After Muse, every major platform will need its own agent, an investor says
An investor argues Muse has forced platforms to build proprietary agents, that the harness matters more than frontier models, and that agent loops will drive compute demand.
Muse altered the industry’s trajectory. Every major platform will now be forced to field a proprietary agent to defend against value expropriation and disintermediation. The defensive playbook will vary from walled gardens to strict Agent-to-Agent (A2A) handshakes designed to ring-fence proprietary data, margins, and customer relationships. What is now unequivocal: A. The harness is the product: Mass consumer adoption won't be gated by model availability, but by agentic scaffolding and execution harnesses (the OpenClaw / Muse paradigm). Once the harness solves task execution, consumer usage inflects vertically. B. Frontier IQ is unnecessary for task completion: You do not need frontier-tier reasoning for day-to-day execution. Commodity "average-IQ" models perform exceptionally well once paired with a deterministic harness and the right toolchain. C. A massive new vector for compute: The narrative that inference efficiencies would compress compute demand is wrong. Multiplying agent loops and autonomous workflows creates a step-function increase in token consumption. Portfolio Positioning: I still see zero durable moats forming at the application layer, only mass experimentation with CapEx at all time highs. With every usual suspect (Google, Amazon, OpenAI, Anthropic, etc) poised to ship competing agentic interfaces in short order, app-layer pricing power will get competed away. Consequently, I am taking profits in $META. I owned it when consensus doubted the logic of building out proprietary compute and questioned incremental ROIC, while the near-term advertising monetization runway was obvious. Roughly +25% in 3 months and I'm good. The market's mispricing has now shifted down the stack. The line every buyside is tracking is the capex 2nd derivative (accel/decel). Consensus is obsessing over "peak capex" and questioning cycle sustainability at the exact moment a blockbuster product architecture has emerged. And we just got a huge new demand vector (see "C"). Everyone is about to clone and deploy this agentic harness playbook and the aggregate compute demand required to run it is about to accelerate again. Time to adjust the portfolio accordingly.
This post quotes another post: x.com/tropicalvalue/status/2102120077353226574