The concept of boundaryless performance reflects how AI is not just supporting performance but transforming its very nature. With technological advances in AI architecture, conversational interfaces, and knowledge management, we’re entering an era where support doesn’t just complement work, it becomes indistinguishable from it.
In part one of this series on boundaryless performance, we’ll explore how AI architectures enable this shift and consider implications for organisations and workers.
Performance support: A brief history
Performance support in the workplace has a rich history dating back to the 1960s, when practitioners first used flowcharts to analyse performance problems and decide whether workers required job aids. This journey from manual job aids to electronic performance support systems – from programmed instruction to workflow learning – reflects our ongoing quest to support employees at their moment of need.
The workplace performance landscape has seen continuous evolution, but what we’re witnessing now represents something more profound. Taking place is a fundamental shift where the lines between learning, support, and performance are dissolving entirely. Organisations possess vast knowledge that often remains trapped in silos, documents, and individual expertise. What AI-enabled approaches offer is the capability to apply knowledge seamlessly in the flow of work itself.
We’re now entering the era of boundaryless performance, where AI is not just supporting performance but transforming its very nature.
The blurring boundaries
Think about how workplace support has traditionally functioned. You encounter a challenge, pause your work, seek information, then return to your task. This process creates a natural boundary between ‘doing the work’ and ‘getting help’ that’s increasingly disappearing.
Traditionally, we’ve distinguished between three distinct activities:
- Training/learning (acquiring capability).
- Performance support (accessing information when needed).
- Performance (executing the work).
These categories made sense in a world where information access required physical movement or context switching. But the boundaries between these activities have been eroding for years, and AI is now accelerating this convergence.
Consider how certain workplace tools have evolved. Technologies that were once distinct support mechanisms have gradually become so integrated into workflows that we no longer think of them as separate ‘support tools’ but simply as part of how we work.
The distinction often comes down to whether something requires a context switch. Tools that interrupt workflow remain ‘support’ while those that blend seamlessly into execution become part of performance itself.
AI assistance is following this same trajectory: from separate tool to integrated capability. When virtual assistants become so seamlessly integrated that users don’t perceive them as separate entities, the traditional boundaries between support and performance collapse entirely.
Jay Cross, who envisioned Workflow Learning back in 2004, described a world where learning and work were inseparable. He likened it to service-oriented architecture, where ‘smart’ software would guide workers by connecting them with the right information or expert. Cross’s vision, once aspirational, is now becoming reality through AI integration. As we half jokingly noted, he was always right, but he’s ‘becoming more right’ as technology catches up to his foresight.
Traditional performance support systems waited for users to identify a need and seek help. Emerging boundaryless performance systems flip this paradigm, proactively engaging workers with timely information.
The triage model: An architectural foundation for next-generation performance
Leading organisations are now implementing what we call the ‘triage model‘: a unified user interface that connects to multiple specialised AI capabilities behind the scenes. This architectural approach represents a significant advancement over traditional chatbots or standalone AI assistants.
In pioneering organisations, the approach looks deceptively simple from the user’s perspective: a single chat interface provides access to all support needs. Behind this simple interface lies a sophisticated system that routes queries to appropriate specialised models based on the nature of the question and the user’s permissions.
What makes this triage model particularly powerful is the specialisation of the underlying models. Rather than trying to create a single AI that specialises in ‘everything’, the triage model connects users to AI systems with deep expertise in specific domains: sales, marketing, regulatory compliance, product specifications, and more.
In practice, as users move between topics in a conversation, different specialised systems handle their queries, but they have no idea this is happening. The system creates seamless handoffs between specialised models during a single conversation flow. For example, if someone suddenly shifts from a marketing question to a regulatory one, the system passes the query to the regulatory knowledge base and then returns to the marketing context when appropriate.
This approach offers several critical advantages:
- Precision: Each model can be deeply specialised in its domain, maintaining accurate information without being diluted by data from other areas.
- Modularity: Updates to one domain don’t risk contaminating models handling other domains.
- User experience: The complexity is hidden from users, who experience a single, coherent conversation.
This approach mirrors how human expertise works in organisations. We don’t expect every individual to know everything; we connect people to appropriate subject matter experts as needed.
Hybrid approaches: Technical implementation for security and compliance
While the promise of AI-enabled performance support is compelling, organisations face significant implementation challenges, particularly around security, accuracy, compliance, and reliability. The solution emerging in many leading organisations is a hybrid approach that combines the strengths of both managed chatbots and large language models (LLMs).
Traditional pre-programmed chatbots offer security and reliability but are limited in scope. Comparatively, pure LLMs provide remarkable flexibility but raise concerns about accuracy and data security. The hybrid approach combines these technologies in creative ways, using AI to rapidly program traditional managed chatbots with more comprehensive response capabilities while maintaining security advantages.
This approach allows managed chatbots to consult LLMs for specific tasks without sharing sensitive context. When a user phrases a question in a complex way, the managed chatbot can send just that question to an LLM for rephrasing into simpler language, without revealing who asked it or why. This offers significant compliance benefits, particularly around GDPR and other privacy regulations, as user data can be removed from the managed chatbot without leaving traces in the LLM.
Making this approach truly boundaryless is the disappearance of the line between ‘being trained’ and ‘doing the work’.
Proactive vs. reactive: How boundaryless performance support manifests for users
Perhaps the most visible shift for employees is the evolution from reactive to proactive support. Traditional performance support systems waited for users to identify a need and seek help. Emerging boundaryless performance systems flip this paradigm, proactively engaging workers with timely information.
The scheduled check-in concept exemplifies this shift. Instead of waiting for employees to remember to look for updates, AI assistants initiate contact at strategic times. Advanced systems can proactively inform employees about process changes, explain their rationale, outline implications, and describe required adaptations. This transforms information flow from passive documentation to conversational updates, from ‘pull’ information seeking to ‘push’ information delivery.
Voice interfaces enhance this experience further, making interactions more natural and reducing friction. These conversations create valuable audit trails that document when information was shared and how it was received. This is a significant advantage in regulated industries.
Making this approach truly boundaryless is the disappearance of the line between ‘being trained’ and ‘doing the work’. Learning isn’t a separate activity that happens before performance. It’s integrated into the performance itself, delivered precisely when needed in a conversational format.
Boundaryless performance: where are we now?
Boundaryless performance isn’t a binary state but a spectrum. Organisations are at different points in the journey, with the most successful implementations focusing on niche, specific domains rather than generalised systems.
In part two of this series on boundaryless performance (due to be published on 25 March 2025), we’ll examine real-world use cases, unsolved challenges and future implications.
