The story everyone tells about AI and commerce is about the storefront: shoppers buying through ChatGPT, products surfacing in AI answers, the scramble to be visible when an assistant does the shopping. That's real, and I've written about agentic commerce elsewhere. But it's the less interesting half of what's happening.
The more consequential shift is inside the building. The next generation of DTC brands won't be defined by how they sell to AI. They'll be defined by how much of their own operation runs on it. And the gap between a brand that bolts AI onto a 2019 org chart and one that's built AI-native from the core is going to look, in a couple of years, like the gap between a brand on Shopify and a brand still running a custom Magento build in 2015.
The signal is already in the data. Deloitte's 2026 retail outlook found that roughly 68% of retailers plan to adopt agentic AI within the next 12 to 14 months. That's not a fringe experiment anymore. It's the consensus direction. The open question isn't whether you adopt it. It's whether you adopt it as a feature you add or as the way the company is built.
I've spent a career chasing operating leverage: the structures that let a small team produce output wildly out of proportion to its headcount. At WIN, a lean team ran a portfolio of brands through hundreds of millions in revenue, and the thing that made it possible was being ruthless about what humans should and shouldn't spend time on. AI is the most powerful lever for that question I've ever seen. It's also the easiest to waste. This is where it actually creates leverage, and where it doesn't.
The shift isn't headcount
reduction. It's output
per person.
Let's kill the cynical framing first, because it leads people to the wrong strategy. AI-native operations are not primarily about firing people to cut cost. The brands that approach it that way tend to gut their capability and end up slower, not leaner. The actual prize is different and bigger: dramatically more output from the same small team.
Think about what a 10-person DTC brand could realistically do in 2019. One person on support could handle a few hundred tickets a week. One analyst could refresh a handful of dashboards. One creative could produce a limited number of ad variations. Capacity was a hard ceiling, and the only way through it was to hire. Every increment of growth in operating work meant another seat, another salary, another manager eventually.
The AI-native version breaks that link. The same 10 people, with agents doing the repetitive volume underneath them, produce what used to take 40 or 50. Support covers ten times the tickets with the same headcount. The analyst ships analysis on demand instead of weekly. The creative tests dozens of variations instead of a few. You're not removing the humans. You're removing the ceiling on what they can produce, and that changes which problems a small company can credibly take on.
"The goal was never a smaller team. It was a team whose output stopped being capped by the number of hours in its week."
AI-native and AI-added
are two different
companies.
The difference between an AI-native brand and an AI-added one isn't the tools. Both buy the same software. The difference is the default. In an AI-added brand, a human owns each function and occasionally reaches for an AI tool to go faster. In an AI-native brand, an agent owns the repetitive core of each function by default, and a human steps in for judgment, exceptions, and the parts that need taste.
That sounds like a subtle distinction. It isn't. It changes how you hire, how you structure work, and how the company scales. The AI-added brand still hires a person every time the operating load grows, because the human is the unit of capacity and the AI is a nice-to-have. The AI-native brand hires only when it hits a genuine judgment or relationship bottleneck, because agents absorb the volume growth. One company's cost base grows with revenue. The other's mostly doesn't.
This is why "we use AI" is a meaningless claim now. Nearly everyone uses AI in the trivial sense. The brands pulling ahead aren't the ones with the most AI tools in their stack. They're the ones who redesigned the work so that agents do the default and humans do the exceptions, which is a structural choice, not a software purchase. You can't buy your way into it. You have to rebuild the workflow around it.
Where the leverage
actually is, and where
it isn't yet.
Not all work is equally automatable, and the brands that win sequence this correctly. The test is simple: automate work that is high-volume, rules-based, and low-judgment, where a wrong answer is cheap to catch and fix. Avoid, for now, work that is low-volume, judgment-heavy, or where a mistake is expensive or irreversible. Get that filter right and the rest is execution.
The functions that automate cleanly today are the operational connective tissue: tier-one customer support, order and returns status, review responses, cart-recovery and lifecycle messaging drafts, inventory and reorder monitoring, and first-pass analysis of your own data. These are exactly the areas the current generation of agents handles reliably, because the work is defined, the inputs are structured, and errors surface fast. This is the connective tissue that used to eat a disproportionate share of a small team's week.
