NEA's Tiffany Luck: AI IPOs, personal agents and ROI reckoning

NEA partner Tiffany Luck laid out how the tokenmaxxing-to-ROI shift is changing enterprise AI budgets, the case for personal agents, and where value is actually being created. She spoke with host Rebecca Bellan on the Equity podcast on Jun 17, 2026 and tied current market moves to concrete operational changes inside companies.

What changed: how did tokenmaxxing run into an ROI reckoning?

Tokenmaxxing was the hottest trend in Silicon Valley earlier this year, pushed by CEOs who encouraged employees to push AI usage as far as it would go. The bill arrived quickly: Uber reportedly blew through its annual AI budget in a few months, some companies cut Claude licenses for parts of their org, and Meta killed its internal leaderboard. Luck framed that sequence as a shift from experimentation to accountability: firms now need ways to measure return on AI spend rather than simply maximizing token consumption.

How are enterprises actually adopting AI today?

Enterprises are mixing and matching models and bringing engineers closer to customers to deploy AI, rather than committing to a single provider. Luck said forward deployed engineers are becoming a "Trojan horse" for AI adoption, embedding systems and workflows where business value becomes visible. She also noted startups are stepping in to help enterprises track return on AI spend and that organizations are switching licenses and model use by team or use case instead of standardizing on one stack.

Where is value being created in the AI stack?

Luck argued that value is emerging at multiple layers, not just at the model layer. She referenced consumer opportunities for so-called "magic moments" while saying infrastructure, tooling, deployment teams and integrators are all generating real returns. That perspective reframes investment: buying tokens or a single model is not the whole story; operational engineering, product integration and measurement tooling matter for ROI.

Why it matters

The move from unchecked token consumption to explicit ROI demands will reshape vendor relationships and procurement. If companies continue to mix providers and buy tooling that measures spend and outcomes, vendor differentiation will shift from raw model capability to deployment, instrumentation and cost transparency. Forward deployed engineers becoming a common route for adoption means startups that enable operational embedding have leverage; business buyers will prioritize demonstrable return over brand or headline model metrics.

What to watch

Track performance and disclosures from this year’s AI IPOs and whether new enterprise tooling for ROI measurement gains adoption. Also watch whether more companies publicly cut model licenses or report line-item AI budgets: similar moves already showed up when Uber reportedly exhausted its annual AI budget in months and organizations began trimming Claude licenses.

Tiffany Luck got her start persuading companies that e-commerce was the future, and she is now focused on where AI creates repeatable business value, especially in consumer-facing "magic moments" and in the engineering work that makes those moments reliable.