Convergence
Four operators argue the platform layer is the LLM, the apps are agents, and the moat is post-training on proprietary data — not pre-training a base model and not the surrounding software. Implication: foundation-model labs are running a commoditize-the-complement playbook; defensibility for application companies sits in proprietary data + integration + ecosystem.
Operators
- Mark Petty (Gartner) — LLM platform is the new OS, agents are the new apps, MCP registries are the new app stores. LLM platform is the new OS; agents are the new apps; MCP registries are the new app stores.
- Asha Sharma (Microsoft AI) — The economic moat in AI is post-training on proprietary data, not pre-training a base model and Treat the product as a living organism with a metabolism, not a shipped artifact. The economic moat is post-training on proprietary data, not pre-training; product is a living organism with metabolism.
- Evan Spiegel — Software is not a moat — ecosystems, hardware, and distribution are. Software is not a moat — ecosystems, hardware, distribution are.
- Qasar Younis — Position AI as intelligence put into things that already exist, not as a new thing. Position AI as intelligence into existing things, not as a new thing.
- Tomasz Tunguz — AI labs are running the commoditize-the-complement playbook; tag features as core or complement quarterly. Foundation labs run the commoditize-the-complement playbook; tag features as core or complement quarterly.
- Amole Naik — In AI products, capability overhang is the central growth problem. In AI products, the central growth problem is capability overhang — distribution + data integration close the gap.
Variation
- Petty maps the taxonomy.
- Sharma names the moat layer (post-training).
- Spiegel and Younis name the defensibility mechanism (ecosystem, integration into existing things).
- Tunguz names the strategic threat (lab as commoditizer).
- Naik names the symptom in product (capability overhang).
- Convergence: don't bet defensibility on the model — bet it on what surrounds the model.
Implication
For application companies: data flywheels, ecosystem positioning, and hardware/distribution control are where moat lives. Quarterly review which features are at risk of being absorbed by the next foundation-model release; assume "commoditize-the-complement" pressure is permanent. For platform companies: post-training pipelines are the IP.
Sources
- ins_llms-are-new-os-agents-are-apps — Mark Petty
- ins_post-training-as-the-moat — Asha Sharma
- ins_product-as-organism — Asha Sharma
- ins_software-is-not-a-moat — Evan Spiegel
- ins_intelligence-into-things-that-already-exist — Qasar Younis
- ins_commoditize-the-complement-ai — Tomasz Tunguz
- ins_capability-overhang-product-problem — Amole Naik