For over a decade, enterprise learning has operated on a flawed assumption: that employees will voluntarily leave their workflow, navigate to a separate system, and dedicate focused time to learning. We called it "self-directed learning" and measured success by completion rates. But the data tells a different story - most employees interact with their LMS only when compliance forces them to. A 2026 TalentLMS report found that multitasking during training reached 70% - the highest in three years - while only 10% of employees report that compliance training has impacted their actual work practices.
This isn't a motivation problem. It's an architecture problem.
The Real Cost of "Destination Learning"
According to Asana's Anatomy of Work Index, knowledge workers switch between an average of 9-10 different applications per day and spend 60% of their time on "work about work" - chasing updates, attending unnecessary meetings, and toggling between tools. Every context switch carries a cognitive tax: research by Gloria Mark at UC Irvine found that it takes an average of 23 minutes and 15 seconds to fully resume a task after an interruption.
When we ask employees to "go to the LMS and learn," we're asking them to pay that tax twice: once to leave their work, and once to return to it. Meanwhile, McKinsey research shows that knowledge workers already spend nearly 20% of their workweek - a full day - searching for internal information or tracking down colleagues who can help.
The result is predictable: learning gets postponed until it becomes mandatory, consumed in a distracted state, and forgotten quickly. As Ebbinghaus's forgetting curve demonstrates - and as modern replications confirm - without reinforcement, people forget roughly 50% of new information within an hour and up to 70% within 24 hours. One-time learning events delivered in a context-free environment are fighting against basic neuroscience.
The L&D team reports high completion rates; the business sees no behavior change. Both are telling the truth. The fundamental question isn't "how do we get employees to learn more?" It's "how do we make knowledge available at the moment it's needed, without requiring a context switch?"
The Paradigm Shift: From Knowledge Repository to Knowledge Layer
Josh Bersin coined the phrase "Learning in the Flow of Work" back in 2018. But for years it remained aspirational - a concept with no technology to deliver it at scale. His 2022 follow-up research found that while 78% of companies regard L&D as a top C-Suite priority, only 12% effectively deliver learning in the flow of work.
What's emerging in 2025-2026 is the technology catching up to the concept. A new architecture for organizational learning is taking shape - one where the LMS stops being a destination and starts being an invisible layer on top of the employee's existing workflow. The employee never "goes to learn." Instead, relevant knowledge surfaces exactly when and where it's needed.
This shift has three key characteristics:
- Contextual awareness: The system understands what the employee is doing right now - which application they're in, which process step they're on, which customer they're serving - and pushes relevant knowledge proactively.
- Zero-navigation access: Knowledge is consumed within the current workflow. No new tabs, no login screens, no searching through course catalogs. The friction approaches zero.
- AI-mediated delivery: Instead of browsing pre-structured content, the employee asks a question in natural language and gets a synthesized answer drawn from the organization's entire knowledge base - courses, policies, procedures, past decisions.
This isn't a feature upgrade. It's a fundamental rethinking of what "learning" means in an organization. Learning stops being an event (a course, a workshop, a module) and becomes a continuous, ambient capability - like electricity. You don't "go to the power plant." You flip a switch.
Who's Building This Future - and How They Differ
Multiple players are converging on this vision from very different starting points. Understanding where each comes from helps explain their strengths and blind spots:
The LMS vendors adding a workflow layer
Docebo Companion (public beta, April 2026) is a clear example: a browser side panel that surfaces AI answers and training content inside any web application the employee is using. The advantage: the learning management backbone already exists - courses, certifications, compliance tracking, reporting - and now gets a delivery mechanism that meets employees where they are. The MCP Server integration (GA July 2026) takes this further by making platform content queryable from Claude, Copilot, and ChatGPT natively.
The risk: These are still LMS-first tools. The contextual intelligence is new, and how well it actually understands workflow context (vs. just matching URLs to content) remains to be proven at scale.
April 2026 Release Notes
The AI-native platforms that started from search
Sana (acquired by Workday for ~$1.1B, November 2025) took the opposite approach - start with universal semantic search across all enterprise knowledge, then add learning management on top. Now rebranded as Sana Enterprise, it connects to Slack, Drive, Notion, Salesforce, and HRIS data. Josh Bersin called this merger "a bold new strategy for AI" in enterprise learning.
The risk: Brilliant at finding information, but weaker at the structured learning journey - compliance tracking, certification management, regulatory audit trails. Not every knowledge need is a search query; some require deliberate skill-building over time.
sanalabs.com
The digital adoption platforms (DAP) that own the UI layer
WalkMe (acquired by SAP, September 2024) and Whatfix don't teach knowledge - they guide behavior. Step-by-step overlays walk employees through complex processes in real time. Gartner's Digital Adoption Platforms market is growing rapidly for a reason: for procedural tasks, nothing beats a guided flow that physically shows you where to click next.
