Messy Founder
Resource

Founder Lesson: Agent Businesses Collapse When Customers Do Not Own the Runtime

VeChain CEO Sunny Lu argued on Cointelegraph's Chain Reaction that always-on AI agents need blockchain-based accountability—digital passports, decentralized credibility scores, smart-contract settlements—so humans trust software that acts without constant supervision. Days earlier, Animoca chairman Yat Siu warned that traders building strategies atop third-party agents inherit hidden dependencies: model updates, tool access changes, or policy shifts can break edge without notice.

Two founders, one structural problem

Lu sells infrastructure for reputation; Siu sells games and finance while urging skepticism about agent control. Both point to the same founder lesson: distribution without custody of the runtime is fragile. If your startup wraps someone else's agent API, you are a feature until you are not.

Messy Founder readers have seen this movie with mobile app stores, Facebook pages, and Shopify plugins. Agents compress the timeline. A platform policy change on Tuesday erases Thursday's revenue.

What accountability could mean

Lu envisions passports for agents and creators, reputations that accumulate on chain, and settlements enforced by contracts. Even if you doubt public chains, translate the idea: signed attestations, immutable audit logs, customer-held API keys, and exportable workflow definitions.

Founders should ask investors: "What happens if OpenAI, Anthropic, or Google revokes our tier?" If the answer is shrug, valuation should reflect platform risk.

Monetizing expertise without coding

Lu compared agent-building to early app stores liberating non-programmers. The upside is real—domain experts ship vertical agents. The downside is commoditization: if everyone can spawn agents, moats come from data, trust, and distribution—not prompting skill.

Playbook

Own the customer relationship and workflow state. Negotiate API terms with escape hatches. Keep human-visible approvals for money movement. Build community reputation off-platform (email, events) so algo changes do not zero your audience overnight.

Founders building "agent businesses" in November 2026 win by planning for platform divorce on day one—not by pretending the divorce will never come.

Operational teams should treat vendor announcements as triggers for regression tests, not as finish lines.

Document who approved production changes, which prompts were pinned, and what rollback looked like when the first anomaly appeared.

Security and compliance stakeholders need plain-language briefings: what data leaves the perimeter, what logs exist, and what happens on failure.

Leaders who invest in evaluation harnesses this quarter usually spend less on firefighting next quarter—especially when defaults shift underfoot.

Operational teams should treat vendor announcements as triggers for regression tests, not as finish lines.

Document who approved production changes, which prompts were pinned, and what rollback looked like when the first anomaly appeared.

Security and compliance stakeholders need plain-language briefings: what data leaves the perimeter, what logs exist, and what happens on failure.

Leaders who invest in evaluation harnesses this quarter usually spend less on firefighting next quarter—especially when defaults shift underfoot.

Operational teams should treat vendor announcements as triggers for regression tests, not as finish lines.

Document who approved production changes, which prompts were pinned, and what rollback looked like when the first anomaly appeared.

Security and compliance stakeholders need plain-language briefings: what data leaves the perimeter, what logs exist, and what happens on failure.

Leaders who invest in evaluation harnesses this quarter usually spend less on firefighting next quarter—especially when defaults shift underfoot.

Operational teams should treat vendor announcements as triggers for regression tests, not as finish lines.

Document who approved production changes, which prompts were pinned, and what rollback looked like when the first anomaly appeared.

Security and compliance stakeholders need plain-language briefings: what data leaves the perimeter, what logs exist, and what happens on failure.

Leaders who invest in evaluation harnesses this quarter usually spend less on firefighting next quarter—especially when defaults shift underfoot.

Operational teams should treat vendor announcements as triggers for regression tests, not as finish lines.

Document who approved production changes, which prompts were pinned, and what rollback looked like when the first anomaly appeared.

Security and compliance stakeholders need plain-language briefings: what data leaves the perimeter, what logs exist, and what happens on failure.

Leaders who invest in evaluation harnesses this quarter usually spend less on firefighting next quarter—especially when defaults shift underfoot.

Operational teams should treat vendor announcements as triggers for regression tests, not as finish lines.

Document who approved production changes, which prompts were pinned, and what rollback looked like when the first anomaly appeared.

Security and compliance stakeholders need plain-language briefings: what data leaves the perimeter, what logs exist, and what happens on failure.

Leaders who invest in evaluation harnesses this quarter usually spend less on firefighting next quarter—especially when defaults shift underfoot.

Operational teams should treat vendor announcements as triggers for regression tests, not as finish lines.

Document who approved production changes, which prompts were pinned, and what rollback looked like when the first anomaly appeared.

Security and compliance stakeholders need plain-language briefings: what data leaves the perimeter, what logs exist, and what happens on failure.

Leaders who invest in evaluation harnesses this quarter usually spend less on firefighting next quarter—especially when defaults shift underfoot.

Operational teams should treat vendor announcements as triggers for regression tests, not as finish lines.

Document who approved production changes, which prompts were pinned, and what rollback looked like when the first anomaly appeared.

Security and compliance stakeholders need plain-language briefings: what data leaves the perimeter, what logs exist, and what happens on failure.

Leaders who invest in evaluation harnesses this quarter usually spend less on firefighting next quarter—especially when defaults shift underfoot.

Operational teams should treat vendor announcements as triggers for regression tests, not as finish lines.

Document who approved production changes, which prompts were pinned, and what rollback looked like when the first anomaly appeared.

Security and compliance stakeholders need plain-language briefings: what data leaves the perimeter, what logs exist, and what happens on failure.

Leaders who invest in evaluation harnesses this quarter usually spend less on firefighting next quarter—especially when defaults shift underfoot.

Operational teams should treat vendor announcements as triggers for regression tests, not as finish lines.

Document who approved production changes, which prompts were pinned, and what rollback looked like when the first anomaly appeared.

Security and compliance stakeholders need plain-language briefings: what data leaves the perimeter, what logs exist, and what happens on failure.

Leaders who invest in evaluation harnesses this quarter usually spend less on firefighting next quarter—especially when defaults shift underfoot.

Operational teams should treat vendor announcements as triggers for regression tests, not as finish lines.

Document who approved production changes, which prompts were pinned, and what rollback looked like when the first anomaly appeared.

Security and compliance stakeholders need plain-language briefings: what data leaves the perimeter, what logs exist, and what happens on failure.

Leaders who invest in evaluation harnesses this quarter usually spend less on firefighting next quarter—especially when defaults shift underfoot.

Operational teams should treat vendor announcements as triggers for regression tests, not as finish lines.

Document who approved production changes, which prompts were pinned, and what rollback looked like when the first anomaly appeared.

Security and compliance stakeholders need plain-language briefings: what data leaves the perimeter, what logs exist, and what happens on failure.

Explore More

Discover more resources

Browse y/our curated collection of tools, guides, and resources to help you build, grow, and scale.

Browse Blog

Share Your Story

Have a story to tell?

Join the network and share your journey. Your experiences can inspire and help others on their path.

Share Your Story