Enterprise AI agent projects often face a difficult gap between an impressive first demonstration and a fully operational deployment. DeepFabric says it has narrowed that gap with Kenco, putting six AI agents into live operations in just three months across commercial, operations, transportation and client services.
The partnership is now set to expand, with 20 supply chain agents planned across Kenco’s North American 3PL operations over the next 12 months.
That deployment speed stands out at a time when many agentic AI projects fail to progress beyond the pilot stage. Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, citing factors including rising costs, uncertain business value and insufficient risk controls. Moving an AI agent from a controlled pilot into production therefore remains far from standard practice.
A partnership that started with a proof of concept
The relationship between the two companies began with an inbound call from Kenco’s Chief Digital and Information Officer, Pal Narayanan. He had been following DeepFabric’s work with other supply chain organizations and 3PLs and believed the company’s approach to AI deployment was aligned with Kenco’s own technology strategy.
Kalyan Kommineni, founder and CEO of DeepFabric, told FreightWaves that Narayanan understood both the company’s work in the supply chain sector and the philosophy behind its platform.
For DeepFabric, the objective is straightforward: get an agent into production as quickly as possible while keeping risk to a minimum.
The company begins with a proof of concept. After signing an NDA, DeepFabric takes the customer’s data and demonstrates how the agent performs using that real-world information. The approach gives customers an early view of how the technology will operate before it reaches a live environment.
The proof-of-concept phase is also designed to expose weaknesses before they affect customer operations. Rather than beta-testing agents directly on live accounts, DeepFabric uses a controlled, non-production environment where failures can be identified and corrected.
“Agents fail all the time,” Kommineni said, emphasizing that those failures are deliberately kept outside production.
According to Kenco Chief Operating Officer David Caines, none of the six agents caused service disruption when they were ultimately moved into production.
Automating audit work creates a major opportunity
Kenco’s scale provided a particularly strong use case for automation. The company operates 141 distribution facilities and manages 43 million square feet of warehouse space across 33 states and Canada.
Across such an extensive network, every handoff generates documentation and creates another point at which information must be checked and reconciled.
DeepFabric identified audit work as an area where AI agents could remove a significant amount of repetitive manual effort. The company says this type of work is often something operators would prefer to delegate to an AI system rather than handle themselves.
Audit processes also vary depending on the parties involved. Within a 3PL environment, a warehouse may need to validate a carrier, the carrier may need to validate the customer, and the customer may in turn validate the warehouse. Each relationship creates a separate reconciliation process with its own documentation.
According to Kommineni, every connection within that chain represents a potential failure point, making reliability in production a critical requirement.
The freight auditor is now among DeepFabric’s three most widely deployed agents, alongside its proposal manager and inventory manager. The company reports that its technology can reduce audit spending by 45%, while cutting request-for-proposal response times by as much as 30%.
Measuring AI agents starts with the right baseline
For companies considering AI agents, one of the more difficult questions is determining how success should be measured.
Kommineni argues that organizations should first establish what they actually want to accomplish rather than focusing exclusively on efficiency.
A company might want to handle more work with its existing workforce, significantly increase output while adding employees, or reduce the number of labor hours required. The appropriate key performance indicators will depend on that objective.
The metrics also differ from one agent to another.
A proposal manager, for example, can be evaluated according to how many additional bids a team is able to pursue without increasing headcount. An audit agent can instead be measured by the number of manual hours required for every 100 invoices.
DeepFabric works with customers to establish their KPIs before deployment, determine the baseline, measure performance after the agent is introduced and then assess the resulting improvement.
Without that initial baseline, however, it becomes difficult to establish whether an AI agent has actually delivered measurable value.
Finding the right balance between AI models
Model selection is another challenge facing organizations deploying AI agents.
Kommineni rejects the idea that companies must choose between relying on a single model for everything and deploying dozens of highly customized models across their operations.
In his view, the most practical answer lies somewhere between those two extremes.
DeepFabric therefore assigns individual use cases to the models it believes provide the appropriate balance of capability and cost. The company also takes responsibility for managing the associated complexity, including token economics and execution.
That means customers do not have to continuously manage a model landscape that can change from one day to the next.
Kenco separates agentic AI from generative AI
Kenco distinguishes between two areas of AI investment.
The first is agentic AI, where systems perform defined tasks within established guardrails. The second is generative AI, which is used to create content and identify insights.
DeepFabric provides the technology for Kenco’s agentic AI layer. At the same time, Kenco’s teams retain authority over the agents’ outputs through inspection and override mechanisms incorporated into the platform.
The approach keeps human oversight in place while allowing repetitive operational tasks to be handled by AI.
From consulting recommendations to execution
Kommineni says DeepFabric’s emphasis on execution comes partly from his experience as a former consultant.
In consulting, he explained, providing clients with a strategy or recommended plan was often easier than helping them carry that plan through to implementation.
DeepFabric was built with the opposite philosophy in mind. Rather than simply advising customers on where AI agents could be useful, the company aims to help them execute those recommendations and bring the agents into production.
For Kenco, that execution-first model has already translated into six agents operating across its business. With 20 agents planned over the next 12 months, the partnership now moves beyond an initial deployment and toward a broader expansion of AI-driven automation across the 3PL’s North American operations.
