Success depends on aligning several elements into a coherent environment:
- Content platforms
- Knowledge management structures
- Permission models
- Governance frameworks
- Agent-driven workflows
This alignment requires architectural thinking rather than tool acquisition. It involves decisions about data location, integration strategy, and governance maturity that extend beyond any single vendor.
Architecture determines advantage
AI adoption is accelerating, but competitive advantage will not be determined by who deploys the most tools. It will depend on who builds environments where intelligence, context, and governance reinforce each other. Organizations that treat agents as structured infrastructure rather than experimental features are likely to outperform peers over time. Their systems will quietly enable faster preparation, higher quality outputs, and more consistent decision-making without increasing risk exposure. The future of legal AI will not be defined by louder technology but by smarter ecosystems.
