正在加载视频...
视频加载失败
this PDF is a f*cking cheat code for anyone running coding agents Diogo Almeida, founder of Jev and co-author of InstructGPT, published a PDF on building a Jev harness the promise: > up to 200x faster > up to 400x cheaper the speed and the cost both live in... show more
102,441 次观看 • 2 天前 •via X (Twitter)
13 条评论
200x faster and 400x cheaper sounds great — but faster and cheaper than what baseline, and on which tasks? If the router sends a hard case to the small model by mistake, the cost of the wrong answer could wipe out the savings. How does Jev handle the mis-routed cases?
Worth spinning up just to test the loop separation
200x/400x 听着吓人，但 harness 的价值不在倍数，在把每次重发明的流程变固定套路：规则、记忆、验证循环写死，agent 不用每轮重新猜。但让它按 PDF 重建 setup，等于把钥匙交出去，权限验证得先锁死，不然省下的时间一次事故就还回去。你实测过 token 对比吗？
This prompt is worth checking out
jev doing the boring decisions for claude is it open source or api only?
Where is the PDF?
if p(destructive) is read as a probability, how calibrated is it? same issue as ASR confidence scores. cheap is nice, but the false negative rate on rm -rf type commands is the number that matters
As agents become more persistent, the value of their work may depend increasingly on how well they preserve context across long-running tasks. Continuity could become as important as raw intelligence.
把流程搭顺，AI干活能又快又省，真不错。
@bot digest
a PDF from an InstructGPT co-author isn't just a guide - it's a proven methodology
Interesting direction. For those speed and cost comparisons, what quality threshold and task mix were used? The routing trade-off is most useful when we can see failure rates alongside the savings.
The interesting question is where the savings come from: fewer frontier calls, less context, or fewer retries? That breakdown would show which parts of the harness generalize across coding tasks.
![this harness stack is f*cking gold TypeSafe AI + LangChain finally dropped the layer that makes agents affordable at company scale: jev running inside the langchain agent loop as middleware the problem every big team hits: each fork in the loop is another full LLM call. at 10 agents that's a bill. at 10,000 it's the budget > routing before: every task goes to the frontier model now: jev picks the least costly model that can finish it. lookups go fast, architecture goes powerful > guardrails before: the "is this command dangerous?" check sat in the closed parts of claude code, codex, cursor now: AutoModeMiddleware(tools=["bash"]). one line, any agent, blocked before the tool runs > decisions before: one LLM call per question now: every question about the same state in one request, answered in parallel > audit before: "the model decided" now: the answer, the probability and the confidence stay in agent state > adoption no rewrite. pip install langchain-typesafe and drop it into the agent you already run TypeSafe reports up to 200x faster inference and 400x lower cost than comparable LLMs on classification. the LLM still reasons and writes. jev takes the thousands of small calls in between already live: browserbase agents for fractions of a cent, a live trading agent, email triage at scale full loop mapped below: 7 agents, 4 jev gates per step count the forks in your agent loop tonigh](https://image.24vids.com/tw-2104876056893104461/amplify_video_thumb/2104866254146203648/img/J8LnvuFvwbGYaJ7N.jpg)
