Updated · 2 episodes · 1 show · 2 source notes
张阔 / Zhang Kuo
Overview
Zhang Kuo is the Alibaba international-business executive interviewed in E231 and revisited in E255 about agentic cross-border commerce, Accio Work, and the limits of translating frontier-model progress into reliable business execution.
Current Profile
Across the two interviews, Zhang presents B2B agents as governed operating systems grounded in trade data rather than generic assistants. E231 follows the route from a loose product idea through design, supplier selection, transaction, logistics, and after-sales. E255 sharpens the engineering and evaluation thesis: commercial capability is the product of model, harness, and context; it must be measured on real tasks and routed across models at a cost businesses can sustain.
His human-agent boundary remains stable. Agents should absorb repeated execution, information integration, monitoring, and workflow coordination, while people retain product taste, business goals, resource allocation, consequential approvals, and accountability for the result.
Key Characteristics
- Frames cross-border B2B as a high-value agent domain requiring supplier, pricing, logistics, tariff, transaction, and feedback truth.
- Describes Vertical Agent Architecture as “model × Harness × Context.”
- Uses real-work evaluations and end-state completion to distinguish model scores from commercial capability.
- Treats routing, post-training, infrastructure, and runtime design as a joint cost-and-quality problem.
- Keeps people responsible for goals, taste, payment, capital allocation, and major operating decisions.
Evidence
- Agentic sourcing thesis - E231|从B2B到A2A:Agent新基建,如何让“一人企业”做全球生意? connects Zhang to B2B to A2A, stepwise verification, business feedback loops, persistent memory, and the One-Person Company possibility.
- Architecture and evaluation - E255|模型越来越强,为什么用户没感觉?再访阿里国际站总裁张阔 records the three-layer formulation, the 107-task ecommerce evaluation, and the claimed 61% frontier-model pass rate.
- Operating boundary - both E231|从B2B到A2A:Agent新基建,如何让“一人企业”做全球生意? and E255|模型越来越强,为什么用户没感觉?再访阿里国际站总裁张阔 preserve human review for high-stakes commercial judgment and execution.
Qualifications
Both sources are interviews centered on products and systems Zhang helps lead, so growth, speed, benchmark, cost, and customer-outcome claims remain provider-reported. The supplied material does not include independent benchmark reproduction, long-term merchant controls, or detailed failure cases. E231 uses “Axio,” while E255 uses “Accio Work”; this page treats them as the same product branch without independently resolving the naming difference.
What Changed
- Added the model–harness–context architecture to Zhang’s established sourcing thesis.
- Added real-task evaluation and the distinction between autonomous completion and assisted usefulness.
- Added model routing and affordability as deployment requirements.
- Clarified the retained human role in product, allocation, payment, and business-goal decisions.
Relationships
- Alibaba - company and international-commerce platform context.
- Axio / Accio Work - sourcing and merchant-operations product associated with Zhang’s agent thesis.
- Agentic B2B Sourcing - workflow application spanning idea, supplier, transaction, logistics, and feedback.
- Vertical Agent Architecture - three-layer system formulation articulated in E255.
- Business Agent Benchmark Gap - evaluation problem illustrated by the ecommerce task set.
- B2B to A2A - broader agent-to-agent trade direction described in E231.
- Model-Responsive AI Native Organization - organization diagnostic articulated in E231.