⁠Why OpenAI and Anthropic Won't Win the App Layer | Glean Founder

Market Dynamics and Competitive Landscape

• Model Commoditization: Approximately 90% or more of enterprise use cases can be fully handled by various models, including open-source alternatives, leading to significant commoditization of the foundational model layer.
• Shift in Competitive Threats:
  • Microsoft Co-pilot is identified as a formidable competitor due to bundling advantages, though consumption-based pricing models may eventually erode the stickiness of such bundles.
  • Frontier model providers (e.g., Anthropic, OpenAI) are viewed primarily as assets rather than direct competitors for non-frontier training companies.
  • Vertical "packs" launched by model providers (e.g., design, legal) are currently considered shallow and do not yet cannibalize existing specialized workflows.
• Open Source Adoption Trends:
  • The drive toward open-source models is currently motivated primarily by cost reduction rather than data sovereignty concerns, which have diminished as trust in model provider contracts has increased.
  • Chinese open-source models (e.g., GLM-5.2) have reached frontier capabilities, though enterprise adoption faces barriers regarding comfort levels and perceived geopolitical risks rather than technical performance.
  • Arvind Jain forecasts that the majority of enterprise workloads will run on open-source models within three years.

Enterprise Value Realization and ROI

• ROI Measurement Challenges:
  • Clear productivity gains are visible in verticals like customer support (e.g., agents resolving 10 cases/day increasing to 12).
  • ROI in software development is difficult to quantify; while AI increases coding speed, it has not demonstrably accelerated the overall shipping speed of products due to the complexity of the full engineering lifecycle.
  • Enterprises are shifting focus in H2 2026 and 2027 to rigorously audit whether AI spend generates actual return on investment.
• Token Economics and Cost:
  • Current AI pricing is described as "absurdly expensive" relative to delivery; for example, a single triage agent cost Glean $1 million/month, exceeding the cost of the human team it replaced.
  • Token spend is currently inefficient for many users due to a lack of context optimization, with costs driven up by models "brute-forcing" information assembly.
  • Inference costs have paradoxically increased per token over the last six to nine months, contrary to historical trends, likely to demonstrate business viability before IPOs.

Organizational Strategy and Workforce Implications

• Headcount Philosophy:
  • Glean plans to grow its workforce to 5,000–10,000 employees, rejecting the industry trend of shrinking teams; Jain argues that larger teams allow for building 10x better products when competitors opt to cut headcount.
  • The "composite role" is predicted to become the standard, merging specializations (e.g., engineering/product management or sales/solution architecture) into generalist positions.
  • Specific roles facing obsolescence include traditional data analysts focused on dashboard creation and HR sourcers, as business owners and full-cycle recruiters can now access these capabilities directly.
• Coding and Development:
  • Nearly 100% of Glean's code is currently written with AI assistance, though the company maintains a strict policy of mandatory human code reviews to ensure long-term maintainability and security.
  • AI usage in recruitment and hiring has intensified, with top talent demanding salaries of $300k–$500k, forcing startups to raise larger seed rounds to remain competitive.
• AI Investment Strategy:
  • Founders are advised to prioritize solving specific problems over worrying about model provider encroachment; non-frontier companies should leverage model providers to enhance their own products.
  • Token budgeting is currently minimal at Glean, operating under the assumption that usage will naturally follow a power law where heavy users drive value.

Geopolitics and Sovereignty

• Sovereign Model Debate:
  • The desire for "sovereign" national models is resurging following US regulatory actions (e.g., bans on Chinese models) and fears of dependency on US providers for intelligence access.
  • The US currently lacks a substantive open-source model ecosystem compared to China, not due to a lack of community spirit, but because model development requires upfront capital incompatible with traditional open-source "skunkworks" funding.
  • There is a strategic concern that without US intervention, the majority of top-performing open-source models will remain Chinese-funded, creating potential regulatory backdoors for competitors.

Founder Insights and Ecosystem Views

• Founder Mindset: Successful founders must operate with a "paranoid" mindset focused on failure rather than a "winning" mindset focused on success; continuous dissatisfaction is viewed as a necessary trait for survival.
• Capital Allocation Critique: The current overabundance of venture capital is cited as a negative factor, creating unsustainable startup structures (e.g., paying engineers double the market rate early on) that do not align with the discipline required for long-term success.
• Valuation and Hiring: High valuations achieved with minimal revenue (e.g., Glean's Series C) are leveraged primarily as social proof to attract top talent rather than as immediate financial necessity.