Hi all—Blake here with a few notes from the field.

Coming off a stretch of travel that included FMI, NGA, and a series of retailer visits, one pattern feels hard to ignore: the ambition around AI and agentic commerce is real. But the operational groundwork to support it is thin at best.

AI was the headline (again) this year. The language has matured. Fewer surface-level buzzwords, more talk about agents, automation, autonomous workflows.

But walking the floor and sitting in panels, there was a noticeable tension. In some conversations, AI meant internal efficiency. Meeting summaries, email drafts, small lifts that can seem impactful. In others, it meant something much deeper. Intelligence woven into product discovery, into substitution decisions, into fulfillment routing. Into how the entire system thinks.

Seeing those two conversations happening side by side is what keeps me up at night.

Because what’s forming right now isn’t a gap in awareness—everyone is aware. It’s a gap in execution. In discipline. In whether AI is being treated as a feature or as infrastructure.

As always, stick around for the Featured Insights at the end! We pulled together a couple of pieces from the past month that are worth your time.

🧠 AI Fundamentals That Actually Matter
💬 Featured Insight: Matt Van Gilder, NexChapter
💬 Featured Insight: Gary Hawkins, CART

001 // AI Only Works If You Know How To Direct It

For some, AI is still a faster Google. A prompt goes in, a response comes back. Helpful, but limited.

For others, it’s closer to delegation. The task is defined. The constraints are stated. The context is laid out. The output is reviewed and run again.

That difference seems small until you see the results.

Without structure, AI tools like ChatGPT and Gemini drift toward the center. It gives you something reasonable, something safe. Something that could apply to almost anyone. With structure, the output becomes specific instead of generic.

The retailers pulling ahead aren’t relying on better models. They’re clearer about what they’re asking for, and more deliberate about how they use each tool.

That clarity is doing more work than the model itself.

002 // The Future Isn’t One AI Tool. It’s AI Systems.

There’s still a quiet hope in parts of the industry that a single AI product will handle every use case. Marketing, pricing, fulfillment, customer service, analytics. One login, one interface, problem solved.

That’s not what we’re seeing in practice.

The teams making progress are assembling systems. One model for broad reasoning. Another for validation. A separate tool for creative work. Workflow layers that move decisions into real processes instead of leaving them in a chat window.

When those pieces start working together, the role of AI changes. It’s no longer a productivity shortcut, but starts behaving more like infrastructure.

The competitive pressure isn’t around who can generate better marketing copy. It’s around orchestration. How pricing decisions adjust, how promotions are optimized, how orders are routed, how substitutions are handled.

These are coordination problems.

The retailers who understand that are building systems that think across departments. The ones waiting for an all-in-one solution may be waiting for something that doesn’t exist—and never will.

003 // The Real Shift Is Moving From Assistants To Agents

For a while, most of us were talking about AI assistants. Tools that help someone do their job a little faster. Draft the reply, suggest the summary, offer a recommendation.

That’s useful, but it keeps a person in the middle of every step. Now something else is starting to appear.

Instead of discovering that an order is falling behind during fulfillment, the system sees the pattern forming. It looks at what’s being picked now, what’s still in queue, and which orders are at risk. It flags the right people early, while there’s still time to correct course. The team steps in because the system surfaced it—not because the delay has already happened.

The shift sounds small until you consider what it does to flow.

Across grocery operations, agents are beginning to handle routine decisions that once required constant oversight: identifying at-risk orders, adjusting pick sequencing, escalating bottlenecks before they spread, and routing work based on live conditions instead of static plans.

In grocery, this is beginning to surface in areas that are deeply operational—substitution decisions, customer service responses, personalized merchandising flows.

None of this feels flashy, but it changes how work moves.

And once agents are embedded into workflows, the question isn’t whether AI is “helpful.” It’s whether your operation is structured to let it act.

004 // Ownership And Control Are Becoming Strategic AI Questions

At first, speed was the priority. Get something live, connect to a platform, see what it can do. That was reasonable.

But who owns the data once it moves through these systems? Can it be moved somewhere else? What happens to costs as usage grows? How much room is there to shape the system around your own workflows?

There’s a quiet realization setting in that renting intelligence isn’t the same as building on top of it.

This doesn’t mean every retailer needs to train models or manage infrastructure. It does mean the structure of the relationship matters. If the intelligence sits inside a closed box you can’t examine or adjust, you’re accepting limits that may not be obvious at first.

In grocery, small structural decisions add up. Margins are thin. Over time, control over how AI touches your data and operations stops being a technical question and becomes a strategic one.

