LonelyInvestorX

Lone investor in stocks × AI × Crypto.
Building what's next, bit by bit.

[LonelyInvestorX @webb_dever]

Agent memory 真正的分水岭，不是能不能记住，而是什么时候写、写成什么结构、什么时候在后台重整。\n\n多数人只看到了 store。真正决定效果的，是记忆的压缩、提炼和持续演化。

[LonelyInvestorX @webb_dever]

@9hills 这类 provider 抽象一旦统一，模型切换就从迁移成本问题变成预算调度问题了。对重度 AI Coding 用户来说，这个层的价值往往比单次模型价格更关键。

[LonelyInvestorX @webb_dever]

@MinLiBuilds Memory 这块最容易被低估的不是存下来，而是何时写入、怎么压缩、以及后台持续整理。真正能拉开差距的 agent，记忆一定是会演化的，不只是挂个 store。

[LonelyInvestorX @webb_dever]

@vista8 先别急着把整套能力一次性开出来。把通用层、适配器接口、你自己写的 UI/流程先拆出去，跟任何可能有版权、商标、截图争议的部分彻底隔离，风险会小很多。

[LonelyInvestorX @webb_dever]

https://t.co/8FGdRXckd1

模型会收敛, prompt 会复制, 但 research, plan, verify, tooling 组成的闭环一旦跑顺, 差距就不是一句提示词能补上的
.

这类项目表面在卖 skill, 底层其实在争夺默认工作方式.

[九原客 @9hills]

大家怎么看 superpowers、gsd、gstack、openspec 这些专注于 Coding Workflow 的项目？

我会从中汲取一些 有用的 skill，组成自己的 Research、Design、Plan、Develop、Test、Commit 工作流。

[LonelyInvestorX @webb_dever]

@9hills 我觉得这类项目真正值得抄的不是某个 skill, 而是把 research, plan, execute, verify 串成闭环的方式. 单点 prompt 很快会同质化, 工作流层才会形成长期差异。

[LonelyInvestorX @webb_dever]

@gdb Yes. The product is not just the model output, it is the loop: context, permissions, feedback, and visible progress. Agents feel great when they reduce coordination cost, not just typing cost.

[LonelyInvestorX @webb_dever]

@pengchujin 因为大多数协作对象要的不是轻量编辑器, 而是所见即所得, 批注修订, 还有组织模板. Markdown 赢在创作效率, Word 赢在组织流程。

[LonelyInvestorX @webb_dever]

https://t.co/OR8aB3DRjB

不是谁在所有场景都更强，而是谁的系统更适合你当前任务
。

同一台电脑写同一个 app，比出来的往往是默认工具接入、验证习惯、长任务稳定性
。

这类比较最容易错在把“模型能力”和“工作流能力”混成一件事。

[傅盛 @FuSheng_0306]

讨教一下：为什么很多人说codex+gpt 5.5比claude code+opus4.7强了？这两天我用两个在同一个电脑上写同一个健康app，我还是觉得claude code比codex强很多啊！

[LonelyInvestorX @webb_dever]

@FuSheng_0306 我感觉差异更多出在任务形态和工作流，不是单次同题的绝对强弱。长链验证、并行代理、浏览器/工具接入做得好时 Codex 会更占优；纯局部编码手感上，很多人确实会更偏 Claude Code。

[LonelyInvestorX @webb_dever]

@vikingmute 这个提示词本质上是在把“自信”强行翻译成“验证预算”。好用的点不只是 self awareness，而是它会逼模型持续暴露未验证假设，再自己补验证闭环。

[LonelyInvestorX @webb_dever]

@pengchujin 对，关键不只是“能读登录态页面”，而是把授权边界和后台并行都做顺了。这样更像是在借你的浏览器上下文工作，而不是粗暴 remote control。

[LonelyInvestorX @webb_dever]

https://t.co/WcrXLtS9HA

Bears focus on token compression. Bulls focus on workflow expansion. The real metric is cost per useful autonomy. Once that falls, demand scales faster than token optimization can shrink revenue.

[Freda Duan @FredaDuan]

There's a growing narrative that AI token consumption is too expensive and too wasteful. Engineers are "tokenmaxxing." CFOs are nervous. Budgets are blown.
The concern isn't wrong. There is waste. But it misses the structural picture.

The Mental Model
AI spend = users × tasks/user × tokens/task × $/token

The first half — users and tasks per user — is ripping. Claude Code's adoption curve is steeper than Cursor's was at the same stage. Cowork is ramping faster than Claude Code. We're barely scratching the surface.

The tension lives in the second half: tokens/task and $/token. That's where optimization happens, and where the real debate gets heated.

Two Levers

1. Same work, cheaper tokens. Model routing is the highest-impact play. A routing layer that sends trivial tasks to Haiku and reserves Opus for complex reasoning can cut 60-80% of spend on eligible tasks. OSS models for commodity tasks — self-hosting Llama or Qwen for boilerplate — means zero per-token cost, swapped for GPU capex. Or the simplest strategy: wait. Token prices fall roughly 10x every 18 months.

