Jonathon's AI Wiki

Opportunity AI vs Efficiency AI — 12 Thought Starters (NLW)

6 min read

Source: raw/10_Ways_to_Think_Bigger_with_Opportunity_AI.md — Creator: Nathaniel Whittemore (NLW), The AI Daily Brief weekend “long read / big think” episode · URL: https://www.youtube.com/watch?v=XsOoumeXtm4 · Platform: YouTube (yt-podcast) · Published: September 2026, about a week after GPT-6 Astra became available. NLW published an interactive companion page on aidbrief.ai with a “make it mine” personalizer for each idea.

NLW’s standing distinction is between efficiency AI (doing your existing work faster, cheaper or better) and opportunity AI (work you could not do before). This episode tackles the blank-page problem that stops people finding the second kind. It offers twelve concrete “thought starters”, several of them marketing formats, and a homework exercise for building a video pipeline. It is an opinion and ideas episode: no results or data, but the formats are specific enough to try this week.

Key Takeaways

  • Efficiency AI is the foundation; opportunity AI is where companies pull ahead. “There is absolutely nothing wrong with efficiency AI”. It is “where a lot of the initial value” comes from. But efficiency gains “will become table stakes”, and “the companies that lean into finding and discovering new opportunities, even if those opportunities are currently orthogonal to what they do”, are likely to race ahead.

  • Why now: new models are uneven. GPT-6 Astra drew complaints from coders:

    • “the smartest and dumbest model I’ve ever worked with” (a user NLW quotes)
    • “our effective spend looks doubled” (OpenCode’s Dax)

    Its excitement instead centres on video editing and 3D design, work “not currently part of our day-to-day.” NLW calls it one of the first models that “dances inside this difference.”

  • The blank-page problem is real. People carry only “a smaller, more familiar inventory shaped by our job, our tools, our experience.” NLW says he found his own opportunity uses by “stumbl[ing] and bumbl[ing]” into them. One example is a site that turns episodes into shareable chunks and feeds a social and video pipeline.

  • Shortcut: copy other jobs. Many “opportunities” are things other roles already do that you never could. “Look around at what people with other jobs are doing that you think is really cool.”

  • A recurring pattern is the “what-if machine”. Interactive proposals and decision simulators both let someone change an assumption and see the consequence.

  • Some ideas are both kinds at once. A proposal the client can reshape, or your expertise packaged as a product, opens new ground and removes repetitive back-and-forth.

The Twelve Thought Starters

The episode title says 10; the companion exhibit lists 12. NLW skipped #10 and #11 on air.

#Thought starterNLW’s concrete example
1Marketing people can play (games)See the three game patterns below
2Video production pipelinesHis AIDB clips are made by “a custom-built Claude-run pipeline that ingests scripts, the raw video material” and produces the videos, not a dedicated clipping product such as Opus
3Product demos people can exploreLet the buyer’s question choose the path (“open this, isolate that component… inspect the result”), not a fixed sales-deck order
4Proposals clients can shapeExpose the scope–resources–time model so “we want it sooner” returns “we can finish sooner if the team can attend more often” without another round trip
5Simulators for business decisions”We need another hire” hides several disagreements; a simulator forces everyone to state the demand, stage limits and training time they are assuming
63DAstra with Blender: 3D walkthroughs of Zillow listings, and learning experiences where rotating an object helps
7Customer stories turned into filmsCombine an interview, screenshots and a process recording into a mini-documentary that shows the problem, the change and the judgment behind it
8Training simulationsPractise difficult customer or management conversations with immediate feedback on judgment
9Physical products you can prototypeA teacher’s object with removable pieces; for his podcast studio, acoustic panels shaped by the room’s measurements
10Browser featuresSkipped on air
11A test crew for your websiteSkipped on air
12Your expertise as a productTurn part of your judgment into something people can work through themselves. The episode’s own interactive exhibit is the example, and it works internally too

The three game patterns (#1):

  • The rules carry the argument. An adviser to growing businesses builds a short game where adding people also adds coordination work, so the player experiences the thesis.
  • Play reveals why the problem matters. A consultant gives a prospect a five-minute fictional launch where sales, product and support collide.
  • Play lets people express taste. A “make this awkward apartment work” challenge places the products inside a problem the customer understands.

NLW’s caveat: “game design is of course about more than just being able to tell a coding agent what to build”, and most games fail. What changed is that a solo operator can now try one over a weekend.

Try It

NLW’s homework for #2, a video pipeline:

  1. Write and record. Have AI write a 60-second educational marketing script for what you sell, and record it “in the simplest way possible just with your iPhone.”
  2. Build the pipeline. Give the video to Claude Code or Codex. Ask it to design a visual motif, transition elements or layered graphics, and “a full production pipeline so that all you have to do is drop the source video in.”
  3. Iterate once. “Give it at least one round of feedback and see if you can push it farther.” Then ask whether recording yourself reading a script is now a low enough bar for video to become a regular tool.
  4. No obvious use case? Write a 90-second movie with AI and build the pipeline around that.

For #7, a customer-story film, he suggests a one-shot test first: put a customer journey’s raw assets into Claude Code with Fable 5.1, or Codex with Astra, and ask it to architect the whole piece. One-shotting is not the long-run method, but it shows the capability quickly.

Open Questions

  • Ideas, not evidence. The episode gives no examples of results for any thought starter, so which formats actually convert or retain customers is untested here.
  • **Thought starters #10 and 11 (browser features, a website test crew) exist only on NLW’s companion page, which was not fetched.
  • Model claims are anecdotal. The Astra comparison (strong on video and 3D, weaker on some coding) rests on a handful of posts NLW quotes, not on benchmarks.

Graph View

Backlinks