Presence isn't selling a model — it's selling a discipline for deploying one
What stopped me in OpenAI's Presence wasn't the model — they don't even foreground which one it runs on. What stopped me was that the platform sells policies, guardrails, approved actions, simulations and a Codex-powered improvement loop as a bundle, not even self-service — OpenAI's own Forward Deployed Engineers carry it in by hand. They say their own English-language phone support "resolves 75% of inbound calls without human intervention" — but the thing actually being sold isn't that number, it's the discipline wrapped around resolving the call.[1]
I'll confess: this both excites and worries me. It excites me because I'm watching the thing I called "the harness wins" become institutionalized — the model layer is commoditizing, value is moving up, and now that migration has its own product category. It worries me because pricing, geographic limits and contract terms are all undisclosed; for a developer who wants to be the chef, this is still a closed kitchen.[1]
Glow: handing security itself to agents
The second story I read that same day was Glow — an endpoint-security startup that hit a $1.2 billion valuation on a $180 million Series A. But what interested me wasn't the funding, it was the method: Glow uses its own AI agents to monitor the software and AI tools running on employee devices. As CEO Roi Tiger put it, "AI lands on the endpoint in a way we've never seen" — and the thing watching that endpoint is no longer a human, it's an agent.[2]
A chef now has to manage not just the agents cooking the meal, but the agents supervising those agents — a kitchen that grows layer by layer. This is the natural extension of my "the model layer is commoditizing" thesis: first the model got cheap, then the orchestration layer became a business, and now the layer that secures that orchestration is spinning up companies of its own.[2]
From July 16 to today: the thesis is now a category
On July 16 I wrote that the model layer was commoditizing. On the 19th I showed the agent layer had genuinely gotten cheaper. On the 21st I measured it with Writer's own research: a 33-61% gain across six models when only the orchestration changed. On the 22nd I tied Poolside's small-but-well-routed model and Google's fast small tier — held-back flagship to the same lesson. What I'm seeing today isn't a trend anymore, it's a market: Presence sells the orchestration, Glow sells securing that orchestration — neither sells the model itself.[1], [2], [3], [4]
Last week's test hasn't resolved yet — whether Laguna S 2.1 gets adapted in three separate open-source projects is a month-long question, not a one-day one. But I'm adding a new test today: when Presence's pricing becomes public, I'll watch whether it's priced like an API call or like a consulting contract — the answer will tell us whether the orchestration layer is really commoditizing, or whether OpenAI has just locked the door behind its own engineers.[1]