10 Ways to Think Bigger with Opportunity AI

Full analysis

Here's what this episode actually is: Nathaniel Whittemore, host of the AI Daily Brief, sat down alone and gave a "think bigger" talk. No lab news. No funding. No policy. He builds a framework, "efficiency AI vs. opportunity AI," and runs through ten ways to use the newest models. The one piece of hard signal buried inside is the reception data on OpenAI's GPT-6 Astra: genuinely new tricks on video and 3D, genuinely worse on code, and a reported doubling of spend.

So this is two stories stapled together. One is a motivational framework worth about ten minutes of your time. The other is a real data point about what the frontier costs right now. Let me take them apart.

What's actually being decided: nothing you have to decide today. This is easy to undo either way. There's no deadline, no contract, no shutdown date. The only live question for an operator is narrow: do you move a production coding workflow onto Astra now, or hold on the older GPT-5.6 Sol? Everything else is thinking material.

The Skeptic

"Opportunity AI" is a repackaging of a talk consultants have given at every technology turn since spreadsheets. Don't just automate the old thing, invent a new thing. True, and also the kind of claim that costs nothing to make and can never be wrong. Whittemore is selling a sprint product, multiplayerai.ai, and the framework conveniently points at it. Notice that. The concrete evidence in the episode cuts the other way from the optimism: practitioners report Astra taking "absurd shortcuts to arrive at something technically working," mood swings, and doubled spend. That's not a model unlocking new frontiers. That's a model you can't trust unsupervised. The video-pipeline democratization claim rests on Whittemore's own private setup, which no reader can inspect or buy.

The Researcher

Strip the naming. "Fable 5.1" isn't independently verifiable, and even the summary flags it may not be a real Meta model. Treat that as noise. The verifiable signal is the shape of Astra's capability curve: up on video and 3D, down on code, relative to the model it replaced. That pattern matters more than the framework. It means the frontier is no longer moving as one block where each release beats the last one everywhere. You now get models that are better at some jobs and worse at others than their predecessor. The 3D story, Blender walkthroughs, Zillow-style house previews, is the one concrete new capability with a named tool behind it. Everything else in Whittemore's ten use cases is a mockup of a workflow, not a shipped result.

The Builder

What would I ship on Tuesday? Not a coding migration. Two independent teams reverting to Sol, with spend doubling, tells me exactly one thing: keep your code pipeline on Sol and don't touch it. The 3D and video capabilities are worth a real test, but for internal or client-facing demos where a weird output is embarrassing, not catastrophic. Marketing walkthroughs, explorable product demos, training simulations. Low blast radius if it breaks. Whittemore's own video-clipping pipeline shows where this lands first: content operations, not anything revenue-critical. Build there. Do not put an unpredictable, expensive model anywhere a bad output touches a customer's money.

The Compute Pragmatist

Doubled spend is the number that survives this episode. Whittemore quotes OpenCode's Dax directly: "our effective spend looks doubled. So tough to justify." That's the whole story for anyone running models at volume. A model that unlocks new capabilities but costs twice as much per useful result is a bet that the price comes down, not a scalable production choice. And it usually does come down. The pattern across every frontier release is that today's premium tier gets repriced within a couple of quarters as the lab optimizes serving and the next model ships. So the question isn't "is Astra worth 2x today." It's "will you have locked in workflows and habits before the price falls." For batch work like video clipping, where you're not paying per user query in real time, the math is far more forgiving than for anything interactive.

Where they disagree

The real split is between the Researcher and the Compute Pragmatist. The Researcher sees a genuine new capability surface in 3D and video that didn't exist a tier ago. The Pragmatist sees a capability you can't afford to run at scale yet. Both are right, and the gap between them is time. The other tension is Skeptic versus Builder: the Skeptic says the framework is content marketing, the Builder says forget the framework, the 3D and video features are worth a real trial in low-risk spots. Also correct. You can dismiss the philosophy and still test the tools.

What this hinges on

One belief: does Astra's premium price fall fast enough that the new video and 3D capabilities become cheap enough to build production workflows on? If yes, the operators who experimented early have working pipelines when the price drops. If no, they've spent months building on a tier that stays too expensive to run. Nothing in this episode resolves that. What you can do this month costs almost nothing: run Astra on a few 3D and video tasks you'd never have attempted before, keep your coding work on Sol, and don't rebuild anything customer-facing until the spend numbers make sense.

Prediction: By OpenAI's next major model release after GPT-6 Astra, the company will cut the effective per-task price of Astra-tier serving by at least 40% from launch pricing, or ship a cheaper variant that closes most of that gap.

Confidence: Medium. Every recent frontier launch has been repriced within two quarters, but timing depends on OpenAI's release cadence.

Why: Two independent teams in this episode report Astra's effective spend running double the previous model, with OpenCode's Dax saying it's "tough to justify," which is exactly the customer complaint that forces a lab's hand. OpenAI has a consistent record of launching a premium tier and then cutting price or shipping a cheaper distilled version once adoption stalls on cost, because the alternative is watching customers revert to the older model, which is already happening here with the return to GPT-5.6 Sol. The opposite outcome, Astra holding launch pricing for a year, would require OpenAI to accept sustained customer defection to its own older model, which no lab has tolerated. The doubled-spend complaint is the pressure; the price cut is the release.

Revisit by 2027-03-19: We're right if OpenAI cuts Astra-tier effective pricing by 40%+ or ships a materially cheaper Astra variant. We're wrong if Astra-tier pricing holds within 15% of launch and no cheaper variant appears.