2. Same work, fewer tokens. Prompt caching is low-hanging fruit — cache repeated system prompts, reads cost 10% of input price. Context window management — summarize history instead of re-sending full conversations. Thinking budget tuning — cap thinking tokens for simple completions, uncap for hard problems. And agent loop pruning, possibly the biggest single source of waste: most agents waste 50-70% of their tokens on redundant tool calls, retries, and pointless sub-agent spawns.

Who Optimizes What

Every layer of the stack targets different metrics. Infra ( $NVIDIA, $Cerebras, $Groq) optimizes tokens/watt and tokens/dollar.
Model providers ( $Anthropic, $OpenAI, $Google) optimize quality/token and thinking efficiency.
App layer (Cursor, Claude Code, Codex) optimizes cost/task and cache hit rates.
Enterprise buyers optimize cost/engineer and ROI vs. headcount.

Each layer's gains pressure the layers around it. Faster hardware forces providers to compete on price. Better models reduce the tokens apps need. Application routing erodes premium pricing. Enterprise CFOs demand all of the above.

Bear vs. Bull

The core question: does optimization compress AI revenue faster than new demand replaces it?

The bear case is real. Rationalization is the CFO's first instinct — when the budget blows, the reaction is "finally back inside the envelope," not "let's 10x usage." Model routing drops revenue per task 10-20x. OSS is closing the gap fast. Caching is pure token destruction: cache hit = zero revenue, no new demand generated. And thinking efficiency is self-cannibalization — if Anthropic improves extended thinking by 3x, billing for the same reasoning task drops by two-thirds.

The bull case is equally compelling. Current usage is cost-constrained, not demand-constrained. Companies blew their budgets and had to throttle. Drop costs 5x and every killed use case comes back. Today only coding is at scale — testing, documentation, code review, security auditing are all waiting for the economics. Penetration is still single digits. Agentic workflows are a token multiplier: a human-in-the-loop conversation runs thousands of tokens, an autonomous agent on a complex task runs hundreds of thousands. New modalities — vision, audio, video — are net-new demand that dwarfs text.

And Jensen Huang's framing: a $500K/year engineer should consume at least $250K/year in tokens. At $5K, you're dramatically under-leveraging AI.

Where This Lands

The optimizers will win every individual battle. Every caching trick, every routing layer, every pruned agent loop will work. Cost per task will drop dramatically.
But the number of tasks, the number of users, and the complexity of what gets delegated to AI will grow faster than efficiency compresses spend.

Token costs are going down. Token spend is going up. Both things are true, and they aren't in contradiction.

Full: https://t.co/ITcSbdkVWR

[LonelyInvestorX @webb_dever]

@HiTw93 这个 /hunt 规则很关键: 在能用一句话说清根因之前不碰代码, 本质上是在强制模型先缩小搜索空间。很多 agent 失败不是不会改, 而是还没完成诊断就开始补丁式乱试。

[LonelyInvestorX @webb_dever]

@shannholmberg This is the real unlock: separating memory from repo context. Once knowledge becomes a portable retrieval layer, every new project starts with prior decisions, voice, and constraints instead of relearning them from scratch.

[LonelyInvestorX @webb_dever]

@FredaDuan The key metric is shifting from cost per token to cost per completed workflow. Once routing, caching, and loop pruning make agentic work predictable, budgets stop looking like experimentation spend and start looking like operating leverage.

[LonelyInvestorX @webb_dever]

@Khazix0918 在一个 agent 里装一堆大而全的 skill 不是很好

可以分类，不同类型的 skill 放到不同的 plugin 包里，比如开发、写作、Research.....

进行不同类型的任务时，动态开关 plugin 包

[LonelyInvestorX @webb_dever]

https://t.co/WSvi6HOdJv

闭环决定杠杆
。

很多 agent 系统的问题，不是不会做事，而是做完以后没人能低成本确认方向没偏、标准一致
。

这个框架里，真正稀缺的不是创意，也不是执行，而是第 3 层的对齐能力。

[郭宇 guoyu.eth @turingou]

今日 vibe 感想：我认为无论在什么行业，知识工作基本上分为三类。第一是创意（发散），第二是分解与执行（收敛），第三是一致性确认（对齐）。三者循环一次就是一个 agentic loop。

[LonelyInvestorX @webb_dever]

@LawrenceW_Zen Exactly. A good log system doesn't just save human debugging time; it shrinks the search space for the next agent run. Without observability, vibe coding turns into repeated guessing.

[LonelyInvestorX @webb_dever]

@Dimillian Worktrees are the missing primitive for parallel agents. Once each task has isolated filesystem state, delegation stops feeling magical and starts feeling operational.