The Silent Revolution: How Squint’s AI Agents Are Digitizing the ‘Tribal Knowledge’ of the Factory Floor

By PYMNTS
September 11, 2026

For decades, the backbone of modern manufacturing has relied on a paradox: while the machinery is cutting-edge, the process of optimizing human labor has remained stubbornly prehistoric. Ask a veteran factory technician how to resolve a recurring mechanical glitch, and the answer is rarely found in a digital manual. Instead, it sounds like folklore—a series of anecdotal cues: “If it smells like ozone and the pressure gauge flickers, tap the left panel three times.”

This “tribal knowledge” is the lifeblood of the production line, yet it is notoriously fragile. It resides in the minds of veteran workers and, upon their retirement, often walks out the door, leaving a vacuum of expertise that can cripple output for months.

Squint, an innovative industrial AI startup, is aiming to solve this by digitizing the tacit expertise that keeps factories running. By moving beyond traditional robotic automation and focusing on the cognitive layer of manufacturing, the company is bridging the gap between human intuition and machine precision.

The Founding Thesis: Building a Context Layer

Devin Bhushan, founder and CEO of Squint, spent years observing the limitations of industrial technology before launching the company. He encountered a recurring roadblock: manufacturers were attempting to implement advanced Machine Learning (ML) and Artificial Intelligence (AI) on top of a vacuum.

“There was a lot of stuff that was in people’s heads that had never been documented,” Bhushan told PYMNTS. “No amount of ML or applied AI could actually help them because there was no context for that AI to work on top of.”

Squint’s fundamental insight was that before an AI agent can optimize a workflow, it must first understand the reality of the work being performed. The company began by building a “context layer”—a digital infrastructure that aggregates video footage of workers, informal documentation, and data streams from Enterprise Resource Planning (ERP) and maintenance systems.

By linking a video clip of a technician utilizing a wrench to the specific maintenance schedule they were following, Squint creates a searchable, intelligent repository of human expertise. Because manufacturers are rightfully protective of proprietary processes—such as the specific recipe formulations at a company like Pepsi—Squint avoids a “one-size-fits-all” model. Instead, each customer receives a tailored context layer, ensuring that sensitive data remains siloed and secure.

The Evolution of Model Efficiency

Early attempts to build this context layer using massive Large Language Models (LLMs) proved to be a bottleneck. The process was slow and prohibitively expensive, often requiring 10 to 14 days of compute time per customer due to the sheer volume of video data.

Squint pivoted, moving away from bloated, general-purpose models. The company developed a specialized, 2-billion-parameter model designed for one singular, high-utility task: analyzing industrial footage and mapping it precisely to existing records. This shift allowed for a massive increase in speed and a drastic reduction in cost, enabling Squint to deploy its solutions at a scale that was previously impossible.

The Lean Manufacturing Agent: A Digital Stopwatch

With the context layer firmly in place, Squint has moved from passive documentation to active optimization. The recent launch of the “Lean Manufacturing Agent” represents a significant leap forward in factory management.

Historically, Lean and Six Sigma methodologies relied on a stopwatch, a clipboard, and days of tedious manual data entry. Plant managers would stand on the floor, timing cycles and identifying bottlenecks by hand. Squint’s agent replaces this with real-time video analysis.

By uploading footage of a work cycle, the agent performs a gapless time-and-motion study. It classifies every second of activity as either “value-added,” “necessary,” or “waste.” The agent then flags specific frames where efficiency dips, offering actionable recommendations that a manager can accept, modify, or reject. The agent’s scope is elastic; it can zoom in to monitor a single workstation or pull back to analyze an entire production line, rebalancing workloads in real-time the moment one station falls behind its target cycle time.

Supporting Data and Real-World Impact

The impact of this technology is not merely theoretical. Customers utilizing Squint’s platform have reported significant improvements in operational metrics, with some citing a reduction in scrap rates by as much as 50%.

Beyond scrap reduction, the AI agents are transforming safety and training:

• Safety Protocols: In one use case, a conversational AI agent assists operators during complex pre-work safety briefings. Instead of reading through a static, 150-point checklist, an operator can narrate their observations to the agent, which then turns those notes into a prioritized, digital safety plan.
• Intelligent Maintenance: For partners like Carolina Handling, the agent acts as a diagnostic bridge. It takes unstructured problem descriptions from the field, cross-references them with maintenance history, and provides a clear diagnosis, a parts list, and an assessment of which technician possesses the specific skill set required for the repair.

These interventions are drastically reducing training time for new hires, as the AI acts as a 24/7 on-floor mentor, guiding workers through complex tasks with the collective wisdom of the entire veteran staff.

The Future: Toward the “Instant Changeover”

While Squint’s current impact on efficiency is profound, Bhushan’s vision extends toward a much more radical goal: the “instant changeover.”

In the current manufacturing paradigm, transitioning a facility from one product to another—such as shifting a snack plant from producing Cheetos to Doritos—can take months or even years of retooling, documentation, and process recalibration. This latency prevents manufacturers from reacting to market demand in real-time, often leading to supply chain lags.

Bhushan envisions a future where the factory floor is so contextually aware and the processes so well-mapped that a facility can pivot its output as fast as raw materials can be delivered to the loading dock. If a factory could switch products with the agility of a software update, the implications for the global economy would be seismic. Manufacturing capacity would no longer lag behind consumer demand; it would flex in sync with it.

Implications for the Industrial Workforce

The rise of the industrial AI agent inevitably raises questions about the future of human labor. However, Squint’s model suggests a shift in the nature of the work rather than an obsolescence of the worker. By offloading the documentation, timing, and diagnostic heavy lifting to AI, human workers are freed to focus on high-level decision-making and creative problem-solving.

As manufacturing continues to move toward a more digitized, data-driven model, the "tribal knowledge" of the past is being transformed into a tangible corporate asset. The factory of the future is not one devoid of humans; it is one where the knowledge of the humans who built it is finally, and permanently, preserved.

By capturing the nuance of the factory floor—the smell of a machine, the specific way a veteran holds a tool, the rhythm of a perfect cycle—Squint is not just documenting history. It is creating a playbook for a more agile, efficient, and resilient manufacturing future.