Key Takeaways
- Tibo's AI video cost math shows "unlimited" plans are unsustainable: a single 30-second clip costs $15–30 after typical iterations.
- Levelsio argues the internet is "dying" as AI replaces traditional Q&A and content platforms, pushing training data toward real-world capture.
- Ben Tossell predicts an agent era marked by more personal software, agent-maintained services, and a shift to personal brands.
- Tw93 shares a pragmatic indie dev playbook for the AI era, emphasizing product sense, weekly shipping, and data analysis.
- Claude Code now lets sessions hand off tasks via message summaries, a step toward multi-agent workflows.
1. Agent Era: Content, Traffic, and Data
Levelsio contends that people increasingly ask AI instead of posting on Stack Overflow, Reddit, etc., which both kills user traffic and stops the creation of new training data. He cites the sharp decline in Stack Overflow questions as evidence, and argues AI companies will have to harvest data from real life and user data instead. — via 1
Levelsio calls Meta's camera glasses a "genius" move because they turn a fashion item into a real-world data collection device for AI training. While privacy concerns remain, it highlights a shift toward training AI on sensory data as internet text becomes scarcer. — via 1
Ben Tossell outlines a mixed agent-era future: more creators and personal software, software engineering still in demand but transformed, SaaS "dying" while existing SaaS firms still profit, and the internet eventually serving agents and unique personal projects. He also expects people to stop caring whether content is AI-generated, making personal brands more important, with agents buying creators extra time. — via 1
Arvid Kahl observes that after @nikitabier left, his social feed reverted to random videos, rage bait, and junk, questioning whether the change is coincidence. This anecdote illustrates the fragility of curated feeds and algorithmic content quality. — via 1
2. AI Video Generation Costs Expose Unsustainable "Unlimited" Pricing
Tibo, who spends over $300K/month on AI models, breaks down video generation costs: Seedance 2.5 at ~$0.10 per second, a 30-second video at ~$3 in pure model cost, and 3–10 iterations per finished video drive the final cost to $15–30. This makes a $59/month "unlimited" plan mathematically impossible — two good videos would already lose money for the provider. — via 1
Tibo further argues that "unlimited" offers are almost always accompanied by silent rate limits, renaming "unlimited" to "enhanced fast," or shortening usable days — often all three at once. He prefers showing real prices over renaming after checkout, concluding that "prices that don't work mathematically are scams." This is a warning for both consumers and AI product founders. — via 1
3. Building for the Agent Era: Indie Tactics, New Tools, and Success Signals
Tw93 shares his indie dev philosophy: a product engineer is a mix of user research, product, engineering, operations, data, and business, with coding only ~30% of the job. He emphasizes that what you don't build matters more than what you build, and products should work without manuals. — via 1
In the AI era, Tw93 says the coding moat is lower, so token budgets should go to requirement analysis, data mining, and content usability. He advises shipping weekly, avoiding AI buzzwords and paid promotion, open-sourcing products, going global, and writing English content to build brand trust; data analysis is key to finding hidden issues. — via 1
Claude Code now allows sessions to send messages to each other, passing task summaries (not history/files) so another session can take over mid-task. This is a concrete step toward multi-agent orchestration and long-running task handoff. — via 1
Levelsio tried to reduce Claude's increasingly obscure output by forcing it to use ASD-STE100 simplified technical English, but ultimately deleted Claude anyway. This anecdote underscores ongoing frustration with AI verbosity and the limits of prompt-based control. — via 1 2
Jack Ellis has removed Google Analytics from all personal and client sites, switching to Fathom, with historical data stored in D1. This move reflects growing privacy-consciousness among developers and the appeal of lightweight, simpler analytics. — via 1
SuperX has hit $200K in revenue, up from only $800/month a year ago, as shared by Tibo, who adds "you just have to do it." This is a concrete indie success story reinforcing the power of shipping. — via 1
Ben Tossell is launching "ben's sessions," a new series walking through his agent sessions with pros and cons. It's a useful resource for learning agent workflows in practice. — via 1
