Podcast Atlas
entity Updated 2026-08-24 Topics: Technology

枫言枫语

枫言枫语 is the podcast/show context for Vol. 160 一年多以后，再聊AI写代码Vibe Coding, Vol. 161 从开发自己的 OpenClaw 聊起, Vol. 162 科技快乐星球44: 新模型“SOTA们”齐贺新春, Vol. 164 从苹果聊到软件未来：Agentic Software 真的要来了？, Vol. 165 做客声东击西：「龙虾」和 vibe coding 正如何改变我们的思维, Vol. 166 闲聊: 从 Gemini 到 AI 的加速与混沌, Vol. 169 高考只是个开始，Don’t Waste Your Life, Vol. 170 Fable 5 重出江湖，GPT 仍需努力, and Vol. 167 Token 如流水，Agent 似朝阳. Vol. 160 revisits AI coding after a year of tooling change, using NewSpot to connect Vibe Coding, Claude Code, YOLO permissions, AI search, testing, human taste, and the need to slow down. In the OpenClaw episode, hosts Justin Yan and 自立 move from a planned “科技快乐星球” topic into a full discussion of Open Claw, self-built personal agents, Vibe Coding, and agent-native product design. Vol. 162 returns to that roundup format, using new model releases, Xcode Agentic Coding, Codex/Claude Code comparison, Gemini, AI video, shopping agents, and cloud-chip infrastructure to add Model Workflow Fit and Agentic Commerce. Vol. 164 uses Apple and App Store as a platform entry into Agentic Software, Atomic Capability Services, Tencent Meeting, AI-era communication, and generated-content devaluation. Vol. 165 adds the 声东击西 crossover, where Justin Yan, 徐涛, and 王俊玉 use “龙虾” and vibe coding to connect personal agents with non-technical builders, internal AI hackathons, and digital-employee management. In Vol. 166, they widen the lens to Gemini, coding agents, AI Product Fragmentation, AI Workforce Monitoring, and the social effects of AI acceleration. Vol. 169 applies the same AI-uncertainty lens to College Major Choice, University Opportunity Density, and Learning How To Learn. Vol. 170 returns to hands-on workflow through Fable 5, Codex, Superpowers, GrillMe Skills, and the cost/product implications of Token-Driven Software. Vol. 167 fills in the product-news and personal-agent bridge: Apple platform AI, OpenAI/Microsoft cloud shifts, Project Glassfin, AI Content Provenance, Medical AI Marketing Risk, Codex remote operation, and IM-based Open Claw/Hermes Agent workflows.

Vol. 171 假如我们有无限 Token adds the show’s explicit Unlimited Token Workflow turn. Instead of only asking which model is strongest or which coding tool feels better, Justin Yan and 自立 ask how work changes when long-running agents, multiple subscriptions, and abundant token access make it plausible to run many coding, testing, translation, research, migration, and one-off software tasks in parallel.

Vol. 172 Codex 卖重置套餐，DeepSeek 峰谷调价，苹果重回 5 万亿等 adds the next operating snapshot. The show moves from abundant-token imagination into concrete AI Subscription Economics and Peak-Valley AI Inference Pricing: reset purchases, DeepSeek time-window pricing, OpenRouter routing, Apple Intelligence availability, agent sandbox escapes, password-manager access, and assistant-commerce commissions all become ordinary workflow questions rather than abstract AI strategy.

Source Position

  • The episode is grounded in practical personal experimentation rather than only market commentary.
  • Vol. 160 adds a retrospective, practice-heavy Vibe Coding snapshot where AI coding is useful only when paired with tests, review, product taste, and permission discipline.
  • Its contribution to the wiki is a builder-side view of Agent Native Software, AI Skills, On-Demand Apps, and Agent Permission Boundaries.
  • Vol. 165 adds a crossover view: the same OpenClaw/vibe-coding shift matters to media operators, product founders, and non-technical employees who can now prototype workflow tools.
  • The show links personal automation, software taste, AI coding practice, and commercial skepticism around AI Inference Cost Structure.
  • Vol. 164 adds the show’s software-design thesis: agents may force old SaaS products to expose reusable capability atoms while making communication, writing, and review more important human skills.
  • Vol. 166 adds a more chaotic industry snapshot where practical Agentic Workflow experience sits beside product-strategy, workplace, health, and human-connection concerns.
  • Vol. 169 adds an education and career-preparation branch: gaokao is treated as the start of a four-year learning loop, not a final optimization problem.
  • Vol. 170 adds a stronger model-capability snapshot: One-Shot AI Coding improves, but quota, review, skill selection, and product taste still decide whether the output becomes useful software.
  • Vol. 167 adds a platform-and-trust snapshot where AI Inference Cost Structure, AI Governance And Compliance, and Agent Permission Boundaries become daily product concerns rather than abstract AI strategy.
  • Vol. 162 adds an earlier model-roundup snapshot where Model Workflow Fit, Agentic Commerce, MaaS Infrastructure, and AI Plus Terminals connect coding tools, product entry points, video models, shopping agents, and infrastructure.
  • Vol. 171 adds the abundant-token workflow snapshot: Token Maxxing, Agentic Workflow, AI Use Pacing, and Human Judgment Under AI become one problem because the host can start more AI work than he can comfortably inspect, publish, or turn into finished products.
  • Vol. 172 adds the cost-and-boundary snapshot: heavy AI users now manage reset pricing, peak/off-peak API costs, model routing, browser permissions, service-entry incentives, and hardware-platform limits inside one daily tool stack.

Connections