Week of November 9, 2026
A non-technical creator who built a one-person content team with Claude Code, plus another who shipped 75 animated websites in one day. Also, Trae Stephens on why the best founders never hunt for ideas.
This Week's Top Videos
Claude Built Me a 1-Person Content Team (It Does Everything Now)
By The Next Wave
A self-described non-technical creator used only a $200/month Claude subscription to build a custom content command center in two weeks. It ranks video ideas using Ahrefs search data, writes native short-form scripts, and drives Adobe Premiere through MCP to auto-edit shorts. The real unlock is that Claude Code now handles the technical thinking for you.
- Claude Code improved sharply in the last two weeks. A creator whose coding knowledge stops at HTML built a full suite of custom apps with no spending beyond the $200/month Claude plan.
- Connect Claude to Ahrefs so your ideation app ranks video ideas by search volume and your channel's context, rather than relying on guesswork.
- Claude can drive Adobe Premiere or the free DaVinci Resolve through MCP. It edits shorts in the background, and you can jump in and take over the edit at any time.
- Repurposing long-form clips into shorts no longer works. Have Claude write a native short-form script from your long-form video, then record it yourself so the content still comes from you.
- Teach Claude your editing style by pointing it at a Dropbox folder of shorts your human editor already made. It studies them and copies the format, such as yellow centered captions and a split screen with the talking head on the bottom.
- Build reusable skills first, such as a scripting skill and a short-format skill. Then chain them in a command center app that starts automatically when you plug in a hard drive containing the finished video, its transcript, and the raw footage.
Why the Best Founders Don't Look for Startup Ideas | Trae Stephens
By South Park Commons
Trae Stephens argues the best founders never hunt for startup ideas: a good quest finds you, through an obsession that won't leave your head. He backs it with his own path from intelligence officer to Palantir, where Alex Karp hired him after 5 minutes, and claims VCs can spot great founders just as fast, passing on 95% of pitches.
- Don't start from the identity of 'I'm an entrepreneur' and then search for an idea. Stephens says that sequence produces 'really dumb companies.' The strongest companies start with someone obsessed with a problem who only later realizes a company is the right vehicle.
- Treat starting a company as an act of last resort. The host cites Zuckerberg's view that most companies fail, so you should only build one when it is the best available vehicle to bring an idea to life.
- Stephens found his quest from inside the customer. As a US intelligence analyst, he fought to buy Palantir and his agency told him 'We don't buy things. We build things.' That frustration with procurement became his path into defense tech.
- Great founders and hires are obvious fast. Stephens says that within about 5 minutes you can usually tell whether you want a 10-year relationship with someone. He estimates 95% of pitch meetings are clear passes, with no hidden VC KPI checklist behind the decision.
- Stephens rejects the idea of untapped markets in defense. He says founders who believe they are operating in free space are wrong, so every defense startup needs a clear plan to displace incumbents.
- Sequence your ambition. Stephens notes that SpaceX did not build Starship from day one, so founders should start with a wedge product that earns the right to pursue the moonshot.
10 Habits PhDs Must Unlearn To Build a Startup | Startup School
By Y Combinator
Your PhD training is actively sabotaging your startup: novelty is overrated and plagiarism is a valid strategy. The speaker's company Focal burned 3 months training a retail people-tracking model that customers refused to pay for. AI founders should start with a human in the loop, decide with 70% of the information, and ship before they fully understand why it works.
- Build a Stanford-style startup, not an MIT-style one. Start from one boring customer problem and iterate on it for years instead of patenting a black-box invention nobody wants. Instacart has spent 15 years on grocery delivery, a problem it didn't invent.
- Copying is fine if you execute better. Lyft vs Uber, Anthropic vs OpenAI, and Instacart vs earlier delivery attempts show that novelty is rarely required and often not preferred.
- AI startups should validate with a human in the loop before spending on compute. Focal spent 3 months training a people-tracking model for retailers. Hand-labeling a few videos a day would have produced the same insight, and the customers said they would never pay for it.
- Talk to the people who have the problem, not investors. Ask them directly: if we built X, would you pay money for it? What people say they would pay for differs from what they actually pay for.
