# Launch Examples Function as Usage Instructions ## Claim The examples a vendor chooses to demonstrate a new capability become the de facto instructions for how developers use it, regardless of the tool's actual comparative strengths. <!-- atomicity: 1 idea ✓ --> ## Scope & conditions - Applies to any general-purpose capability launch — a model, an API, a framework — where the vendor picks a small set of demo use cases to illustrate a broader capability. - Strongest where the audience has no other signal yet: a brand-new release, before independent benchmarks or community use cases accumulate. - The mechanism is imitation under uncertainty, not evaluation — developers copy the shown use case because it's the visible, working example, not because they compared it against alternatives. - Breaks down once independent comparative evidence accumulates; early demo-driven usage can and does shift once users find the tool suits other tasks better or worse than the launch examples implied. - One confirmed quote so far; treat as a candidate pattern until a second clean instance is verified, beyond the informal parallel noted below. ## Evidence > [!quote] ID:rw_01m1qd8avkv3ghjv2spcy6j0je > "The blog openai posted has a lot of examples with blender and 3D game stuff to show that it's good at spatial reasoning. So I think that encouraged people to go the 3D route." Angie Jones asked on X (2026-09-05) why developers were suddenly using a newly-released model for 3D work, and a reply attributed it directly to OpenAI's launch blog choosing Blender and 3D-game examples to illustrate spatial reasoning. The product name captured as "Astra" and the reply's author are both unverified and should be confirmed before citing this publicly. A separate source read the day before — Dave Winer's observation that some developers use sample code without ever reading the spec — describes a similar mechanism in a different domain (API design rather than model marketing), but that source isn't in this run's material and isn't cited here as confirmed second evidence; it's noted as an unverified parallel worth checking against its original text before treating this as a two-source pattern. ## Links - [[3. Resources/Permanent Notes/A Capability Demonstration Is Not an Operating Model]] — adjacent, not identical: that note says a demo doesn't prove a system can hold an ongoing job; this one says a demo, whether or not it proves anything, still determines what people build. - [[3. Resources/Permanent Notes/A Vendors Capability Claim Needs a Paired Comparison Not a Demo]] — contrasts: that note is about what a demo can prove to an evaluator; this one is about what a demo causes an audience to do, which happens whether or not the demo is rigorous evidence. - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] — home: how AI vendors' launch choices shape downstream developer behavior is a recurring thread there. ## Tests / falsifiers - Find a launch where the vendor's chosen examples and the tool's actual most-common use case diverge over time (post-launch usage data, or a vendor statement about "surprising" use cases). Convergence would weaken the claim; persistent divergence toward the demoed examples would support it. - Confirm the Winer parallel from its original source before citing both as independent instances of the same pattern. ## Source - [been on a plane for the last 24h so may...](3.%20Resources/Readwise/Full%20Document%20Contents/Tweets/been%20on%20a%20plane%20for%20the%20last%2024h%20so%20may....md) — a 369-character tweet-plus-reply capture; the tweet's author is Angie Jones (@blackgirlbytes), the reply's author is inferred from thread structure and unconfirmed; verify the product name "Astra" before citing.