Bypassing the LMS: Bites Raises $40M as Workflow Learning Takes Over

The contradiction

Funding is flowing into AI training. The balance sheet is not moving.

Bites secured a $40 million Series A round on November 2, 2026 for its artificial intelligence frontline training platform. MIT's Project NANDA reports that roughly 95 percent of enterprise generative AI pilots show no measurable profit and loss impact. Between those two facts sits the question this edition keeps returning to: what, exactly, are organizations measuring?

A deck built from technology blocks extends from a tall pier of AI adoption toward a hollow pier of measurable impact. More technology lengthens the deck but a gap remains until leadership mindsets are placed.

Nearly 90%of enterprises use AI regularly in at least one business function.McKinsey

Roughly 95%of more than 300 enterprise generative AI pilots produced no measurable profit and loss impact.MIT's Project NANDA

Measurable impact on the bottom line

Adoption is wide. The span stops short of the bottom line.

Conceptual diagram. Adoption and pilot figures are as reported by McKinsey and MIT's Project NANDA, via an analysis in Chief Learning Officer.

Story one: frontline training

Thirty-second modules, and what a checkmark can verify

Bites secured a $40 million Series A round on November 2, 2026 for its artificial intelligence frontline training platform. The company built a system that ingests dense, historically ignored operating procedures, standard safety protocols and fifty-page compliance manuals, then uses generative AI to convert them into video modules running between thirty seconds and two minutes.

The delivery mechanism is as notable as the conversion. The platform bypasses the legacy corporate learning management system entirely, the portal where employees reset forgotten passwords every six months just to click next on a pixelated slide deck. Instead, Bites pushes modules through channels frontline teams already check during their shifts, primarily WhatsApp and Microsoft Teams. Bites announced that its platform generates completion rates three to five times higher than conventional enterprise systems.

Vendor claimThe three to five times figure is Bites's own announcement.

Solving that distribution bottleneck is a real operational achievement. Reaching a deskless worker on a factory floor, in a healthcare facility or across a retail environment has historically been an uphill battle. The standard still holds, though: course completion is not capability.

Follow one module from manual to workplace, then turn up the volume.

Illustrative volume of what a dashboard can count. Not data.

Select a step to see what it establishes.

What the tool does

Procedure converted

Generative AI ingests operating procedures, safety protocols and fifty-page compliance manuals and turns them into video modules of thirty seconds to two minutes. This is a capability of the tool. It does not show that the modules teach well.

What the tool does

Delivered where people already are

Modules arrive through WhatsApp and Microsoft Teams, bypassing the legacy learning management system and meeting deskless employees on channels they already check mid-shift. Reaching them at all has historically been the hard part.

Vendor claim

Marked complete

Bites announced completion rates three to five times higher than conventional enterprise systems. That is the vendor's claim, and it measures completion.

Verified, narrowly

Recalled for a minute

After a thirty-second video and a two-question micro-quiz, the organization has verified brief cognitive recall in that isolated minute, and that the worker was holding the device and interacting with the screen.

Not verified by a quiz

Done in the workspace

Behavioral execution in the physical workspace is the step a quiz cannot reach. For high-stakes procedural competence, on-the-job behavioral observation cannot be bypassed.

Not shown by completion data

Capability

Resilient organizational capability is what the whole chain is meant to build. A dashboard of green checkmarks can create the illusion of a trained workforce without showing it.

Conceptual diagram. Filled blocks mark what a dashboard can count. Dashed blocks mark what a completion count cannot show.

When a frontline worker watches a thirty-second video on WhatsApp and answers a two-question micro-quiz, the organization has verified brief cognitive recall in that isolated minute. It has verified that the worker was holding a phone and interacting with the screen. It has not verified behavioral execution in the physical workspace.

Consuming safety training should never feel identical to swiping through a social media feed.

Consider an employee driving a forklift in a crowded, high-volume logistics warehouse. Faster video consumption on their phone does not mean they have the procedural competence to operate safely and avoid driving into an industrial racking system. For high-stakes procedural work, automated digital quizzes cannot bypass the necessity of on-the-job behavioral observation.

