Pinecone 2.0 — Edo Liberty, Pinecone

Englishالعربية Deutsch Español Français हिन्दी Bahasa Indonesia 日本語 한국어 Português Русский 中文

컴퓨터/소프트웨어AI/미래기술

스크립트

00:00:00interface I want to there are many many cool ideas that make this work but I

00:00:06want to tell you a little bit about just one of them okay what you're seeing is

00:00:11the interface for Nexus okay you will see at the top a question that the the

00:00:17agent issued and you still I see on the left the different steps that it went

00:00:23through okay what you see on the bottom right is probably the most interesting

00:00:28you see generated code on the bottom right what Nexus does which is very

00:00:34different than other systems it's what we call a runtime coding agent okay unlike

00:00:40other coding systems in the you're used to where the task is take a large code

00:00:49base and then help me edit it right but when the edits are done what I have is a

00:00:55piece of software and that is deployed that is the artifact okay what is running

00:01:00does not contain the model anymore it's just the code okay this is not what's

00:01:05happening here the in query time you should think about the engine essentially

00:01:10building something like a Jupiter notebook something like like a Python repo right

00:01:16and they literally write code and execute it and write code and execute and if they

00:01:20get the answer they know what to do with it and if it's you know if it's not what

00:01:24they expected they rewrite that piece of the code and what you get in the end is a

00:01:30is a piece of code that completes the task that gets the information that you want to

00:01:36get okay that is save-able that's rerunable okay but it's also incredibly flexible

00:01:42because now the the answer is got not by by a query to a database it's right it's

00:01:50written by code which is in of course incredibly flexible okay that code is by the way

00:01:57interestingly enough if you replace if you write software this way the the amount the

00:02:05the amount of prompting that you need is significantly reduced we went down from

00:02:08having something like a hundred and fifty thousand tokens to give our agents all the

00:02:12tooling they need to less than a thousand tokens to specify all the

00:02:16interfaces they need to get information out of Nexus okay and so the answer the so what

00:02:23it looks like when you run it is like this okay so on the left of course you'll see

00:02:29Nexus with no QL answering the question and then on the right is so is the same task

00:02:38given to an agent with all the tooling to complete the task okay but remember this

00:02:43on the right it's severely handicapped it's like the employee on the first day

00:02:48on the job they have to read a bunch they have to figure out where the data is

00:02:51they have to figure out what the task is they have to write a bunch of code it's

00:02:55just it's a severely handicapped and this I don't know if you okay stopped but

00:03:02needless to say this is by far slower it's a lot more expensive in terms of

00:03:08token consumption but interestingly enough it's also a hell of a lot less

00:03:12accurate because again they don't have the right context the right objectives and

00:03:17so on in your company I'll just say that we work obviously with a bunch of

00:03:25early access customers already you see some results here but I'll just sort of fly

00:03:30through them and say that across different domains across different

00:03:35enterprises across different kinds of tasks you will universally see the same

00:03:40thing that if you shift to this paradigm you get significant cost savings 77 90 you

00:03:50know percent token token reduction is pretty common it gets a hell of a lot

00:03:58faster anywhere from 20 30 percent to sometimes 77 faster okay but and the

00:04:07most important thing is it actually becomes a lot more accurate as well in the

00:04:11same time so this is really it's a slam dunk I'll just wrap up by saying that the

00:04:20nexus is coming out of early access and into public preview literally tomorrow so go try it out

00:04:50Thank you.

핵심 요약

Pinecone's Nexus shifts agent architectures from heavy prompt-based tooling to a runtime coding engine that generates and executes code directly, reducing token consumption by up to 90% and improving response speeds by up to 77%.

하이라이트

Nexus uses a runtime coding engine that generates and executes code on the fly during query time.
Switching to runtime coding interfaces drops token overhead from 150,000 tokens down to under 1,000 tokens.
The runtime coding architecture cuts token consumption costs by 77% to 90%.
Query response speed increases by 20% to 77% compared to traditional tooling agents.
Nexus moves from early access to public preview.

타임라인

Runtime Coding Agent Mechanics

Traditional coding agents output software artifacts meant for deployment, whereas Nexus generates and executes code in real time during the query.
The query engine operates like an automated Jupyter notebook by iteratively writing, executing, and rewriting code until it arrives at the correct result.
Replacing pre-prompted tool descriptions with runtime interfaces reduces prompt size from 150,000 tokens to less than 1,000 tokens.

Standard software-editing agents edit a codebase and output a final script that runs without the underlying model. Nexus instead functions as a runtime coding agent that keeps the model active inside the query execution loop. It constructs temporary Python environments to run code, analyze outputs, and refactor failing steps immediately. This programmatic access eliminates the need to pass thousands of tokens of static tooling descriptions into the context window.

Performance and Cost Comparison Across Enterprise Workloads

Agents relying on massive pre-loaded context and tool definitions run slower, cost more per query, and suffer from lower accuracy.
Early access enterprise deployments show token cost reductions between 77% and 90%.
Execution speeds improve between 20% and 77% alongside higher answer accuracy.

Providing agents with static tooling descriptions forces the system to re-read context repeatedly, mimicking an employee on their first day trying to navigate unfamiliar data. This redundant processing causes slow execution speeds, high token expenses, and lower output precision due to missing exact objectives. Moving to dynamic code generation solves these constraints by executing precise tasks on demand, leading to simultaneous improvements in speed, cost, and accuracy across enterprise domains.

커뮤니티 글

아직 글이 없습니다. 이 영상에 대한 첫 번째 글을 작성해 보세요!

이 영상에 대해 글쓰기