Copilots Everywhere. Autonomy Nowhere.

The team at Authentica just got back from NRF’s Big Show in New York, and one theme stood out: agentic AI has moved from pilot projects to mainstream strategy in retail supply chain. People aren’t asking whether AI can help—they’re asking how to move from copilots that assist to agents that complete the job. The bottleneck isn’t model capability. It’s reliable execution across systems.

Consider three-way match for supplier payments. An invoice comes in. Someone pulls the PO from the ERP, confirms receipt in the WMS, checks that quantities and pricing align, and decides whether discrepancies need escalation. Then they route the decision across systems to release or dispute payment, using email and chat along the way. It touches three or four platforms, requires judgment, and happens hundreds of times a week. You can’t hard-code it because the edge cases vary. So it stays manual.

We saw lots of AI agents that help people do these workflows faster. But we saw very few agents that can actually run them safely and autonomously end to end. That’s not a failure of imagination or models—it’s a failure of trust in cross-system execution. And it’s leaving money on the table.

Adoption is High. Autonomy is Rare.

The pattern shows up in industry numbers. A recent Fluent Commerce survey found that 70% of retailers are piloting or have partially implemented agentic AI—but only 8% have fully deployed it across operations, and just 5% describe their AI as mature. A Gartner survey tells a similar story: 75% of IT application leaders are piloting or deploying AI agents in some form, but only 15% are considering, piloting, or deploying fully autonomous agents.

Even the shiniest new tools unveiled at NRF are unlikely to move the needle. SAP, Oracle, Blue Yonder, and Manhattan Associates all announced new AI agents for order, inventory, and supplier management. But each focused on surfacing insights, gathering information, and flagging exceptions—not completing fully delegated tasks. The theme across booths was consistent: AI that helps people work faster, not AI that executes cross-system workflows end to end.

The Gap Isn’t Model Capability. It’s Cross-Stack Risk.

Supply chain is a “can’t break” environment. Mistakes create missed shipments, incorrect payments, compliance exposure, and cascading disruption.

That’s why supply chain systems are designed for control. ERP, WMS, TMS, finance platforms, and supplier portals all enforce boundaries and approvals. The tradeoff is familiar: siloed data, manual handoffs, and slow execution—but the risks stay managed.

The moment an agent acts across systems, hard questions surface. Is the data consistent across platforms? Who’s authorized to approve? What’s safe to auto-run? Can you produce an audit trail? How do you prevent duplicate actions or recover from failures?

Because these questions are difficult, the market has converged on a safer default: keep orchestration light and execution narrowly scoped, if it happens at all. That’s why “agentic” in supply chain still mostly means assist and recommend.

Why “True Autonomy” Can’t Equal “One Platform”

To make autonomy viable, the big vendors are pitching a solution: keep it contained within their own ecosystem.

SAP talks about “embedded AI” throughout its retail suite, positioning a “closed-loop” operating system that ties planning, execution, and even customer engagement together.

This makes sense for the vendor: if the agent operates inside a single governed platform, it’s easier to control permissions, policies, and audit trails. It’s also a great way to lock in clients.

But most enterprises can’t (and shouldn’t) go all-in on one platform. Real stacks are heterogeneous: systems of record, specialized tools, regional variations, acquisitions, edge workflows. So teams leading AI adoption get stuck in a loop:

Pilots prove value
Production demands governance
Governance requires cross-system control
Cross-system control requires expensive, multi-year replatforming
Autonomy stalls

Gartner predicts over 40% of agentic AI projects will be canceled by the end of 2027, with rising costs from legacy system modifications playing a significant role.

The Path Forward: A Flexible Execution Layer Between Systems

If overhauling your systems isn’t the answer, how can you get more autonomy from AI with what you have today?

That’s what we’re focused on at Authentica.

We give enterprises the capability to deploy fully customized supply chain workflows that run reliably end to end across existing systems. You can pilot them quickly, scale with confidence, and run with governance controls in place. Not a new core. Not a consolidation. A way to close the gaps at the edges of your systems.

Focusing on specific processes one by one may not be as glamorous as a single all-encompassing platform. It’s hard to get excited by freight invoice auditing, three-way match, QA review, tariff classification, PO consolidation, supplier onboarding, or shipping document assembly.

But when agents can actually do these jobs on their own, the results are huge. Hundreds of hours per year returned to teams stuck in manual review. Overpayments worth several percent of COGS recovered automatically. Lower inventory and fewer stockouts because POs went out at the right time. Border delays avoided because documents were correct the first time.

Those results don’t come from a new platform. They come from execution that works across the ones you already have.

Why This Matters Today

The gap between copilots and true autonomy isn’t going to close itself. Suite vendors might be able to close it for you—if you’re willing to wait years and lock in.

There’s another way. Teams that figure out how to safely run autonomous workflows across their existing systems will be compounding results quarter after quarter, while others are still negotiating roadmaps.

The window is now. We’re ready to help.

Ready to see how autonomous workflows can work across your existing systems? Book a 30-minute demo and we’ll show you how.