The Human Clipboard: Closing the Operational Gap in Legal AI

By Katja Nikolaus, Co-Founder, JUNE.

Few technologies have been adopted by legal departments as quickly as generative artificial intelligence. In an international survey by ACC and Everlaw from 2025, 53% of inhouse lawyers said they already use generative AI in their legal work. In Europe, the figure was 61%. The share of those who neither use generative AI nor plan to now stands at 2%. It is used most often for drafting (73%), followed by legal research (53%) and communication and collaboration (51%).

Yet anyone observing the teams that handle complex transactions or large-scale proceedings will encounter a striking bottleneck. Despite substantial investment in AI licences and case management infrastructure, lawyers still rely on a tool introduced five decades ago: the clipboard (Ctrl + C / Ctrl + V).

Lawyers craft precise arguments in the chat window and then spend ten minutes manually pasting paragraphs into Word, adjusting templates, matching case references, filing documents in the document management system and entering deadlines into the calendar by hand.

This friction stems from a structural divide in the enterprise legal tech market: the gap between individual intelligence and process intelligence.

A Thomson Reuters Institute survey shows the same gap. Generative AI has become firmly established in legal work, while agentic systems, those that carry out work steps autonomously, are at an early stage: fewer than 20% of respondents from the legal sector report widespread adoption. Around half are planning or considering deployment. 47% consider it appropriate for legal work, 22% disagree.

Two types of legal AI

A closer look at this gap reveals two distinct AI profiles:

1. Individual AI (the cognitive multiplier): This support sits alongside the individual case handler. It sharpens arguments, reviews clauses, drafts documents and summarises lengthy judgments in seconds. At the same time, it operates in isolation from the live file. It works without knowledge of the procedural history, without a connection to served pleadings, without insight into party roles and detached from the firm’s deadline calendar.

2. Process AI (case management): This system guides the entire team through complex litigation, mass claims and portfolio mandates. It captures incoming mail, extracts metadata, applies firm rules, orchestrates workflows and monitors deadlines. Case management platforms today also include numerous AI functions as a matter of course, such as extraction, semantic search or operational agents, always with a focus on making file handling more efficient. Operational process control follows predominantly defined rules and workflows, while linguistic nuance or unexpected counter-arguments still require legal judgement.

Large legal organisations, however, need both components. Because they have so far been developed as separate product categories, they often still run side by side in silos.

The case handler bridges these systems manually, day in, day out. Every handover between individual analysis and team-based process execution requires tedious intermediate steps: copying, reformatting, checking and filing. The time gained through text generation is quickly consumed by administrative tasks.

The limits of read-only access

Previous integration approaches focused primarily on retrieving information. Through retrieval-augmented generation (RAG) and search connectors, assistants gained the ability to search data repositories and summarise documents.

Reading alone, however, solves only half the task.

An active legal matter is a dynamic procedural state rather than a static PDF archive. In demanding or high-volume proceedings, considerable effort arises along the entire process chain: capturing, reviewing, reconciling, assigning and formally entering events into the file.

A conversational assistant speeds up legal review, yet a real productivity lever for the organisation only emerges when actions can be executed directly in the system of record:

– Filing a newly drafted pleading directly in the matter workspace.

– Creating a telephone note after a client call, automatically linked to the contact.

– Extracting court deadlines from orders and entering them directly in the firm’s calendar.

This shifts the question from reading to writing, and that is precisely where the legal sector’s reluctance begins. In the same survey, respondents name oversight and control as the second most common reservation about systems that carry out work steps autonomously. The objection is justified, because a file has evidential weight, and an entry in it is more than a data record. Write access therefore remains bound to clear governance standards:

– Privilege and ethical walls: Large law firms and legal departments operate under strict access restrictions. At all times, a model may hold only the rights held by the respective user.

– Audit-proof records and traceability: Because a file has evidential weight, automated actions require a complete log of who initiated an action and when.

– Controlled drafting workflows: Responsible lawyers retain final approval over official statements and pleadings. Write operations therefore follow a draft-first principle: they create versioned drafts that await approval by the responsible lawyer before they enter the official file.

Open protocols replace proprietary interfaces

In the past, connecting these systems required complex interface projects. Such point-to-point connections between individual document management systems and AI services were costly and maintenance-intensive.

The Model Context Protocol (MCP) provides a common foundation for this. As an open standard for connecting language models directly to external software, MCP separates the intelligence layer from the operational data layer.

