Context

Jess Hertz is COO of Shopify, the commerce platform that powers over 14% of the US e-commerce market. She joins Bob Safian on Rapid Response (cross-posted into the Masters of Scale feed) to talk through what Shopify's data shows about agentic shopping, meaning AI agents that search for and buy products on a person's behalf, as it happens in real time. The conversation splits into two threads a PM should care about: how a platform shift like this rewrites who wins in e-commerce, and how Shopify rebuilt its own operating model, incentives, and hiring around AI. Hertz came to Shopify from politics and the Biden White House, and she frames much of her thinking through incentive design and first principles.

The Big Idea

AI agents are starting to do the shopping, and that inverts the old rule where the biggest brands bought their way to the top of search. Agents reward the product that actually fits the need, not the one with the most ad spend, and Shopify's bet is to own the infrastructure layer that makes any product discoverable and buyable by an AI.

The shift is already measurable, not theoretical: last quarter, 75% of Shopify's AI-attributed orders came from categories outside the top 100, and an agent buying through Shopify's structured product data converted at twice the rate of one working from scraped web data.

Key Insights

1. Agents reward precision, not popularity

Traditional search rewarded whatever was most popular or had the most ad spend behind it, which favored big brands. AI agents work differently: they parse the actual need and match it, so niche and long-tail products surface that a keyword search would have buried.

• The data: 75% of Shopify's AI-attributed orders last quarter came from categories outside the top 100 sellers.
• Why it matters for PMs: discovery is shifting from "rank for the keyword" to "be the best literal match for an intent," which changes how product metadata, positioning, and merchandising should be built.
• Example: Hertz needed an under-bunk trunk for her daughter's ranch camp, knew none of the specifications, and an AI agent found a specialized Wisconsin maker that had been doing it for 40 years. That merchant would rarely win a normal search.

2. Structured data doubles agent conversion

Shopify built a product layer for agentic commerce called catalog, which makes a merchant's products understandable, discoverable, and purchasable by an AI rather than something the agent has to guess at by scraping a web page.

When a shopping agent uses Shopify's catalog data, it converts at 2x the rate of an agent working from scraped data. The lesson for anyone selling online: the machine-readable quality of your product data is becoming a direct conversion lever, not a back-office detail.

3. Company speed equals information speed

Hertz argues a company moves only as fast as information moves inside it, so the highest-leverage use of AI is cutting the cost of moving that information. Shopify's internal agent, River, is the concrete expression of that idea.

• What River is: an agent living inside Shopify's Slack. You describe a task in a public channel, and it can read code, run tests, and act.
• The scale: it operates in about 10,000 channels, roughly 90% of employees have used it, and it has run 275,000 sessions and taken 15 million actions in five months.
• Public by design: every request has to be made in a public channel. Hertz calls that "a feature, not a bug," because it lets the agent read context across the company and compounds through network effects, getting better with every interaction.
• Why it matters: the payoff is a lower transaction cost for information, so the company relies less on a human being the connective tissue between teams.

4. Complexity is the moat, not the enemy

A common worry is that big AI platforms will build their own commerce layer and cut Shopify out. Hertz's answer is that commerce is irreducibly complex, and absorbing that complexity for merchants is exactly what makes the infrastructure valuable and hard to replace.

• Shopify acts as one unified operating system across every way to sell: online store, in-person retail, and now agentic channels.
• The Universal Commerce Protocol, an open standard Shopify built with Google, keeps the merchant as the merchant of record even when the sale happens through an AI agent, so merchants are not disintermediated from their own customers.
• Her read on the feared "SaaSpocalypse," the idea that AI would wipe out software companies: Shopify has already come out the other side of it, because a more complex world makes reliable infrastructure more scarce and more valuable, not less.

5. A durable core funds the AI upside

Hertz splits the business into two parts, which is a useful way for any PM to talk about a bet-on-the-future initiative without overselling it.

• The core is durable: almost 90% of quarterly revenue comes from merchants who have been on the platform over a year, and that compounding base has now delivered five straight quarters of growth above 30%.
• AI is upside, honestly sized: agentic GMV (the dollar value of goods sold through AI channels) is still small, but growing, with early signs that Shopify's merchant-facing AI is helping merchants sell more. She does not inflate the agentic number into the headline.

6. Flat headcount, 34% revenue growth

Shopify's headcount has been roughly flat for more than eight quarters while revenue grew 34%, which Hertz offers as the real evidence of AI return on investment rather than usage statistics.

• Engineers can access AI model tokens mostly without restriction, with what she calls "thoughtful speed bumps" and constant monitoring, because the company wants people using the best model for each job.
• She is careful to separate two things: adoption is not impact. Lots of people using a tool does not prove it created value, and Shopify is still working out how to measure impact directly.

7. Align incentives with the real outcome

Hertz's first internal message as COO said the company had "drifted from its core," and the specific drift was the sales team's compensation plan. It rewarded actions that were not tied to whether merchants actually succeeded.