A management dashboard showing a flurry of green checkmarks creates an illusion of a highly trained workforce. Mistaking the velocity of content delivery for the depth of cognitive acquisition and behavioral habit is the risk. It helps explain why so many enterprise AI initiatives are financially flatlining: organizations optimize for digital engagement and consumption volume rather than verifiable operational impact.

Story two: the adoption paradox

Adoption is nearly universal. Measurable return is not.

An analysis published in Chief Learning Officer examines what it calls the enterprise AI adoption paradox. Data from McKinsey indicates that nearly 90 percent of enterprises use artificial intelligence regularly in at least one business function. Tool adoption appears ubiquitous across sectors.

Yet when researchers at MIT's Project NANDA examined the financial outcomes of more than 300 enterprise generative AI deployments, roughly 95 percent of those pilots yielded no measurable profit and loss impact. Server costs and software licensing fees are climbing, yet nothing on the balance sheet is moving in a positive direction.

Nearly 90%of enterprises use AI regularly in at least one business function. McKinsey.
Roughly 95%of more than 300 enterprise generative AI pilots: no measurable profit and loss impact. MIT's Project NANDA.

The Chief Learning Officer analysis concludes that this gap is adaptive rather than technical. Closing it requires a new set of leadership mindsets, which the publication categorizes as the Flux 5 framework. It points to three areas learning leaders must prioritize.

First mindset

Inquiry: from prompting to interrogating

Right now most organizations are focused on prompt fluency, running workshops on prompt engineering and on how to converse with language models. The Inquiry Mindset moves well past prompt construction, to systematically interrogating model outputs and the premises behind them. The analysis advocates structured spaces called Judgment Labs.

In a Judgment Lab, operational teams examine real workplace artifacts generated by AI: a complex project plan, a piece of architectural code, a customer service resolution. Human experts evaluate the artifact. Teams compare the automated output against historical expert analyses and dissect it for missing contextual nuance, subtle hallucinations and unspoken operational assumptions.

The practice builds the organizational muscle to articulate and exercise human judgment. It trains employees not to accept the first generated answer, but to interrogate it for flaws, biases and contextual blind spots.

Put an artifact on the bench, then move the probe deeper.

1
Missing contextual nuanceWhat the output leaves out about the real situation.
2
Subtle hallucinationA detail that reads as fact and is not.
3
Unspoken operational assumptionA premise the output depends on without saying so.
PromptingOutputsPremises

Prompt fluency shapes what goes into the model. It does not examine what comes out.

Conceptual illustration. The artifact is abstract; the three flaw types are those the analysis describes for Judgment Labs.

Second mindset

Humanizing: protect the apprenticeship

This is one of the most urgent warning lights on the enterprise talent dashboard: the need to protect entry-level skill-building pipelines. Generative AI is remarkably effective at automating foundational corporate tasks. It drafts initial marketing copy, summarizes lengthy meetings and generates standard boilerplate code. That repetitive, often tedious work is also the developmental training ground where junior employees practice reasoning and build professional intuition.

If an organization automates away the struggle of learning the basics, it dismantles the pipeline that produces its future senior experts. A sous chef masters knife mechanics, kitchen timing and flavor profiles by chopping thousands of onions and preparing basic stocks day after day. If a machine does all of it, the apprentice never develops the tacit sensory feel needed to eventually run a kitchen as an executive chef.

Run a kitchen as executive chef
Tacit sensory feel
Knife mechanics
Kitchen timing
Flavor profiles

The apprentice builds the base by doing it. Each block rests on the one below.

The remedy the analysis describes is an institutional holding environment: scaffolding and safety mechanisms that let junior professionals experiment, struggle and learn alongside automated tools, rather than allowing software to replace the developmental process entirely.

Conceptual model of the apprenticeship pipeline described in the analysis.

Third mindset

Systems: when a faster step makes a bigger pile

The Systems Mindset asks leaders to map workflows deliberately and avoid local optimization cul-de-sacs. A business unit buys a specialized AI tool that makes one task exceptionally fast, and the team celebrates. But the wider organization was never re-engineered to handle that new velocity, so all they have done is create a bottleneck at the next stage of the workflow.

