一个 AI 创始人的虚荣心、装，和愚昧之巅｜对谈 invoko.ai 创始人梦琪

Summary

This 42章经 episode interviews 梦琪 / Mengqi, founder of invoko.ai / Invoqo, about the painful reset from vertical Agent and 2B growth-agent narratives toward a smaller consumer desktop/browser product, Clico. The source argues that Claude Code, stronger models, and crowded Agent stories make AI software startups easier to copy in concept but still hard to deliver as stable, loved products. Its main contribution is a founder-level case where Founder Ego, Vertical Agent SaaSification, AI Application Layer Moat, Customer Pull, and AI Startup Unit Economics are all tested against real user feedback instead of fundraising language.

Key Claims

• AI application startups face Model Provider Tool Competition because coding agents and model providers lower build cost while making it harder to explain durable differentiation.
• The episode separates concept novelty from product reality: many teams can describe a General Agent or Vertical Agent, but few ship stable, smooth, daily-use software.
• invoko.ai / Invoqo’s early Sourcing Agent path drifted toward Vertical Agent SaaSification when professional users demanded auditability, configuration, and process control rather than a simple end result.
• Founder-facing “sell shovels” logic around One-Person Company demand was weaker than expected because many would-be OPC users had little revenue or payment capacity.
• 梦琪 / Mengqi frames AI product business models as split between high-ARPU token-heavy users and subscription products that depend on users not exhausting token usage, sharpening AI Startup Unit Economics.
• The shift to Clico makes AI Application Layer Moat concrete: reducing the distance from user intent to solution, preserving flow across tabs and desktop context, handling privacy trust, and maintaining many versions matter more than claiming a unique idea.
• Reddit-based user research becomes a practical extension of Cross-Cultural User Research: the team repeatedly posts and comments to test pain, urgency, alternatives, and payment intent among U.S. users.
• Internal growth skills worked better than the earlier external 2B product because the team had complete Context Engineering around its own workflow, making the resulting AI Skills easier to maintain and evaluate.

Key Quotes

“很多特别对的决策都是错的” — Mengqi on self-consistent early startup choices.

“用户才是真正的衣食父母” — the episode’s turn from investor-facing concepts toward user pull.

“软件需要持续维护” — the source’s answer to the claim that AI makes product copies instantly fatal.

Connections

• 梦琪 / Mengqi — founder voice and main retrospective subject.
• invoko.ai / Invoqo — company moving from vertical Agent experiments toward consumer productivity software.
• Clico — browser/desktop product used as the source’s positive case for user experience and intent compression.
• Founder Ego — self-critique around RL narratives, impressive hires, fundraising story, and founder status.
• Vertical Agent SaaSification — concept created for the episode’s core 2B Agent failure mode.
• AI Application Layer Moat — extended by the emphasis on stability, interaction detail, privacy explanation, and maintenance.
• AI Startup Unit Economics and Product Led Willingness To Pay — payment and token-cost boundaries for AI products.
• Cross-Cultural User Research — extended by the Reddit demand-validation method.
• Agentic Workflow, AI Skills, and Context Engineering — internal workflow and skill-building layer.
• Manus, Claude Code, and Model Provider Tool Competition — external pressure shaping the founder’s anxiety.

Contradictions

• No direct contradiction found. The source qualifies the wiki’s more optimistic One-Person Company, Agentic Workflow, and AI Application Layer Moat threads by showing that lower build cost and stronger agents still leave demand validation, maintenance, workflow context, and willingness to pay unresolved.