The functions that don't automate cleanly yet are the ones where the whole job is judgment: deciding what's actually on-brand, setting price and positioning, owning a key relationship, making a large inventory bet. An agent can produce a hundred caption options. It cannot reliably tell you which one is the right one for your brand, because "right" here is a taste judgment trained on your specific point of view, not a pattern in public data. Knowing the difference is the whole skill.
A concrete example makes the filter obvious. Take ad creative. Generating fifty variations of a caption or ten crops of an image is high-volume, low-judgment work, exactly what an agent should own, and a brand that automates it can test far more angles than one producing each by hand. Deciding which three of those fifty actually match the brand's voice and are worth real spend is a taste call with money attached, so a human makes it. Same function, split cleanly down the middle: the agent does the volume, the human does the judgment. Run that test on every task in your operation and the automation map mostly draws itself.
Operating leverage isn't just a productivity story. It's a margin story. Every function you move from a salaried seat to an agent at a fraction of the cost shows up in operating expense, which flows straight to the bottom line. For a brand fighting to defend contribution margin against rising acquisition costs, a lower fixed opex base is one of the few levers fully inside your control. AI-native operations and a healthier P&L are the same project viewed from two angles.
Can AI really run
your support? The
honest answer.
Customer support is where most brands start, and it's the right place, because the volume is high and the bulk of it is genuinely repetitive. Leading brands now automate something in the range of 90% of tier-one support, the where-is-my-order, returns, order-edit, and refund tickets that make up most of the queue. That's not a projection. It's already happening at scale.
But the 90% number gets misread in both directions. The skeptics hear it and say AI can't do support, pointing at the bad chatbot experiences everyone's had. The hype crowd hears it and says fire the support team. Both are wrong, because the number describes tier one specifically: the high-volume, low-complexity tickets. It says nothing about the 10% that are high-emotion, high-stakes, or genuinely novel, and that 10% is where loyalty is actually won or lost.
The right model is agent-first with a fast, visible human escalation path. The agent handles the volume so your humans aren't buried in password resets and tracking-number lookups. The humans handle the angry customer, the influencer with a problem, the genuinely confusing edge case. You haven't removed humans from support. You've moved them from triage to the conversations that matter, which is a better job and a better outcome. Replace the team entirely and you save money right up until the moment a viral complaint shows you what the missing 10% was worth.
The org chart of an
AI-native brand, function
by function.
Here's what it looks like in practice. The point of the table below isn't the exact split, which varies by brand. It's the pattern: the human owns the judgment layer of each function, and an agent owns the volume layer underneath. The human's job title stops describing "the person who does X" and starts describing "the person who decides what good X looks like and reviews the agent doing it."
| Function | Agent owns (the volume) | Human owns (the judgment) |
|---|---|---|
Customer support | Tier-one tickets, order status, returns, refunds | High-stakes escalations, policy, tone of voice |
Lifecycle & CRM | First-draft flows, segmentation, send-time tests | Offer strategy, brand voice, what to say and when |
Creative | Variation generation, resizing, first drafts | The concept, the taste call, what ships |
Analytics | Pulling data, building first-pass reports, anomaly flags | The question worth asking, the decision that follows |
Merchandising & ops | Inventory monitoring, reorder flags, listing updates | Big inventory bets, assortment strategy, supplier terms |
Look at the right-hand column. Every item is a judgment, a relationship, or an irreversible decision. That's not an accident. As agents absorb the volume layer, the human roles concentrate upward into exactly the work that compounds: the calls only a person with stake, taste, and context can make. A brand built this way doesn't have a junior support rep and a junior analyst and a junior media buyer. It has a few senior generalists, each sitting on top of a stack of agents.
Taste, relationships, and
irreversible bets stay
human.
It's worth being precise about what should not be automated, because the brands that get burned are usually the ones that automated the wrong thing to look advanced. Three categories should stay firmly human, and they're the three that determine whether a brand is worth anything in the first place.
