A question I hear more than any other when we talk about agentic AI with customers: “Does your own team actually run this way?”

It’s a fair question. The gap between what technology companies say and how they operate is real, and anyone who’s been through a software evaluation knows it. So, I want to answer it directly.

Yes. We do.

Over the past several months, Unilog’s engineering and operations teams have been building and using an internal AI platform we call UniOps. It’s a multi-agent system that connects the tools our teams use every day and puts a shared intelligence layer on top of them. Here’s what we learned building it, and why it matters beyond our own walls.

The Problem Wasn’t Unique to Us

Like most companies at our scale, our operations teams worked across a lot of systems. GCP for infrastructure. Site247 for monitoring. Jira for development tickets. Freshdesk for support. GitLab for code. Cloudflare for network and security. Each one has its purpose. The challenge is that when something goes wrong, or when someone just needs to understand what’s happening with a customer account, the answer is rarely in one place.

That’s not a Unilog problem, that’s a scale problem. And it’s exactly the kind of problem that distributors and manufacturers tell us they live with every day across their own operations, their ERPs, their portals, their support systems, their catalog tools.

The question we asked ourselves: what would it look like to put an intelligent layer on top of all of it?

What We Put in Place, and What We Found

UniOps uses a multi-agent architecture, meaning it doesn’t just connect to one system. It connects to all of them, and an orchestrator routes queries to the right agent, or multiple agents simultaneously, depending on what’s being asked.

A few examples of what that looks like in practice:

Incident Investigation

When a customer site experiences a slowdown or outage, the response used to require coordination across infrastructure, tech support, and cybersecurity teams. Each team would pull data from their respective tools, compare notes, and work toward a shared diagnosis. That process typically ran between 15 and 25 minutes. With UniOps, a single command now triggers parallel queries across GCP, Cloudflare, and Site247 simultaneously. The orchestrator synthesizes the findings and surfaces where attention needs to go. The same investigation now takes under a minute.

Database Queries

Our support and development teams sometimes need to pull order data or investigate account-level issues directly from the database. Previously, getting that access required routing a request through cloud ops, setting up permissions, and configuring a local environment. That setup alone could take 20 minutes before any actual work began. UniOps provides a read-only natural language interface across 135 database schemas. A developer or support team member can ask a plain-language question and get a structured answer in seconds.

Ticket Synthesis

When a customer escalation comes in, understanding the full picture means reading across Freshdesk, Jira, and sometimes attached documents in both. UniOps can pull comments, linked tickets, and document attachments together and summarize the current state in a single query. What used to require tab switching and manual cross-referencing now takes one question.

Knowledge Retrieval

Internal documentation is now vectorized and semantically searchable. Asking UniOps to explain a technical setup or summarize a process returns a structured answer with source references, faster and more reliably than navigating a wiki manually.

None of this required us to replace existing systems. It required us to connect them and put intelligence on top.

What This Tells Us About the Industry

The operational fragmentation that prompted us to build UniOps is not unique to software companies. It’s also a problem found across distribution and manufacturing industries, and in many cases a more acute one.

Mid-market distributors and manufacturers typically run their businesses across an ERP, an eCommerce platform, a PIM, a customer support system, and a growing stack of integrations. When something breaks, or when a customer calls with a question that doesn’t have an obvious answer, the people responsible have to pull from multiple systems to piece together what’s really happening. That takes time. It introduces error. And it pulls skilled people away from work that requires judgment.

The pattern we proved out internally, consolidating fragmented data access, orchestrating across systems, compressing investigation and response time, is the same pattern that applies to commerce operations. The tools are different. The workflows are different. But the underlying problem is the same.

Companies that learn how to put an intelligent layer on top of their existing systems, without ripping and replacing what works, will be able to move faster and support their customers more effectively. The ones that wait for a clean-slate moment to get started will find that moment doesn’t come.

Where the Human Still Belongs

One thing building UniOps has reinforced: the goal isn’t to remove humans from operations. It’s to remove the low-value work that gets in the way of what humans are good at.

The investigation that used to take 25 minutes still results in a human making a call. The database query that used to require a multi-step access request still has a person interpreting the results. What’s changed is that the investigative work, the data gathering, the cross-system synthesis, happens faster and doesn’t require pulling multiple teams into a coordination loop.

On the roadmap, we are working toward what we’d describe as an auto-healing capability. The scenario: a bot attack generates a traffic anomaly. UniOps identifies the user agent creating the problem, narrows it down, and drafts a rule to block the traffic in Cloudflare. A human reviews and approves it. The agent does the legwork. The decision stays with the person.

That’s the model. Not autonomous in the sense of unchecked. Autonomous in the sense that the system handles what it can handle and escalates with context rather than noise.

The Customer-Facing Version of This

The work happening inside UniOps is informing how we think about what’s possible for our customers’ operations.

Unilog HyperScale AWC (Autonomous Work Completion) is the agent-based capability within the CX1 Platform that applies this same logic to commerce workflows. Today, HyperScale AWC agents are compressing work across PunchOut, onboarding, quote processing, fraud detection, and catalog merchandising at meaningful rates. The architecture is the same: agents that connect to existing systems, orchestrate across them, and handle the repetitive, data-intensive work so your team can focus on what requires judgment.

If you’re a distributor or manufacturer thinking about where agentic AI fits into your operations, that’s a good place to start the conversation.

Learn more about Unilog HyperScale AWC

Suchit Bachalli is CEO of Unilog.