Source: raw/If_You_Want_To_Create_Wealth_In_The_Age_of_AI_Watch_This.md · raw/Stop_caring_about_your_moat._Do_these_3_things_instead.md

Creators: Greg Isenberg, The Startup Ideas Podcast (solo episode, https://www.youtube.com/watch?v=b3tMAOFB71o) · Chris of buildgreatproducts.com, a product designer who says he has spent 15 years designing apps and advising startups (https://www.youtube.com/watch?v=9E03CH0Cknk) · Duration: not recorded · Platform: YouTube (auto-captions, both fetched 2026-11-05). Both are opinion pieces. Chris promotes his paid community and a “product OS” plugin; Isenberg mentions ideabrowser.com, which he does not describe further.

Two creators answer the same worry from opposite ends: if AI makes software, content and answers nearly free, what can a new business keep? Isenberg lists eight kinds of business that AI cannot copy overnight and ranks them. Chris argues that starting from “what’s my moat?” stops people building at all, and gives an order of questions to work through instead, ending with the product’s shape. Read together: pick who you serve and the outcome first (Chris), then choose a category that holds value once you have customers (Isenberg).

Key Takeaways

Isenberg: eight categories that hold up

  1. Atoms. Physical businesses such as a coffee shop or a plumber (“AI is not going to be able to fix that leak”). He expects their valuations to rise as AI money looks for moats.
  2. Permission. Licences, regulation and partnerships that cannot be copied on day one: being a preferred vendor of Salesforce, Microsoft or a mid-sized company; a firm that only a licensed professional may own in some countries (law, bookkeeping), where you partner with the professional and run marketing, growth and AI.
  3. Attention. A media business (newsletters, meme pages, founder-led content) built around a micro-niche, with a human in front where possible. He now has Opus 5.5 edit his short-form video and says it brought “hundreds of thousands of views” in a week.
  4. Timing. Use a trend tool or a personal agent (he names ideabrowser.com, Grok Bot, and what the captions call “Claude Duo”) to monitor your micro-niches and build first. Even if the first product is not AI-proof, the data, brand and trust you collect are.
  5. Belonging. Paid communities (Circle, Skool), weekly virtual accountability events and in-person events. His example is Sam Parr’s Hampton, peer groups for founders doing at least $3 million a year (his understanding). His own Nashville event for founders deploying agents had more demand than his team could meet. The work is choosing the niche, the shared goal, the experience and a “no-brainer” price.
  6. Liquidity. Marketplaces, “one of the most sticky businesses of all time”. A friend’s new weekly auction for collectible cars, final lap.com, is marketed through his Instagram audience. Two angles: build marketplaces for creators who already have attention, and use agents for support so “a team of 1, 2, 3, four, five people” can run a marketplace doing millions in volume.
  7. Unique data. Sell data the labs lack (he names Mercor and Micro1 as companies that sold coding data to the labs), or build a curated paid directory, since people will want trusted collections amid infinite generated content. His example, restaurant reviews free of five-star/one-star polarisation, is consumer and therefore riskier.
  8. Outcomes. Automated AI services: a creative agency run by models, or a system that finds Fiverr and Upwork jobs and bids on them. Defence comes from the best system, data and context, or an exclusive deal.

His ranking: physical businesses and licence/regulation businesses at the top, outcome businesses at the bottom. He expects most good businesses to mix categories (attention plus belonging, permission plus liquidity).

Chris: build in this order, and the moat comes later

  • Moats are “often discovered, not designed” (he quotes an a16z post). Worrying about one up front is a trap that stops people starting. He has seen a community member’s app copied and monetised by another community, so copying is real, but at the start you hold too small a share of the market for it to be worth defending.
  • Code was never a good moat. It protected the time it took to write, not any customer benefit. Four moats still exist in software: network effects, switching costs, brand and data.
  • The order:
    1. Who you help. A person, not a market: “not manufacturing”, but a job role at a company size dealing with specific workflows. Tests: can you name five of them, and do you know where they gather? Start with who already asks you for advice, whose problems you know from the inside, and who you enjoy spending time with. Example: independent physiotherapists who run their own clinic.
    2. What you help them achieve. Write it as before and after, in terms of time, money, confidence, connection or joy. Ask how they would know it worked and whether they would be upset if it disappeared tomorrow. Features are not outcomes: logging habits is a feature, a happier life is the outcome.
    3. The right product shape. Do not default to a dashboard app. His eight shapes that still work: software-enabled services, vertical agent harnesses, headless AI software, agent-native systems, taste-led niche products, ambient consumer agents, proprietary data products, and open-source core products.
    4. Then the moat, once you have customers who would miss you.
  • His own example. An agent harness for Dungeons & Dragons game masters (the product name is captioned “Lawwell”). Before: hours of preparation. After: “the best night of the month” for the players with minutes of preparation. It preps and runs the session alongside the game master and turns it into recap notes covering characters, NPCs and maps. Anyone could copy it; he builds it anyway because he knows the customer.

Try It

  1. Name five people who fit your target customer and where they gather online. If you cannot, narrow the definition (Chris’s test).
  2. Write the outcome as one before/after sentence and check it is not a feature.
  3. Pick a product shape from Chris’s eight before defaulting to an app.
  4. Score your idea against Isenberg’s eight categories and add one that holds value: a licensed partner, a community, a data asset or an audience.
  5. Set a weekly trend-watch routine on your personal agent for two or three micro-niches, with the instruction to report only trends where nobody has built anything yet.

Open Questions

  • Isenberg’s view-count and event-demand claims are his own and unmeasured here.
  • The marketplace URL is captioned “final lap.com”; the exact domain is unconfirmed. The data companies are captioned “Merore” and “micro one”; Mercor and Micro1 are assumed.
  • “Claude Duo” in Isenberg’s list is probably Claude Duet; the captions do not settle it.
  • Chris’s eight shapes are only named in this video; his longer video on them was not reviewed.
  • The D&D product’s name is uncertain from captions.