Features will change. The architecture you choose will stay with you much longer.

005 // The Barrier To Building Custom Technology Is Collapsing

Not long ago, building something custom meant a long process: define the scope, find the budget, and wait for engineering capacity that might never materialize.

That alone stopped a lot of ideas before they started.

Now teams are building small internal tools in days. A dashboard to track ongoing issues. A workflow adjustment. A reporting layer that would have taken months before. The gap between idea and build is smaller than it used to be.

This doesn’t eliminate the need for strong technical leadership. If anything, it raises the standard. But it changes where momentum begins.

The constraint is less about writing code and more about defining the problem. If the issue is clear, a system can be built around it. If it’s vague, nothing moves.

Retailers who can turn operational friction into clear requirements will move faster. Not because they hired an army of engineers, but because the barrier to building is lower.

006 // The Interface Is About To Disappear

Most AI interaction still happens in chat. You type something in. It sends something back. It feels controlled, familiar.

That won’t be the only way for long.

Voice is already common. Cameras and computer vision are starting to become commonplace. Systems can look at what’s in front of them and respond. Instead of explaining the issue, you show it. Rather than drafting instructions, you say them out loud.

In a store, that changes things. A camera catches a shelf problem without someone filing a report. A picking route reveals inefficiencies as it’s happening. A fulfillment bottleneck shows up before a dashboard is built to track it.

There doesn’t have to be a chat window for any of that.

If AI stays in the back office, it will feel like analysis. But it’s moving toward the floor, toward live decisions.

007 // The Industry Risk Is Not Falling Behind On AI

What worries me isn’t that grocers are ignoring AI. It’s that many are adopting it in ways that don’t build anything scalable.

It’s easy to add tools, to run pilots, to layer assistants onto existing workflows. Activity is visible and it feels like progress.

But without clear problem frameworks, without automated processes underneath, without an architecture that connects decisions across systems, the gains tend to stall out.

The separation over the next few years won’t come from who tested the most products. It’ll come from who changed how intelligence fits into the operation. Who built structure around it. Who treated it as part of the infrastructure.

At NGA, the interest was obvious. But the application was shallow. Some teams are restructuring around automation and systems. Others are just using ChatGPT to clean up notes.

There’s still time to close that gap. But it won’t stay that way indefinitely.

At Homesome, these principles are foundational to everything we design and deploy.

Most of our recent work hasn’t been about launching standalone AI features. It’s been about stitching intelligence into the places where decisions already happen—how products are surfaced, how offers are personalized, how orders are routed, how fulfillment issues are identified before they cause delays, and much more.

If those systems stay disconnected, AI just adds noise.

The industry conversation will keep evolving. New terms will replace old ones. But underneath it, the fundamentals won’t move much.

We’ll keep sharing what’s proving out in real operations—and what still holds up once the pilot phase ends.

by Matt Van Gilder, Principal & VP of Omnichannel at NexChapter

Most regional grocers aren’t falling behind because they lack ambition or technology.

They’re struggling because their omnichannel operating model was never rebuilt to support the growth they layered on over time.

A Practical Omnichannel Maturity Guide for Regional Grocers reframes omnichannel challenges as a leadership and sequencing problem, not a digital one, and introduces a clear, executive-level framework to diagnose where cracks are forming before they become costly failures. Through the Omnichannel Maturity Grid, the guide helps leaders see whether their business is fragmented, overextended, or truly positioned to scale, and, more importantly, what to fix first.

If your teams feel stretched, initiatives stall, or AI and retail media aren’t delivering on their promise, this guide offers clarity, prioritization, and a path forward grounded in real-world grocery operations.

Read the full guide →

by Gary Hawkins, CEO at CART: The Center for Advancing Retail & Technology

Retail is entering a deceptively dangerous phase of its technology evolution.

On the surface, progress appears steady. AI is being embedded across forecasting, pricing, promotions, supply chain, marketing, labor, and customer engagement. Dashboards are smarter. Models are faster. Recommendations are sharper.

But beneath that progress sits an uncomfortable truth: Most of today’s AI is trapped inside yesterday’s enterprise architecture. Siloed applications. Fragmented data. Disconnected decisions. Human-mediated workflows.

The result? AI delivers incremental efficiency, not structural advantage. Retailers are doing the same things slightly better—when what’s required is the ability to operate fundamentally differently.

Becoming an agentic enterprise is not a single deployment. It is a multi-phase evolution in how data, intelligence, and decision-making flow through the organization. Retailers who understand—and sequence—this evolution will compound advantage.

Those who don’t will quietly automate themselves into irrelevance.

Read the full article →