The risk: High maintenance cost - each flow must be rebuilt when the target application changes its UI. And this approach only works for "how to do X in this specific tool," not for conceptual understanding, judgment calls, or transferable skills.
walkme.com
The enablement tools that own a specific workflow
Spekit (named a Visionary in the 2025 Gartner Magic Quadrant for Revenue Enablement) takes a narrow but deep approach: its AI Sidekick understands CRM context, deal stage, and persona to surface the right playbooks and coaching inside the sales workflow. Degreed does something similar for skills development - aggregating learning into a unified skills profile with a browser extension for capturing knowledge from anywhere.
The risk: These tools solve for one function brilliantly (sales, skills) but don't generalize. You can't use Spekit for onboarding engineers or Degreed for compliance training in manufacturing.
The ecosystem giants betting on ubiquity
Microsoft Viva Learning embeds learning inside Teams and Outlook, with Copilot now offering AI-personalized recommendations. The Copilot Academy adds structured AI skill-building directly within the tools people already use.
The risk: Only works for Microsoft-centric organizations. If your frontline workers use Salesforce, ServiceNow, or industry-specific tools as their primary workspace, Viva Learning is invisible in the wrong places.
The Deeper Question: What Changes When Learning Is Invisible?
If we take this trend to its logical conclusion - knowledge available instantly, contextually, without friction - several things change fundamentally:
Completion rates become meaningless. When an employee gets a 30-second AI answer that solves their problem, did they "complete" something? The metric shifts from "hours consumed" to "problems solved" or "errors prevented." Most L&D teams are not equipped to measure this. As the TalentLMS L&D Benchmark 2024 found, 67% of employees stay at jobs that offer progression opportunities - but measuring "progression" through completion rates misses the point entirely.
Content architecture matters more than content production. In a traditional LMS, you can get away with messy metadata and poor tagging because humans browse visually. In an AI-mediated system, the machine needs to find and synthesize the right content programmatically. Poor tagging means wrong answers. Outdated content means dangerous answers. As Harvard Business Review research on information overload shows, the problem isn't lack of knowledge - it's poor knowledge organization. The L&D team's most critical skill becomes information architecture, not instructional design.
The LMS becomes a "Source of Truth" for AI agents. With protocols like MCP (Model Context Protocol) gaining adoption, your LMS isn't just serving employees directly anymore - it's serving the AI assistants that employees use. When someone asks Copilot "what's our refund policy for enterprise customers?", the answer should come from your verified, version-controlled training content - not from a hallucination. A 2026 HBR study found that cognitive fatigue from managing multiple AI tools is already a measurable problem - making accurate, trustworthy AI answers even more critical.
The skills gap becomes visible in real-time. When every knowledge request is logged - what employees ask, when they ask it, in which application, how often the same question recurs - you get a live map of organizational knowledge gaps. This is vastly more actionable than annual skills assessments or post-training surveys. According to Bersin's research, spending on corporate learning now exceeds $1,400 per employee per year - but without real-time gap visibility, most of that investment is directed by guesswork.
What This Means for Your Strategy
If you're an L&D leader evaluating this shift, here's what I'd recommend thinking about:
- Audit your content for "AI readability": Can your training content be meaningfully chunked, tagged, and retrieved by an AI system? Or is it locked in 45-minute SCORM packages with no granular metadata? The content itself might be fine - but its structure may be invisible to AI-mediated delivery.
- Map knowledge needs to workflow moments: Instead of asking "what courses do people need?", ask "at which moments in which applications do people get stuck, make errors, or need guidance?" The answers will reshape your content strategy entirely.
- Don't choose yet - but start experimenting: The market is still consolidating. The Workday-Sana merger, SAP-WalkMe acquisition, and Docebo's AgentHub launch all happened in the last 18 months. No single vendor has the complete answer. But the organizations that start structuring their knowledge for AI-mediated delivery now will have a significant advantage when the dust settles.
- Preserve deliberate learning for what it's good at: Not everything should be micro-delivered in the flow of work. Complex skill-building, mindset shifts, leadership development, collaborative problem-solving - these still need dedicated time and space. The goal isn't to eliminate the LMS as a destination; it's to reserve it for learning that actually requires immersion, and handle everything else invisibly.
Bottom Line
The shift from "go learn" to "learning comes to you" isn't a product feature - it's a paradigm change in how organizations think about knowledge. The technology is ready: contextual delivery, AI synthesis, browser extensions, MCP protocols. The question for most organizations isn't which tool to buy - it's whether their knowledge architecture is ready to be delivered this way.
The LMS isn't dying. It's disappearing - in the best possible sense. Like good infrastructure, the best learning system is one nobody notices they're using.
Research and resources cited in this article:
Gloria Mark, UC Irvine - The Cost of Interrupted Work (PDF)
Asana - Anatomy of Work Index 2023
McKinsey - The Social Economy: Unlocking Value Through Social Technologies
Replication and Analysis of Ebbinghaus' Forgetting Curve (PubMed)
Josh Bersin - Learning in the Flow of Work (2018)
Josh Bersin - Growth in the Flow of Work (2022)
TalentLMS - The State of Workplace Learning 2026
HBR - Reducing Information Overload in Your Organization
HBR - When Using AI Leads to "Brain Fry" (2026)
Model Context Protocol (MCP) Specification
Docebo Inspire 2026 Announcements
Josh Bersin - Workday and Sana's Bold AI Strategy (2026)