Tenex 的 AI 原生研发实验：结果定价、项目记忆与企业落地的最后一公里

Tenex ties engineering compensation to delivered outcomes, but Alex says its hardest incentive problem is rewarding architecture and client work that falls outside coding tickets.

CJ describes MetaHarness as a repository-centered workflow that preserves plans, decisions, and client context alongside code, so agents can recover the rationale behind earlier work.

CJ estimates that Tenex engineers roughly doubled their output, while acknowledging that story-point criteria require quarterly recalibration as models improve and the value of work changes.

CJ favors cross-model review, while Tenex's cost discussion suggests that model escalation should be justified by a task's economic value rather than by conspicuous token consumption.

Among roughly one hundred enterprises Tenex is speaking with, Alex estimates a steep adoption drop from general-purpose model access to non-engineering agents operating in production.

CJ argues that background agents need standardized processes because inconsistent human workflows make it difficult to define correct behavior, build evaluations, and delegate work reliably.

Alex identifies data permissioning as a major enterprise bottleneck and says Tenex has productized a gateway after initially building bespoke controls for individual clients.

Alex questions whether bespoke enterprise demands permit software margins, but the host argues that proprietary data, specialized interfaces, integrations, and habits can protect vertical products from general models.

Alex says replacing SaaS should be assessed against ongoing cognitive load and business risk, while the host emphasizes faster iteration and better experiences rather than savings alone.

CJ sees automation as an opportunity to redeploy domain experts, whereas Alex warns that displacement is real and argues that employers and employees both carry responsibility for reskilling.