Why AI Tooling Won't Outcost Your Developers by 2028

Introduction

Katrina Wong, Chief Marketing Officer at New Relic, has spent more than 20 years building go-to-market motions at enterprise technology companies including Twilio Segment, Hired, Zuora, Salesforce, and SAP. Her current work sits unusually close to the developer audience — New Relic's observability platform is the instrumentation layer engineering teams rely on to monitor, troubleshoot, and optimize software in production. That vantage point gives her a direct read on a question a lot of technology and marketing leaders are quietly modeling into next year's budget: what happens when AI tooling costs keep climbing? Her answer pushes back hard on the prevailing forecast.

The Case Against the Gartner Forecast

Gartner has predicted that by 2028, the AI tools developers use could cost more than the developers themselves. Wong disagrees, and her reasoning is a useful lesson in how to read any technology forecast. "It's almost assuming that nothing changes between now and 2028," she said. Straight-line projections built on current token pricing ignore the most reliable variable in this market: aggressive, continuous i ovation aimed squarely at bringing costs down. She noted Jensen's recent framing that what you spend on tokens should be the cost of your developer — then argued the community is already working to route around that ceiling.

Open Models as the Pressure Valve

The specific mechanism Wong points to is the ongoing shift toward open models, which she described as what the community is actively discussing right now. Because everything in AI is accelerating, she expects the affordability problem to get solved well before 2028. For marketing leaders, that's a procurement signal as much as a prediction. Multi-year commitments priced against today's token economics are effectively a bet that costs stay elevated — a bet the people closest to the tooling aren't making.

The Real 2028 Divide Is Adoption, Not Cost

Competitive pressure alone makes the doomsday pricing scenario unlikely. Nobody is going to pay $250,000 a year for AI and tokens when there is this much competition for the business. The more consequential split is between operators who build with agents effectively and those who don't. The teams that adopt well will produce a disproportionate amount of output per person. Non-adopters and laggards are the ones who genuinely struggle — and that's an organizational problem, not a line-item problem.

What This Means for Marketing Organizations

Engineering won't become as expensive as your engineers; it will get relatively inexpensive, while the people doing it accomplish far more. That implies fewer engineers doing dramatically more work, which forces change across every function that depends on engineering capacity — including marketing ops, data infrastructure, and the internal tooling roadmap. If your team's AI plan is a budget defense, you're solving the wrong problem. The defensible position is capability: who on your team can actually operate these systems well.

The Endgame Wong Is Betting On

Asked what else the next few years might bring, Wong offered a single prediction with a clear operator's bias: "Maybe we will get to one-to-one marketing and one-to-one selling where it's highly personalized." She framed it plainly as the goal for people in marketing — hers included. It's a notable answer coming from a CMO who sells to developers, an audience historically resistant to broad-brush campaign tactics. If token costs fall the way she expects, genuine one-to-one execution stops being a budget question and becomes an execution question.

Key Takeaways

Three things worth carrying into your pla ing. First, treat 2028 cost projections skeptically — they assume a static market in the most dynamic category in enterprise software, and open models are already applying downward pressure. Second, stop framing AI as a cost-control exercise and start framing it as a capability gap; the wi ers aren't the teams that spent less, they're the teams that built. Third, plan for the second-order effects of leverage: smaller technical teams shipping far more, and the organizational change that follows. The affordability problem will likely solve itself. The adoption problem won't.

--- One note on sourcing: the transcript provided covers only the closing lightning-round segment (roughly three minutes), so this post is built on the Gartner disagreement, the open-models argument, and the one-to-one personalization prediction — the substantive material available. If you have the full-episode transcript on "Turning Data into Decisions," I can expand this with the earlier discussion.