FROM HBR

Generative AI has produced two competing views about which workers will benefit most. One view is that AI will narrow performance gaps by giving lower performers templates, answers, and coaching directly in their workflow. The other is that AI will widen those gaps by eliminating routine but necessary work, which will allow top performers to spend more time on higher-value activities where they already have an advantage. At the individual task level, high and low performers may both benefit, but the more important question is which effect dominates on net in real jobs, where workers perform bundles of activities. The answer matters for how firms hire, train, organize teams, and adjust practices built for their current talent mix.

Two of us are economists who study technology; the third is the CEO of Terret, a company that helps organizations deploy AI in revenue teams. To study the effects of AI among leading companies, we decided to focus on sales, which is a useful setting for understanding how AI affects performance in jobs that combine routine and high-judgment work. In our study, the results of which have not yet been released elsewhere, we conducted interviews with eight C-level revenue executives from the Bain Capital Ventures CRO Advisory Board, held more than a dozen advisor meetings with sales leaders, and conducted a survey of 236 sales professionals at Stripe and Checkr.

One of our main findings is that while AI broadly helps workers, the largest benefits seem to be accruing to top performers, at least in leading-edge AI adopting firms. In this article, we’ll unpack why that is. The short answer is that while AI saves time, it also allows top performers to augment their own strengths. To enable that, leaders need to address four important concerns: how to hire and interview for AI-era talent, how to redesign entry-level roles so they still function as a talent pipeline, how to decide what AI capability to build versus buy, and how to strike the balance between centralizing context and decentralizing discovery. Although our focus is on sales, the underlying tension—whether AI mainly closes skill gaps or widens them—is likely to play out across jobs that bundle routine and judgment-heavy tasks, from consulting to customer success to knowledge work more broadly.

The Rise of the Autodidacts

Why do top performers benefit more from AI?

The explanation we expected going in was simple: AI would automate routine work, and top performers would use the freed time to handle more leads, more accounts, or more customer conversations. We saw some support for this. The higher-performing reps in our survey realized greater time savings than lower-performers, and managers reported that after AI adoption they assigned higher-value tasks to 44% of top-quartile reps, compared with 12% of bottom-quartile reps.

But a more important mechanism also emerged: Top performers are more fluent with how to use—and even build—AI tools, and apply them across the deal lifecycle rather than only to tactical questions. Beyond saving time on research, content creation, and documentation, they use AI as a thinking partner to sharpen strategy, learn from past interactions, pressure-test objections, and rehearse negotiations. This broader use helps them improve win rates. Managers were also 16 percentage points more likely to report quality improvements for top-quartile reps than for bottom-quartile ones, suggesting these gains didn’t come at the expense of the customer experience.

One might think that tools to improve preparation would offer greater benefits for lower-performers. For instance, lower-performing customer-service agents have been shown to benefit when AI suggestions automatically show up on their screen in the middle of calls. But here the process is different. Sales reps must actively choose the tools they use, design the workflows, and get comfortable with a technology suite that is fast moving and quickly changing.

AI literacy is central to the performance effects that we observe. In our survey, people rated both their AI literacy and the benefits they’d gotten from AI on a scale of 1 to 5. Each one-point step up in literacy was associated with a 0.71-point average increase in reported benefits. AI literacy is not measuring simple trust in AI (which has little correlation with the gains we observe). It’s the ability to use existing AI tools, learn new ones, evaluate outputs, check errors, and know when the human should remain in control. The distinction matters because AI tools are changing too quickly for static training or traditional learning approaches to cover emerging use cases.

Top performers don’t wait for trainings or a demonstration. They’re autodidacts, which puts them in a better position to benefit from AI. The autodidacts build literacy through trial, rapid iteration, comparisons with others, feedback, and social information exchange. In manager responses, the recurring success factors were executive role modeling, encouragement to experiment, peer-to-peer knowledge sharing, and concrete examples from power users.

Our interviews surfaced numerous examples of top performers constantly seeking ways to improve their skills outside of company-sponsored training. For example, one rep at the AI-powered verification platform company Checkr built a negotiation copilot before these tools became widely available. According to the company’s COO, Lindsey Scrase, this kind of creative, account-specific use was what distinguished the best from the rest. As AI tools have evolved, Checkr reps now generate highly personalized prospect messaging, run workflows that track a large book of business, and surface customer signals that combined make a measurable difference in pipeline generation and customer experience, according to Scrase. Similarly, at the fintech company Brex, an employee built an AI tool over a weekend because it seemed useful, and the company later rolled it out more broadly.