A marketing or product team might accelerate content generation fivefold. If the legal or compliance team still reads every word at human speed, the output simply piles up in an unreviewed backlog. The local win fails the larger operating system.

Content generationBefore the tool

Waiting for review

Legal and complianceReads every word at human speed

Reviewed and released

Both stages move at the same pace. Nothing waits.

Conceptual simulation. Only the fivefold figure comes from the example; both stages are assumed to start at the same pace for illustration.

At the extreme

When AI output becomes AI input

Carried to its logical extreme, the implications are deeply concerning. Imagine an enterprise that ignores all three mindsets. It builds no Judgment Labs to verify automated outputs, and it has automated away the entry-level staff who historically caught anomalies in operational data. What happens when AI-generated content becomes the unquestioned input for other internal AI systems?

The error cascades. The enterprise constructs a closed loop of accelerated hallucination. The entire corporate system becomes faster at generating and consuming its own unverified synthetic data, drifting untethered from physical and operational reality. That scenario is a governance and operational nightmare.

Physical and operational reality

A closed loop of accelerated hallucination.

Judgment Labs. Experts examine AI artifacts against historical expert analyses.

Entry-level pipeline. Junior staff who catch anomalies in operational data stay in the system.

Workflows mapped. A faster step does not push unchecked output into the next stage.

Conceptual model. The analysis describes the extreme case where all three mindsets are ignored; the intermediate states are illustrative.

Story three: the talent market

Candidates now ask how you will keep them relevant

Workers see new tools arriving in their daily workflows, watch foundational tasks being automated and recognize that their existing skills face rapid depreciation. Candidates are reacting. A workforce study released on November 2, 2026 by Marsh reveals a shift in what attracts people to an employer. Historically, employers competed for top talent mainly through compensation and healthcare benefits.

Before you look: which did more candidates cite as a primary reason to choose an employer?

Candidates: primary reason to select an employer
Competitive salary
60%
Learning and upskilling opportunities
58%
Career advancement prospects
52%
Marsh workforce study released November 2, 2026. Share of candidates citing each factor.

Learning and upskilling opportunities are now cited by 58 percent of candidates as a primary reason to select an employer. Career advancement prospects stand at 52 percent. Both closely rival competitive salary, cited by 60 percent.

In an era of rapid automation, long-term economic security depends on continuous capability development. An employer that hands people an AI tool to speed up output but offers no structured development of higher-order judgment can look like a serious career risk, because the employee could leave three years later less competitive than when they arrived. Candidates sit in interviews and ask hiring managers directly how the organization plans to keep them professionally relevant.

Skills-first hiring is arriving faster than the plumbing

The pressure is forcing a broader move toward skills-first talent evaluation. Marsh reported that 52 percent of surveyed organizations now prioritize skills-based assessments over traditional credentials such as university degrees. Seventy-six percent are rewriting legacy job descriptions around dynamic competencies rather than static job titles, and 81 percent are deploying functional skills assessments to measure actual capability.

The execution is messy. Marsh found that 61 percent of those organizations cite a severe lack of integration between their new functional assessment tools and their core HR enterprise systems. They are gathering skills data but lack the infrastructure to store, analyze or apply it across workforce planning. Only 23 percent of employers report being prepared and actively using AI technologies in recruitment workflows with established governance frameworks.

Organizations moving toward skills
Prioritize skills-based assessments over traditional credentials
52%
Rewriting legacy job descriptions around competencies
76%
Deploying functional skills assessments
81%
Where the infrastructure lags
Cite a severe lack of integration between assessment tools and core HR systems
61%
Prepared and actively using AI in recruitment, with governance frameworks
23%
Marsh workforce study. The 61 percent figure refers to the organizations moving toward skills-first evaluation; the 23 percent figure refers to employers.

There is also a budgeting irony here. When quarterly revenue contracts and budgets tighten, learning and development has historically been the first line item reduced, because training expenses appear conspicuously on the balance sheet. The true costs of cutting capability development stay hidden.