Her fix came from a behavioral-economics view that how you design compensation shapes outcomes. She realigned sales comp to Shopify's core model, which is that the company only makes money when its merchants make money, so that the incentive and the mission pointed the same direction instead of pulling apart.

Mental Models & Frameworks

X-shaped people, not T-shaped

The old ideal was the T-shaped person: broad general knowledge plus one deep vertical of craft. Hertz argues AI is pushing toward X-shaped people who have several spikes of expertise and can move across domains fast, because AI lets them get to a "7 out of 10" in a new area quickly (she now does her own data work despite a legal background).

• How it works: you keep genuine depth in your original craft but add new areas far faster than before, becoming a specialist in more than one thing rather than a pure generalist.
• The team angle: the real leverage is how these differently-shaped people fit together, like Tetris. Team composition, matching people's edges so they interlock, becomes the accelerant, not any one person's range.
• When to use it: rethinking hiring profiles, role definitions, and team design in an AI-heavy org, instead of hiring only for a single deep specialty.

Keep your identity lightweight

A Shopify phrase Hertz loves: do not get so attached to your own ideas that you cannot change them. Wake up willing to be smarter than yesterday, treat ideas as non-personal, and you can pivot without carrying baggage.

• She describes herself as "deeply unsentimental," which she treats as mostly an asset: it lets her drop an approach, even her own idea, when a better trade-off appears.
• The counterbalance is important: lightweight identity works because it sits on top of a fixed anchor. Shopify's anchor is its mission (be a 100-year company, grow entrepreneurs everywhere). The mission is fixed; how you get there is free to move, which is why constant change feels curious rather than chaotic.

Prove AI can't do the job first

Shopify CEO Toby Lutke's March 2025 memo made "reflexive AI use" a baseline expectation for every employee, and part of encoding that was a hiring rule: teams have to show AI cannot do a job before requesting a new hire for it. Hertz frames this less as a headcount cap and more as a way to force AI-first thinking into the company's actual systems and culture, not just its slogans.

Trade-offs & Nuance

Adoption versus real impact

Freely available AI tokens and high usage look like success, but Hertz is explicit that adoption and impact are different things. The upside is speed and broad experimentation; the risk is mistaking activity for value. Her resolution is to keep tokens flexible for engineers while judging the program on hard outcomes like revenue-per-employee, not on session counts.

Serving SMBs and enterprise together

Most platforms drift upmarket and let their smallest customers go, or stay small and watch their best customers outgrow them. Shopify's claim is that one unified, extensible data model lets it serve a merchant "from the kitchen table all the way up to Kim Kardashian and Skims" on the same stack. The evidence: merchants who reached $1M in GMV had a 92% retention rate over the last five years, and those who reached $10M had 97%. The bet is that architecture chosen years ago is what makes serving both ends at once possible now.

Practical Application

Put an agent in your public channels

Follow the River pattern: give teams an AI agent that lives in shared, public work channels and can read context and take action, rather than siloed private copilots. Requiring public requests is what lets the agent accumulate cross-team context and improve with use. The goal is to lower the cost of information moving between teams, which Hertz treats as the real speed limit on the company.

Make product data agent-readable

Treat the machine-readability of your catalog as a conversion lever, since an agent using structured product data converted at twice the rate of one scraping a page. Audit whether your products expose clean, structured, intent-matchable data to AI shopping agents, not just human-facing pages.

Separate adoption metrics from impact

Stop reporting AI tool usage as if it were value created. Define a hard impact measure up front (revenue per employee, cycle time, conversion) and hold the AI program to that, keeping "how many people used it" as a separate, lower-tier number.

Audit incentives against the outcome

Do what Hertz did with sales comp: take your most important incentive system and check whether it rewards the customer outcome you actually depend on. Where a metric or payout rewards activity that is not tied to customer success, realign it so the incentive and the mission point the same way.

Questions to Consider

• Where does our own product discovery still assume a human typing keywords, rather than an AI agent matching a specific intent, and what would we change if agents became the main way customers found us?
• How machine-readable is our product or data to an external AI agent today, and is poor structured data quietly costing us conversions the way scraped-versus-catalog data cost Shopify half its agent conversion rate?
• Which of our AI investments are we currently justifying with adoption numbers (how many people use it) instead of a hard impact number like revenue per employee or cycle time?
• Is there a compensation or incentive system on our team that rewards activity not actually tied to whether the customer succeeds, the way Shopify's old sales comp had drifted?
• What is our fixed anchor (the mission we will not compromise) versus the methods we are still emotionally attached to but could change tomorrow?

Bottom Line

Agentic shopping is inverting e-commerce discovery from popularity and ad spend toward precise fit, and the durable winners will be whoever owns the infrastructure that makes products readable and buyable by AI. Inside the company, Shopify's playbook is to treat information speed as the real constraint, judge AI by impact rather than adoption, and keep the mission fixed while holding every method loosely.

Notable Quotes

"Complexity is both the challenge and the moat for Shopify." (Jess Hertz)

"Adoption is never going to be the same thing as impact." (Jess Hertz)

"Keep your identity lightweight." (Jess Hertz, on a Shopify principle)