Instead of tying teams to a single proprietary interface, an MCP server standardises both resources (read access to context) and tools (the execution of defined actions).

From silo to platform and from platform to assistant: outside legal tech, the direct connection of AI assistants to operational systems is already a reality. In enterprise process automation, agents can, for example, send emails, read and modify company data or post messages in collaboration systems. In software development, coding agents analyse repositories, make changes and create pull requests for human review.

Platform providers have now made this openness a product decision. Since May 2026, ServiceNow has used Action Fabric to make workflows, approval chains and business rules available via an MCP server to any agent, whether built on ServiceNow, with Claude, with Copilot or in-house; according to the provider, every action is identity-verified, permission-scoped and fully auditable.

Since April 2026, Salesforce has offered more than 60 MCP tools through Headless 360, allowing agents to execute platform functions without a user interface. Since July 2026, Salesforce’s own MCP server has also been available in beta.

The opposing position is equally well documented: SAP restricts the use of its interfaces by (semi-)autonomous or generative AI systems that plan, select or execute sequences of API calls, where this takes place outside the conditions set by SAP.

For legal departments and law firms, it is therefore foreseeable that the architecture of the core legal system will increasingly determine which external AI assistants can access its data and functions.

For legal tech, this development can be described as a two-stage shift: first, isolated applications, processes and data sets were consolidated into central platforms. Now the next stage follows: connecting these platforms directly to the user’s preferred AI assistant.

This architecture can be integrated into existing enterprise security frameworks. Through appropriate authentication and authorisation mechanisms, access can be limited to the permissions of the respective user and the company’s security policies.

The integrated legal workflow in practice

Once the individual assistant can securely access the relevant functions of the case management platform via open protocols, a large share of these manual handovers can be eliminated.

A typical pleading is produced in one continuous process:

1. Providing context: Via MCP, the assistant retrieves the current parties, the case reference and the appropriate template directly from the platform, without the lawyer having to search manually for old documents.

2. Individual drafting: In their familiar writing environment, for example Word or Claude, the lawyer instructs the assistant to tailor the argument precisely to the facts of the case.

3. Direct execution: With an instruction such as “File this draft in matter 19/2026 and create a follow-up for Friday”, the assistant calls the corresponding MCP tool. The platform takes over the record, files the draft as a new version in the correct dossier, updates the calendar and asks the lawyer for a brief confirmation.

Legal strategy remains entirely in human hands, while the system handles the administrative filing in the background.

The strategic shift in IT infrastructure

For CIOs, general counsel and legal operations leaders, the focus when evaluating enterprise AI is shifting away from pure model benchmarks towards one central question: does our infrastructure give our AI models secure, bidirectional access to our leading systems of record?

Current studies and industry forecasts point in the same direction. Thomson Reuters’ Future of Professionals Report 2025’ describes the adaptation of workflows and processes, together with the interplay of strategy, governance and individual accountability, as key prerequisites for successful AI adoption.

For 2026, industry forecasts also point to a shift from isolated AI point solutions towards integrated platforms, ecosystems and agent-based workflows. What will matter is no longer simply what an AI tool can produce in isolation, but how securely it can connect to existing systems of record, teams and operational processes, and how transparently those interactions can be traced.

Modern platforms already put this approach into practice. JUNE, whose platform manages complex and mass proceedings for organisations such as Lufthansa, Swiss Life and CMS, makes its architecture available via MCP. With more than 50 case management functions that can be called directly from assistants such as Microsoft Copilot, Claude or ChatGPT, the platform connects legal drafting with central file management, while fully preserving role concepts and human approval steps.

The first phase of legal AI gave lawyers rapid access to text generation. The next phase connects this cognitive capability directly with operational case files.

Only controlled write access to the file makes the detour via the clipboard unnecessary and lays the foundation for end-to-end system integration.

To learn more about what JUNE can do for you see here.

–

About the author:

Katja Nikolaus is Co-Founder and Chief Business Development Officer (CBDO) at JUNE. She supports large law firms and enterprise legal departments in introducing scalable legal tech architectures, process automation and modern working methods for complex or high-volume proceedings. JUNE’s clients include Lufthansa, Swiss Life, CMS and Pinsent Masons.

—

[ This is a sponsored thought leadership article by June for Artificial Lawyer. ]