Rethink Hiring and the Talent Pipeline

If autodidacts are the ones pulling ahead, companies need to select and build for that trait, not just for AI “usage,” in their hiring and role design. According to Kevan Yalowitz, Accenture’s Global Industry Lead for Software and Platforms (which covers sales), A-players can learn and navigate a rapidly changing AI environment, whereas others require structured training and support, making screening for the ability to adapt imperative. Leaders should not simply ask job candidates whether they “use AI.” They should ask how the candidate learns. A useful interview should probe whether the rep has built a workflow, improved one, tested competing outputs, spotted an error, used AI for customer-specific preparation, or decided not to use AI because the customer context was too sensitive or too high-stakes. Asking such questions shifts the focus from tool familiarity to the capacity for curiosity and self-directed learning, and helps identify candidates who look for better ways to work and can explain why the new way improves outcomes. In fact, perhaps the most insightful guidance that came up in our interviews was from a leader who asks candidates to “share with me the AI session you used to prepare for this interview.”

Beyond hiring, leaders must also rethink the entry-level roles that prepare employees for more senior positions. The sales development representative (SDR) role illustrates this challenge. Because much of an SDR’s work in prospecting and pipeline generation is repetitive and rules-based, it’s a natural target for AI automation. But the role is also the typical entry point into a sales career and an important feeder into account executive positions. Responding to that dilemma, Brex CRO Garrett Marker chose to redesign the role, automating 20–40% of the manual work so that ambitious SDRs could use the freed time to take on full-cycle deals independently. SDRs who continued to meet their existing targets while closing a set number of full-cycle deals were automatically promoted to account executive. Choosing not to fully automate the function afforded entry-level workers the opportunity to add value through their own diligence and creativity, separating those eventual high-performers who will be augmented by AI from those who allow AI to dominate the process. As a result, the redesigned role attracted stronger talent, increased internal promotions, and enabled faster ramp.

Technology Choices

The autodidact advantage isn’t just an individual trait—it depends on what tools reps can actually get their hands on, and how easily they can combine them. That puts pressure on the tech stack itself. AI value is not tool-agnostic. The companies we interviewed are rationalizing vendors, rebuilding interfaces, and deciding which parts of the workflow require proprietary context. In the older model, reps switched between a wide range of disconnected tools like Salesforce, Gong, Outreach, ZoomInfo, spreadsheets, Slack, enablement portals, and dashboards. Now leaders are creating a more coherent operating layer that builds complete revenue context by aggregating data across systems (CRM data, transcripts, product usage, contractual information, and public data) into one interface. (Checkr, for example, evaluated roughly 20 vendors for an AI command-center layer that centralized the multiple systems reps used into a single window.)

Should companies buy new tools or build internally? (Christopher recently addressed this question in another HBR article.) Leaders should buy when the task is generic, the market solution is mature, and the tool does not require proprietary context. According to Stripe’s former CRO and current Vice Chair, Eileen O’Mara, Gong, a conversational intelligence platform, is strong for call recording and next-best action, and Stripe would not build that itself. But Stripe does build when the task requires sensitive data, product complexity, or domain-specific logic. Brex makes a similar distinction: Off-the-shelf tools can handle generic tasks such as extracting notes into customer-relationship-management fields, but differentiated value comes from proprietary data and business logic. Jeanne Grosser, COO of Vercel, an infrastructure platform for agents and full stack applications, told us that strong agent performance depends on a well-structured data layer that encodes company-specific definitions of customers, segments, and potential.

These results come from cutting edge companies with deep AI expertise, with the resources to build whatever they want. Other firms may need more nuanced approach for the build-versus-buy decision, including in-house expertise, complexity and execution risk, and time to value. According to Accenture’s Yalowitz, there’s a total cost of ownership question when considering in-house building that includes platform fees, token costs, development resources and ongoing maintenance that must be evaluated case-by-case.

Centralize Context, Decentralize Discovery

If top performers find value that a central team wouldn’t have designed for, that raises a second operating question: What should be centralized, and what should be decentralized?

The evidence supports a mixed model. Companies should centralize the context layer: data, permissions, approved knowledge bases, evaluation tests, production-grade agents, tool registries, and guardrails for customer relationship management. Agent quality and compliance depend on clean data, shared definitions, and hardening. The case for centralization is reinforced by the risks that reps describe: conflicting answers from different tools, inaccurate AI-generated content, stale knowledge bases, and the need for a human in the loop.

But companies should decentralize discovery. Top performers and power users often see customer-specific uses that central teams would not identify or execute against quickly. The survey responses show that top performing teams build local tools, have weekly AI show-and-tell sessions, share prompts among peers, and use vibe-coding tools such as Lovable to prototype applications that a centralized team can then refine and roll out to everyone.

This hybrid model lets the field discover useful applications, then productizes the ones that prove valuable. That means integrating them, securing them, measuring them, documenting them, training on them, and retiring duplicates. Leading companies are making three moves: centralizing context, decentralizing discovery, and diffusing or productizing the best tools and answers.

AI isn’t simply replacing sales work or automatically equalizing sales performance; it’s providing disproportionate lift to individuals and organizations that can build systems around their inherent strengths. Productivity gains depend on whether reps and managers convert saved time into better customer understanding, better deal strategy, and better use of human judgment. The prime managerial task, therefore, is to decide where AI should automate, where it should augment, and what should remain human

Lightbulb © 2026

Privacy Policy