- Apply Bezos's 70% rule: make most decisions with about 70% of the information you wish you had. Larry Ellison pushed SQL to market faster than Stonebraker's technically superior QUEL and won the database wars.
- Drop scientific exactness in marketing. 'Fifteen minutes could save you 15% or more' was never peer reviewed, yet it became one of the most successful slogans ever. Use punchy language without lying.
Get your skills recognized with GitHub Copilot verified proficiency
By GitHub
GitHub now certifies developers without an exam: its proficiency engine scores your actual Copilot usage signals across 5 levels, beginner to expert. Hitting proficient or higher earns a Microsoft Pro Badge via Credly and a verified capability statement on LinkedIn, a recruiter-visible signal that beats self-reported AI skills.
- GitHub Copilot Verified Proficiency replaces exams, labs, and assessments with analysis of your real product usage signals, measured against a defined proficiency framework. Opt in at learn.github.com.
- The framework has 5 levels: beginner, intermediate, proficient, advanced, and expert. Advanced and expert are still under development, so proficient is currently the top level you can earn.
- Results are transparent. You can see which specific usage signals drove your score and browse the other levels to learn what behaviors move you up. That gives you a practical roadmap for leveling up your AI-assisted workflow.
- Proficiency is continuous, not a one-time snapshot. Your level evolves as you keep using Copilot, which turns credentialing into an ongoing signal instead of a point-in-time test.
- A LinkedIn partnership adds GitHub-verified capability statements to your profile through connected apps. Recruiters and hiring managers see skills verified by GitHub rather than self-reported claims.
- Reaching proficient or higher earns a Microsoft Pro Badge. You claim it through Credly and link it to Microsoft AI Skills Navigator. Participation is opt-in, results are private by default, and GitHub evaluates usage signals, not your code or prompts.
Speed up design system documentation with the Figma agent
By Figma
Figma's agent turns design system documentation from a slog into a review task. One prompt generates color swatch grids grouped by family, it drafts component docs by copying your existing conventions, and custom skills published org-wide lock in your team's documentation standards. That matters for any team whose design system outpaces its docs.
- Generate foundation docs with one prompt. Ask the agent to create a color swatch grid for all color variables, grouped by color family and labeled with variable names. Repeat the same approach for typography and spacing scales.
- Draft component documentation by selecting the new component plus an existing documentation page as reference. The agent then mirrors your team's conventions instead of inventing a new format.
- Write a brief overview of each component's purpose in its component description field. Both teammates and the Figma agent read that field when using your library, so it improves future AI-generated designs.
- Convert a back-and-forth refinement conversation into a reusable custom skill with a single prompt. The instructions you iterated on become built into future agent responses.
- Use custom skills for repeatable hygiene tasks beyond docs, such as drafting consistent descriptions, organizing components, and applying meaningful layer names.
- Publish skills to your org so every designer can view and enable them in agent chats. This makes documentation conventions enforceable across the team rather than tribal knowledge.
You Won’t Believe the Websites Claude Can Build Now
By Futurepedia
One creator built 75 designer-quality animated websites in a single day using Claude Code with single prompts and zero hand-written code. His workflow: paste a detailed design prompt, set permissions to auto and reasoning effort to extra, and get desktop, tablet, and mobile-ready sites in 10 to 30 minutes. It works on the $20 Pro plan.
- Detailed prompts drive the output quality. Each template prompt specifies fonts, brand colors, layout, and animations, which is why results closely match the demos. Swap in your own brand colors and business name to customize.
- You don't need a terminal. Open the Claude desktop app, switch from the chat tab to the Claude Code tab, choose a local folder, and paste your prompt. It writes index, style, and script files straight to your computer.
- For hands-off builds, set permissions to auto and reasoning effort to extra with the strongest available Opus model. Medium and high effort still produced good sites. Builds took 10 to 30 minutes, longer when other projects ran in parallel.
- The $20/month Pro plan is enough for one or two sites. The creator used the $200/month top tier for all 75 and never hit its limits, even while running other projects on the same account.
- Edit by pointing, not coding. A built-in toolbar lets you highlight a specific section and ask Claude in chat to change only that part, such as a color or copy block.
- Claude flags real-world problems on its own. It warned that the 555 placeholder area code isn't a working number, which shows it checks content and not just code.