Prioritizing Operational ROI Over Token Consumption and Infrastructure

In this conversation, Factory founder Matan Grinberg outlines the shift from "token maxing" to ROI-focused engineering. The core idea is that while AI leverage is permanent, the current "AI hangover"--where companies realize they are spending massive sums on vanity metrics--is a necessary correction. The hidden result is the end of the "10x engineer" myth, replaced by "load-bearing" polymaths who treat software as a business outcome rather than a coding exercise. For leaders, the advantage lies in resisting the urge to build internal AI infrastructure, which Grinberg identifies as a non-core competency, and instead focusing on the end-to-end customer journey. Those who prioritize operational excellence over raw token consumption will build the durable moats of the next decade.

The Hidden Cost of "Token Maxing"

The initial rush to adopt AI was driven by fear. Grinberg notes that organizations entered a phase of "token maxing" to satisfy board-level mandates for an "AI strategy." This created immediate, high expenditure with no visibility into ROI. The system is now responding with a "hangover" phase, where CIOs are discovering that teams are burning thousands of dollars on low-value tasks--like checking the weather--simply because frontier models were available.

"Phase two was AI at all costs token maxing part of your performance reviews... Phase three is a hangover where you go and look at the bill and it is like oh my god we are spending so much I have no idea what the ROI is."

-- Matan Grinberg

The downstream effect is a necessary contraction in the use of frontier models for trivial tasks. Grinberg argues that the competitive advantage now belongs to organizations that implement "routing"--using smaller, cheaper open-source models for 80-90% of routine work, reserving expensive frontier models only for high-stakes decision-making.

Why the Obvious Fix Makes Things Worse

Conventional wisdom suggests that if a company wants to win, it should build its own AI tools to maintain control. Grinberg refutes this, using the example of law firms spending half a billion dollars to build internal AI. He argues that building AI technology is rarely a firm's core competency. By attempting to do it in-house, they distract themselves from their actual business goals.

"The world going forward, there is going to be nothing that no one can build. Every single piece of software, anyone will in theory be able to build. Now back to the resource allocation though, is it worth your time and your energy to go and build it or should you go to someone else who has already built it or can do it faster?"

-- Matan Grinberg

This reveals a deeper system dynamic: when the barrier to building software drops to near zero, the competitive advantage shifts from technical capability to resource allocation. The firms that win will be those that ruthlessly outsource non-core functions, even if they have the technical ability to build them internally.

The Rise of the "Load-Bearing" Polymath

The industry obsession with "10x engineers"--often measured by lines of code or competitive programming accolades--is a vestige of a pre-agent world. Grinberg suggests that in the age of autonomous agents, the definition of a great engineer is shifting toward the "load-bearing" individual. These are not just coders; they are polymaths who own the full end-to-end product journey, including marketing, sales, and customer outcomes.

This creates a split. The "custodians of code" who focus only on syntax and implementation will see their value diminish. The "builders of factories"--those who design the scaffolding, CI/CD pipelines, and agent workflows that allow the organization to scale--will become the new elite. This shift requires a cultural overhaul: removing the second-class status of sales and marketing teams and integrating them fully with engineering. Companies that fail to do this are like astronauts in space--their muscles will atrophy, and when the market gravity returns, they will lack the operational strength to survive.

Key Action Items

• Implement Model Routing (Immediate): Stop defaulting to the most expensive frontier model for every task. Over the next quarter, audit token usage and route routine tasks to open-source models to preserve budget for high-value reasoning.
• Audit "Core Competency" Spend (Next 6 Months): Evaluate internal AI builds. If you are building infrastructure that is not your core business, stop. Outsource to specialists to free up capital and focus for your primary mission.
• Redefine Engineering Roles (12-18 Months): Move away from measuring "features shipped" as a primary metric. Shift toward business outcome metrics (e.g., customer retention, revenue impact) to align engineering effort with actual value.
• Integrate Sales and Engineering (Immediate): Break down silos. Ensure engineers are involved in discovery calls and salespeople understand the product technical constraints. This creates a cohesive "factory" rather than fragmented departments.
• Optimize for "Load-Bearing" Talent (Ongoing): Stop hiring for competitive programming credentials. Start hiring for agency--individuals who have taken ownership of end-to-end projects, regardless of their background or formal education.
• Adopt "In-Season" Performance Culture (Next 6 Months): Treat your team like professional athletes. Invest in their recovery, sleep, and decision-making capacity. This requires discomfort--monitoring diet, sleep, and focus--but creates a lasting performance advantage over competitors who treat their employees like interchangeable cogs.