Listen to today's episode, 22 October 2026.
LDNUCU edition, Thursday 22 October 2026
Learning expects results. Few can prove them.
71% of professionals expect their learning programs to improve on-the-job performance. Yet only 53% can actually prove it with data. When you factor in that 20% of organizations report having zero credible evidence of program impact at all, you uncover a severe vulnerability for learning departments today.
This is ldnucu.com, Learning and Development News You Can Understand, exploring the critical gap between learning expectations and verifiable outcomes, and how organizations are using new frameworks and technologies to bridge it. Also today: AI agents spreading with little oversight, and the institutions building the foundation beneath corporate learning.
The towers show the survey figures as reported: 71 expect performance gains, 53 can prove them with data, and 20% of organizations report zero credible evidence of program impact.
The survey
Ambitious expectations collide with operational barriers
The initial data comes from Instructure, creator of the Canvas learning ecosystem, which collaborated with HR Dive's studio by Informa TechTarget to survey 1,166 U.S. professionals across human resources, learning and development, technology, and senior leadership.
The findings reveal ambitious expectations colliding directly with operational barriers. 70% of those surveyed expect productivity gains from their learning programs, and 69% expect improvements in role readiness. Yet nearly half of these organizations point to siloed enterprise systems, reporting limitations, and a lack of workforce data as the primary reasons they cannot verify if those expectations are met.
- Who was asked
- 1,166 U.S. professionals across human resources, learning and development, technology, and senior leadership.
- What was measured
- What respondents expect from their learning programs, and whether their organizations can verify it with data.
- What it found
- 71% expect better on-the-job performance, 70% productivity gains and 69% role readiness. 53% can prove performance gains with data. 20% of organizations report zero credible evidence of program impact.
- What it cannot show
- Whether the programs actually improve performance. The figures capture expectations and reported ability to verify. The data comes from a learning technology vendor.
Completion is not capability
The easiest thing to track is not the thing that matters
Melissa Loble, Chief Learning Officer at Instructure, offered a precise critique of this situation. She pointed out that course completion metrics have been the standard currency of learning departments for decades simply because they are the easiest thing to track. A learning management system can easily record who finished a module. The problem is that completion metrics utterly fail to demonstrate actual skill acquisition or performance transfer on the job.
When an organization relies solely on tracking who clicked through to the final slide of a training module, they are measuring compliance and superficial engagement, but they are not measuring capability. This leaves learning and development leaders vulnerable during budget reviews. When the chief financial officer asks for the return on investment, learning leaders cannot tie their programs to operational metrics.
Relying purely on completions is like measuring physical fitness by tracking how many times you swipe your gym membership card at the front desk.You have data showing the person entered the building, but zero evidence they engaged in the work required to build strength: the weight you can lift, the miles you can run.
Climb the staircase of evidence
Choose an approach and see how far its evidence reaches. Then tap any step to read what the reporting says about it.
Completion records
A conceptual map of what each approach is reported to connect. It is not a performance comparison, and every vendor claim stays a vendor claim.
Vendor claims, kept as claims
Canvas Career and the risk of a new vanity metric
To solve this, vendors are pushing new technological architectures, such as Canvas Career, which attempts to connect skills data directly with Human Resources Information Systems and Customer Relationship Management platforms. The goal is to correlate learning engagement with business key performance indicators.
This sounds promising on paper, but automated skills tagging introduces a new risk: the creation of a new vanity metric. If an algorithm flags an employee as possessing a specific leadership skill simply because they watched a 10-minute video series, that does not mean they can lead a team through a crisis. The data might look sophisticated on a dashboard, but it remains disconnected from real-world human capability. Translating a digital tag into observed human behavior is the central challenge.
Schoox and the frontline
The broader learning technology market is responding to this demand for verifiable outcomes by restructuring how platforms operate. This shows clearly in recent industry recognitions. Schoox was named Best Extended Enterprise LMS in the 2026 Talented Learning LMS Top 10 Awards. Extended enterprise systems train people outside direct corporate payroll, such as franchise owners, supply chain partners, and independent contractors.
Schoox's approach specifically addresses the disconnect highlighted by the gym analogy. Instead of merely tracking course completions, they map role-based frontline skills directly to operational metrics. For example, if a retail worker completes a conflict resolution module, the system does not just log a completion certificate; it attempts to correlate that employee's training record with branch customer satisfaction scores or product return rates over the subsequent quarter, tying learning data to concrete business outcomes.