Taste. An agent can generate endless creative, copy, and product ideas. It cannot reliably know which of them is actually good for your brand, because that judgment is trained on your specific point of view and the thousand small decisions that make your brand recognizable. Hand the taste call to the agent and your brand slowly converges toward the bland average of everything the model has seen. The volume is free now. Knowing what's worth shipping is the scarce thing, and it's human.
Relationships. The retail buyer, the top creator, the manufacturing partner, the key wholesale account: these are held by people, with people. An agent can draft the follow-up email. It cannot hold the relationship, read the room on a call, or be trusted with the moment a partnership wobbles. The economics of brand partnerships run on trust that accrues to humans, not systems.
Irreversible decisions. Pricing, positioning, a big inventory commitment, a key hire. These are bets where being wrong is expensive and hard to undo, exactly the opposite of the cheap-to-fix work agents excel at. Use AI to inform them, to pull the data and pressure-test the logic. Don't use it to make them. The asymmetry between a reversible mistake and an irreversible one is the whole reason the human stays in the chair.
How to actually start,
without breaking the
company.
The mistake is trying to make the whole company AI-native at once. That's how you get a chaotic quarter, a burned-out team, and a quiet retreat back to the old way of working. The durable path is the opposite: one proven lane at a time, where each success earns the trust and builds the playbook for the next.
Why this lane: High volume means the payoff is real and visible fast. Low stakes means an early agent error teaches you something cheap instead of costing you a customer.
Why it matters: This review period is where you build the internal trust and the documented standard. Skip it and you'll either over-trust a flawed agent or never trust a good one.
The compounding: Each proven lane makes the next one faster, because you now have a team that trusts the pattern and a playbook for instrumenting, reviewing, and handing off. That's how AI-native becomes the default rather than a project.
The brands that win the next few years won't be the ones with the biggest teams or the most AI tools. They'll be the ones that rebuilt the work so a small group of people with real judgment sit on top of agents doing the volume. That's a return to a very old idea, that leverage beats labor, with a genuinely new tool to express it. The founders who internalize the difference between automating volume and replacing judgment will run circles around the ones who either fear the tools or worship them. This connects directly to the older question of when a founder should add senior operating help: in an AI-native brand, you add it later, and the help you add is more senior.
If you're trying to figure out where AI genuinely pays off in your specific brand versus where it just adds noise, that sequencing problem is exactly what the consumer commerce practice works on. The 80/20 AI playbook for founders is a good next read on which moves return the most for the least effort.
Questions from founders
building leaner, higher-output
teams.
An AI-native brand is one where agents do core operating work by default, rather than software a human operates. The legacy approach hires a person per function and hands them tools. The AI-native approach gives the repetitive, rules-based volume to agents and keeps a small human team for judgment, taste, and relationships. The result is operating leverage: a 10-person team producing the output that took 40 or 50 a few years ago, with a cost base that mostly doesn't grow with revenue.
Start with high-volume, rules-based, low-judgment work where a wrong answer is cheap to fix: tier-one support, order and returns status, review responses, cart-recovery drafts, inventory monitoring, and first-pass data analysis. These are where agents are already reliable. Leave anything involving brand taste, key relationships, or irreversible decisions for humans until you've proven the agent in a low-stakes lane first. Sequence matters more than tooling here.
Only if you hand the taste call to the agent. The risk is real: an agent left to decide what's on-brand will drift toward the bland average of everything it has seen. The fix is structural, not technical. Agents generate volume; a human with a clear point of view decides what actually ships. Keep that division and AI sharpens your voice by giving you more shots at it. Blur it and your brand slowly dissolves into the median. The scarce skill now is knowing what's worth shipping.
The brands that treat it as a cost-cutting exercise tend to gut their capability and end up slower. The real prize is output, not headcount reduction: the same small team producing far more, taking on problems a company that size couldn't credibly attempt before. In practice AI-native brands stay small longer and hire more senior people later, because agents absorb the volume that used to force junior hires. It's less about removing people and more about removing the ceiling on what a small team can do.
Where does AI actually pay off for you?
The hard part isn't the tools. It's the sequencing: which function to automate first, where to keep humans, and how to do it without breaking the team. I've spent a career building high-output lean operations. The form takes two minutes and the conversation will save you from automating the wrong thing.
Start a conversation More about Taylor →