Visible on the balance sheet

Training expense

Scattered across departmental operating budgets

One visible line. Nothing else has moved yet.

The costs named are skill atrophy, operational friction, decreased engagement and employee turnover. They are described as hidden and scattered; no amounts are reported.

Workplace learning is no longer a peripheral support function distributing compliance checklists. It has become a primary recruitment and retention driver. That raises the question learning leaders now face: if candidates demand real skill development, traditional courses fail to deliver procedural competence and isolated AI tools create bottlenecks, how is the profession changing its design architecture?

Story four: DevLearn 2026

From the four-hour course to the navigation system

The answer is dominating the pre-conference sessions at DevLearn 2026. The agenda signals an industry-wide pivot away from rigid, static course design, the event-based model where employees sit in a room for four hours, and toward continuous workflow learning and dynamic performance support.

Programming centers on practical frameworks such as the Five Moments of Need, originally developed by Bob Mosher and Conrad Gottfredson. The consensus spreading across the industry is that organizations cannot train an employee once during onboarding and expect reliable behavioral execution six months later.

A maintenance technician's route through the Five Moments of Need.

How the technician is supported

Stress the system

Select a moment to follow the technician servicing an industrial pressure valve.

New. The technician learns about the valve and its underlying principles for the first time. More. They expand their foundational knowledge to understand how that valve fits into the broader hydraulic machinery. Apply. They stand in front of the machine with a wrench in hand, trying to recall the torque specifications and sequence learned three months earlier in orientation. Solve. The valve malfunctions in an unexpected way and they must troubleshoot an unfamiliar problem. Change. The equipment manufacturer issues an update that alters how the entire mechanism functions.

Conceptual diagram combining the Five Moments of Need with the map and navigation analogy.

Historically, corporate training focused almost exclusively on New and More. Instructional teams packed the concepts into a formal course and hoped the employee would somehow retain them when Apply, Solve and Change arrived in the workplace. Workflow learning embeds performance support directly into the flow of work, at the moment those operational needs occur.

An analogy makes the distinction intuitive. Course-centric training resembles making a driver memorize a complex paper map in a classroom before a cross-country journey. It demands heavy cognitive load, depends on flawless human recall and becomes useless the moment road construction alters the route. Workflow learning functions like a navigation system on the dashboard, delivering contextual, turn-by-turn guidance at the moment a driving decision must be made. It is, in effect, the mandate to build an organizational navigation system.

Platforms like Bites attempt to build that guidance layer by delivering short bursts of information into messaging channels at the moment of application. Here is the critical synthesis: that digital performance support must be paired with the Judgment Labs and Inquiry Mindset described earlier. If someone follows a navigation tool uncritically into a lake because the software hallucinated the route, the entire operating system has failed.

Performance support cannot be allowed to erode critical thinking. Employees need to look up from their screens, inspect their physical environment and question automated guidance whenever it contradicts operational reality. If a mobile module tells a forklift operator to make a specific turn, but an unrecorded pallet of fragile merchandise blocks the aisle, human operational judgment must override the instruction. That is why the Inquiry Mindset is non-negotiable. Organizations are training people to verify information, challenge assumptions and apply expert human judgment over automated recommendations.

The takeaway

A team can generate a hundred training modules by lunchtime at negligible marginal cost.

It is tempting to chase output and vanity completion metrics across analytics dashboards. Yet developing, safeguarding and scaling human judgment will always matter more than the speed of content production. Leaders who use artificial intelligence only to build faster content delivery pipelines, without the critical thinking frameworks to evaluate that content, are optimizing for organizational obsolescence.

A question to take to your leadership team

As AI automates the foundational, repetitive tasks that were the training ground for emerging talent, how will your next generation of leaders and experts learn to solve complex systemic challenges if they never had the chance to struggle through the basics?

Thanks for reading ldnucu.com, Learning and Development News You Can Understand.

Named in this edition: Bites, Chief Learning Officer, McKinsey, MIT's Project NANDA, Marsh, DevLearn 2026, Bob Mosher and Conrad Gottfredson.

Previous Post Next Post

نموذج الاتصال