The fastest-adopted model on Vercel this year cannot write a word

Diogo Almeida, who worked on OpenAI's instruction-following research behind ChatGPT, said he is disappointed by how the technology developed despite describing AI as “lightning in a bottle.” The excerpt indicates he is pursuing an alternative model aimed at addressing perceived limitations in current AI systems, but provides no financial metrics, product specifications, or commercial timeline.

Analysis

This is not a tradable negative read-through for AI infrastructure or public software in isolation; it is a reminder that model capability does not automatically convert into durable enterprise willingness to pay. The key bottleneck remains workflow integration, data governance, reliability and accountability—not incremental benchmark performance. That favors vendors with distribution and proprietary enterprise data (MSFT, NOW, CRM, ORCL) over application-layer companies whose valuation rests primarily on a generic AI feature premium.

Over the next 1-3 months, the relevant catalyst is whether upcoming software earnings show AI products driving paid-seat expansion, net retention, or lower service costs rather than merely elevated customer engagement. A widening gap between AI commentary and disclosed monetization would pressure high-multiple application software first, while hyperscaler capex beneficiaries can remain supported by committed infrastructure spend. Over 6-18 months, persistent weak usefulness at the workflow level would shift value capture from standalone copilots toward systems of record and implementation providers, and could make portions of the AI software cohort vulnerable to multiple compression.

The contrarian risk is that skepticism toward current consumer-facing outputs is misapplied to narrow enterprise use cases. Coding assistance, customer support triage, document extraction and internal search can clear ROI thresholds even if general-purpose models remain unreliable; the investment question is deployment economics and adoption, not whether a model appears broadly intelligent. Treat the underlying commentary as anecdotal unless corroborated by usage, pricing, retention and inference-cost disclosures.

Key Decisions for Investors

• No directional trade on this item alone; set an earnings watchlist for MSFT, NOW, CRM and ORCL, requiring disclosed AI-driven ARR, paid-user conversion or margin evidence before adding exposure.
• Maintain a quality bias within enterprise AI: favor MSFT/NOW over unprofitable application-software names with large AI revenue assumptions but limited recurring-revenue disclosure. Reassess if paid AI attach rates or net retention materially exceed consensus over the next two reporting cycles.
• Monitor IGV relative to MSFT over the next 1-3 months: sustained IGV underperformance alongside weak AI monetization disclosures would support a selective long MSFT / short IGV pair; invalidate if broad software reports accelerating billings and stable operating margins.
• For AI infrastructure exposure, use hyperscaler capex guidance and semiconductor lead times—not commentary on model usefulness—as the decision trigger. A coordinated capex-guidance reduction from MSFT, AMZN, GOOGL and META would be the actionable bearish catalyst for NVDA/AI infrastructure supply-chain exposure.