Meridian and the auditor's standard
That focus on operational metrics aligns with another major industry recognition: Meridian Knowledge Solutions winning the 2026 Training Magazine Network Choice Award for Learning Management Systems. This win underscores enterprise demand for robust, compliance-oriented tracking architecture, especially in highly regulated sectors like pharmaceutical manufacturing or financial services. In these environments, tracking cannot be an approximation of skills based on video views; it must be a verifiable system that proves operational readiness and regulatory adherence to an external auditor. The market is shifting away from centralized course catalogs toward decentralized systems that integrate directly into the daily operational flow to measure actual output.
The next problem
Yet once organizations begin to measure whether an employee can execute required tasks, a new problem emerges: what happens when the tasks themselves completely change?
Governing the agent
From doing the task to setting the boundary
The workplace is shifting from teaching employees how to execute specific tasks to teaching them how to orchestrate autonomous systems. The Project Management Institute announcements from their Global Summit in Detroit illustrate this transition. PMI launched the Advanced Project Leader (APL) credential, designed for senior professionals managing complex, high-consequence initiatives. Alongside this, they announced a partnership with Prosci, a global leader in change management, focusing on AI adoption and organizational transformation.
PMI research explains why this intervention is urgent: 81% of project professionals report increased project complexity in recent years, and 80% of complex projects experience negative fallout from that complexity. To address this, PMI introduced more than conceptual training; they launched the PMI Foundations: Agentic AI for Project Professionals course and Build Lab.
Agentic AI differs fundamentally from traditional chatbots. While a standard conversational assistant functions like a reference tool that answers prompts, an AI agent operates more like an autonomous assistant: given a high-level goal, it navigates corporate systems, makes independent procedural decisions, and executes multi-step projects across networks.
In PMI's Agentic AI Build Lab, learners do not passively watch lectures about artificial intelligence. They build and evaluate an autonomous agent using platforms like Claude or Microsoft Copilot Studio, learning to scope the agent, establish guardrails, and govern its actions within a controlled environment.
Scope the agent
Switch the assistant into an agent, then set the boundary. Everything starts open.
What it does
Guardrails
A conceptual model of the skills PMI's Build Lab describes. It does not predict what any real agent will do.
This hands-on governance lab arrives at a critical moment. PMI cited data from the SAP LeanIX Agentic AI Survey showing that 98% of organizations have deployed or plan to deploy AI agents, yet only 17% have visibility into agent performance, and 48% lack clear governance roles. Widespread deployment of autonomous systems with minimal oversight makes training professionals to govern these tools an urgent risk management priority. If a project manager cannot establish clear operational boundaries for an AI agent, that system could access sensitive personnel files or alter production code.
Source: SAP LeanIX Agentic AI Survey, cited by PMI.
At the same time, the credentialing model PMI is using for the Advanced Project Leader certification presents a scalability challenge. The APL credential relies on accredited organizational partners benchmarking internal corporate programs against PMI standards. Maintaining assessment rigor across disparate global organizations is difficult; an internal academy at a financial institution in London operates with different risk tolerances and workflows than a manufacturing firm in Tokyo. Ensuring consistent capability across decentralized internal academies will test the framework.
Different risk tolerances and workflows. Same assessment rigor?
The workaround
When training stays abstract, people build their own route
The consequences of poor governance are evident in research recently published by CYPHER Learning, which surveyed 400 independent agents across real estate, finance, and insurance. The data reveals an acute lack of governance across distributed workforces, where employees adopt tools rapidly while receiving training that remains disconnected from daily tasks.
The behavioral findings from CYPHER Learning's study are alarming: 88% of surveyed agents have sent AI-generated content directly to clients without editing it first, and 57% do so usually or always. Furthermore, 79% turn to external public tools like ChatGPT for work guidance rather than using official corporate resources.
When professionals forward unedited machine-generated text directly to clients, the risk of factual errors, brand damage, and legal liability increases dramatically. Only 31% of surveyed agents feel confident that their AI use complies with legal and regulatory mandates.
This risky behavior stems directly from inadequate instructional design. While 97% of surveyed agents reported that their affiliated firm provided some form of AI training, 55% labeled that training as broad or generic. When companies offer high-level conceptual webinars instead of practical guidance on daily workflows, employees bypass official channels and create their own workarounds using unvetted public tools.
