The AI Reckoning: Why the Flabby Middle is the Next Battlefield

Jason Lemkin and Rory Driscoll discuss the shift from token-maxing to an ROI-driven enterprise economy. The era of unconstrained AI spending is ending, exposing a dangerous flabby middle in the LLM market. The companies that survive the coming shakeout will not be those with the most impressive models, but those that solve the unit economics of inference. Founders and investors must pivot from growth-at-all-costs to margin-focused operations before the market forces a correction.

The Hidden Cost of Token-Maxing

For the past 18 months, the dominant strategy was token-maxing, which meant throwing unlimited AI resources at every internal process to gain fluency. It was a necessary phase to build organizational capability, but it created a massive, hidden cost structure. As Lemkin notes, companies are now realizing that their IT budgets are bounded.

"The big story of 2027 in AI and the enterprise and all the margins in the enterprise right? Is show me the ROI next year."

-- Jason Lemkin

The transition from go build it to show me the ROI creates systemic pressure on vendors. If a software company cannot prove productivity gains that exceed the cost of the tokens, they become vulnerable. When competitors achieve parity in AI adoption, the productivity advantage vanishes, leaving only the cost burden. Companies will likely need to become 10-15% leaner just to fund their AI infrastructure.

The Peril of the Flabby Middle

The enterprise market is currently split between high-end frontier models like Opus or Sonnet and low-end, inexpensive models like Haiku or Mini. This leaves a flabby middle: workloads that are too complex for cheap models but too expensive to run on frontier models.

Open source, heavily subsidized by Chinese state initiatives, is aggressively targeting this middle layer. The danger for closed-source incumbents like OpenAI and Anthropic is that if they cannot provide a cost-competitive middle-tier product, they will be hollowed out. OpenAI is moving to build custom silicon, known as Jalapeno, in an attempt to vertically integrate and slash inference costs to defend this territory. However, as Driscoll points out, this strategy is risky: these companies succeeded by outsourcing 300 billion dollars in CapEx to hyperscalers, not by becoming hardware companies.

"The whole reason the Open AI and anthropic models work is because other idiots have spent the $300 billion on their behalf."

-- Rory Driscoll

Why Moats are Vanishing

Conventional wisdom suggests that proprietary data or custom workflows create a defensive moat. The reality is more fragile. Modern LLMs can now lift data and entire workflows from one vendor to another in weeks, a process that previously took years of consulting labor.

This moat destruction is a threat to traditional systems integrators like Accenture. Their business model, which relies on billing for bodies to perform manual migrations, is collapsing. When an LLM can perform a data migration in 30 days that previously required a five-year, multi-million dollar consulting contract, the value proposition of the incumbent service provider evaporates. This creates a vacuum that AI-first service providers are filling by offering the same outcomes at a fraction of the price.

"I literally was doing a pitch this week and the founder was going on on about their moats. And I immediately didn't want to invest. Like I just enough your moat can be LLM lifted away."

-- Jason Lemkin

Key Action Items

• Audit Your Unit Economics: Separate delivery costs from revenue immediately. If your margins are negative, you are in a trap. Fix this over the next quarter; growth on revenue you cannot keep is not a strategy.
• Pivot to Agentic Operations: Stop hiring for manual administrative tasks. Build internal agents to handle finance, invoicing, and CRM updates. This pays off in 12-18 months by reducing variable labor costs by 50-60%.
• Master Agentic Workflows: The era of the prompt engineer is dead. The new critical skill is managing agentic loops, which means understanding where they fail, how to debug them, and how to structure workflows so they do not get lazy.
• Prepare for Leaner Teams: If you are a founder, plan for a 15% reduction in headcount over the next 12 months. Reinvest those savings into AI infrastructure to ensure you do not fall behind competitors who are already doing this.
• Avoid Watch-Buying Cultures: Shift your hiring toward small, high-paid teams that work in-office. The 20-hour work week culture is a relic; companies that do not operate at high intensity will fail in the coming ROI-focused environment.