From Tokenmaxxing to ROI: NEA’s Tiffany Luck on the AI Reality Check

The era of "tokenmaxxing"—where Silicon Valley CEOs encouraged unlimited AI usage regardless of cost—is rapidly giving way to a period of intense financial scrutiny. As enterprises move past the initial hype, the focus has shifted from sheer consumption to demonstrable Return on Investment (ROI).

The End of the Tokenmaxxing Era

Earlier this year, the prevailing trend in tech was "tokenmaxxing," a push to integrate AI into every possible workflow to maximize utility. However, the astronomical costs of large language model (LLM) consumption have led to a significant reality check. High-profile reports suggest that companies like Uber reportedly exhausted their entire annual AI budget within just a few months.

This fiscal pressure has forced organizations to implement strict controls, with some companies even cutting Claude licenses for specific departments or retiring internal AI leaderboards, such as Meta. For venture capitalists like NEA partner Tiffany Luck, this shift signals a transition from experimental spending to a disciplined "ROI reckoning."

The Rise of Model Agnosticism and Deployment Strategies

As companies grapple with these costs, a new pattern of enterprise adoption is emerging. Rather than committing to a single provider, enterprises are increasingly "mixing and matching" different models to optimize for both performance and price. This multi-model approach allows companies to use expensive, high-reasoning models for complex tasks while utilizing smaller, cheaper models for routine automation.

To facilitate this integration, a new class of talent is becoming essential: the forward-deployed engineer. Luck suggests these engineers are acting as a "Trojan horse" for AI adoption within large organizations. By working directly on the front lines of implementation, they help bridge the gap between raw model capabilities and specific, value-driven business use cases, ensuring that AI tools actually solve enterprise problems rather than just adding complexity.

Finding Value Across the Entire AI Stack

A common misconception in the current market is that value is concentrated solely at the model layer. While the race for the most powerful LLM continues, Tiffany Luck argues that significant value creation is happening at every layer of the AI stack.

ਉਹ ਸਟਾਰਟਅੱਪਸ ਜੋ ਐਂਟਰਪ੍ਰਾਈਜ਼ਾਂ ਨੂੰ AI ਖਰਚੇ ਨੂੰ ਟ੍ਰੈਕ ਕਰਨ ਵਿੱਚ ਮਦਦ ਕਰਨ ਲਈ ਵਿਸ਼ੇਸ਼ ਬੁਨਿਆਦੀ ਢਾਂਚਾ ਬਣਾ ਰਹੇ ਹਨ, ਤੋਂ ਲੈ ਕੇ ਉਹ ਡਿਵੈਲਪਰ ਜੋ ਉਪਭੋਗਤਾ ਅਨੁਭਵਾਂ ਵਿੱਚ 'ਮੈਜਿਕ ਮੋਮੈਂਟਸ' ਪ੍ਰਦਾਨ ਕਰਨ ਵਾਲੇ 'ਪਰਸਨਲ ਏਜੰਟ' ਬਣਾ ਰਹੇ ਹਨ, ਮੌਕੇ ਵਿਭਿੰਨ ਹੋ ਰਹੇ ਹਨ। ਜਿਵੇਂ-ਜਿਵੇਂ ਉਦਯੋਗ ਪਰਿਪੱਕ ਹੋ ਰਿਹਾ ਹੈ, ਜੇਤੂ ਉਹ ਨਹੀਂ ਹੋਣਗੇ ਜੋ ਸਿਰਫ਼ ਸਭ ਤੋਂ ਵੱਧ ਟੋਕਨ ਪ੍ਰਦਾਨ ਕਰਦੇ ਹਨ, ਸਗੋਂ ਉਹ ਹੋਣਗੇ ਜੋ ਸਭ ਤੋਂ ਕੁਸ਼ਲ, ਇੰਟੀਗ੍ਰੇਟਡ ਅਤੇ ਮਾਪਣਯੋਗ ਇੰਟੈਲੀਜੈਂਸ ਪ੍ਰਦਾਨ ਕਰਦੇ ਹਨ।

ਮੁੱਖ ਨੁਕਤੇ