One hundred agents
Pick a finding and see how many of every 100 surveyed agents it describes.
The survey covered 400 independent agents. Percentages are redrawn here as a share of 100 people. All figures are self-reported by surveyed agents.
Move the guidance into the work
Broad, generic AI literacy training fails to protect organizations from compliance and reputational risk because it leaves workers operating without practical boundaries. CYPHER Learning is addressing this by embedding role-specific training directly into operational platforms through a partnership with Bright MLS, the nation's largest multiple listing service. Instead of requiring an agent to log into a separate learning portal for abstract compliance rules, practical guidance on using AI safely to draft property descriptions is embedded directly inside the software where daily work occurs.
The foundation moves outside
Who teaches the baseline before the worker arrives?
As corporate learning departments struggle to build practical technological capability internally, the responsibility for foundational skills is expanding beyond corporate boundaries. Public institutions and community frameworks are stepping forward to establish baseline digital and AI literacy before an employee ever enters an enterprise.
Microsoft Elevate and the Public Library Association have launched a national initiative, Building AI Skills and Agency Through Libraries, funding no-cost AI curriculum, webinars, and professional Communities of Practice for public library staff. Libraries frequently serve as the first point of contact for community members navigating technology shifts. Through these Communities of Practice, library workers collaborate on helping patrons use AI tools safely for job searches and research while developing critical evaluation habits. Microsoft's TechSpark program reports having upskilled 59,600 individuals in AI literacy over the past three years.
A similar public-sector model is operating internationally. Efekta Education is expanding a major initiative with the Rwandan Ministry of Education, using AI-powered English language instruction to train the country's entire public teaching workforce, reaching up to 150,000 beneficiaries. This is a system-wide workforce capability initiative that simultaneously improves language proficiency and normalizes advanced educational technology across public infrastructure.
Private capital is also targeting the underlying algorithmic architecture that powers adaptive learning. South African education group AdvTech recently acquired a 25% equity stake in MathU, a developer of adaptive algorithms that diagnose cognitive gaps and personalize instructional paths in real time. Rather than presenting static content, adaptive systems identify the precise cognitive hurdle causing learner error and dynamically generate tailored instructional pathways.
Lay a footing
Tap each foundation to place it under the structure and watch what the corporate layer no longer has to build alone.
Nothing laid yet
With no outside foundation, the corporate layer has to teach basic digital literacy from scratch.
The reading of this pattern is the edition's. No measured effect on corporate training demand is reported.
Across Microsoft, Efekta, and AdvTech, a clear pattern emerges: enterprise education providers and public institutions are investing in proprietary diagnostic algorithms and baseline literacy. Community institutions and national programs are establishing foundational evaluation skills before workers enter formal corporate talent pipelines, reducing the need for corporate L&D to teach basic digital literacy from scratch.
The tension of the day
Teach the machines, or teach the humans to set boundaries?
This brings us to the core tension of the day for workplace learning: the friction between the rapid, decentralized adoption of autonomous AI tools by the workforce and the ongoing struggle of learning and development departments to govern these systems and measure verifiable human performance.
As platforms become more sophisticated at diagnosing skills gaps automatically, the defining question for the profession becomes clear: will the future of learning and development focus more on teaching machines to navigate corporate workflows, or teaching humans how to set boundaries for the machines?
Teaching machines to navigate corporate workflows
98% of organizations have deployed or plan to deploy AI agents.
Agents navigate corporate systems and execute multi-step projects across networks.
Adaptive algorithms diagnose cognitive gaps and personalize instructional paths in real time.
Teaching humans how to set boundaries for the machines
Only 17% have visibility into agent performance, and 48% lack clear governance roles.
PMI's Build Lab teaches learners to scope an agent, establish guardrails and govern its actions.
88% of surveyed agents have sent AI-generated content to clients unedited, and 31% feel confident their AI use is compliant.
Balanced. The edition does not settle the question: the same reporting supports both emphases.
Completion records show who finished a module. They do not show skill acquisition or performance transfer on the job.
Agent deployment is widespread, visibility is not: 98% deploy or plan to, and 17% can see agent performance.
Where AI training is broad or generic, surveyed agents report workarounds: 55% call it generic, and 79% turn to public tools.
Listen on YouTube Podcasts