Video wird geladen...

Video konnte nicht geladen werden

Beim Laden dieses Videos ist ein Problem aufgetreten. Dies könnte an einem vorübergehenden Netzwerkproblem liegen oder das Video ist möglicherweise nicht verfügbar.

Today we're introducing Web Skill Factory: a pipeline that turns solved web tasks into reusable, verified code. Most agent "skills" are notes a model rereads and reinterprets every run. Ours are programs. On a WebArena subset with gpt-5.4, reusing the library moved accuracy on held-out instances from 55% to... show more

Microsoft AI Frontiers

12,193 subscribers

31,189 Aufrufe • vor 2 Monaten •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

13 Kommentare

Profilbild von Microsoft AI Frontiers

Microsoft AI Frontiersvor 2 Monaten

Every task a web agent solves leaves a working script behind. Web Skill Factory distills those scripts into a growing library of parameterized skills: code that runs without a model and composes into future tasks, instead of re-exploring a site from scratch.

Profilbild von Microsoft AI Frontiers

Microsoft AI Frontiersvor 2 Monaten

A learned skill is a standalone CLI program. It re-executes in about 40 seconds with zero tokens. Recurring tasks can just be scheduled rather than paying a model to redo the same work.

Profilbild von Microsoft AI Frontiers

Microsoft AI Frontiersvor 2 Monaten

Skills generalize past a single lucky run. We align multiple solves of the same task template, keep the shared workflow as the program skeleton, and lift the differences into parameters. That's what stops a skill from being correct but overfit to one instance.

Profilbild von Microsoft AI Frontiers

Microsoft AI Frontiersvor 2 Monaten

Two gates before anything enters the library. The original solve is checked against a gold answer when one exists, self-verified when it doesn't. Then the distilled skill has to replay standalone. Broken or non-executable skills don't get in.

Profilbild von Microsoft AI Frontiers

Microsoft AI Frontiersvor 2 Monaten

Skills also evolve as they're reused. New solves widen an existing skill in place, and regression replay confirms later updates haven't broken behavior that already passed.

Profilbild von Microsoft AI Frontiers

Microsoft AI Frontiersvor 2 Monaten

Built on WebWright. -

Profilbild von Microsoft AI Frontiers

Microsoft AI Frontiersvor 2 Monaten

Work by @demisama_ and @Adamlu28 🚀

Profilbild von 青雲

青雲vor 2 Monaten

agree with skills-as-programs over skills-as-notes — I've seen the same distinction in my harness work: a skill that carries verification (what 'done' means, how to check) is reusable; one that's just context text gets reinterpreted differently every run. the 55→70 on held-out instances is exactly the delta I'd expect. verified skills are the unit of compounding for agents.

Profilbild von AI Mastery Guide

AI Mastery Guidevor 2 Monaten

55% to 70% accuracy just from reusing skills as code is a big jump

Profilbild von Vantix AI Agency

Vantix AI Agencyvor 1 Monat

Web Skill Factory turns solved web tasks into reusable code

Profilbild von ZenithAi

ZenithAivor 1 Monat

Turning successful web tasks into reusable programs is a major upgrade

Profilbild von Elara AI

Elara AIvor 1 Monat

Turning solved web tasks into reusable verified code instead of notes is brilliant

Profilbild von Aria Tech

Aria Techvor 1 Monat

Code not notes is the leap agents actually needed

Ähnliche Videos

MOST PEOPLE USING CLAUDE CODE ARE LEAVING 90% OF ITS DESIGN SKILLS ON THE TABLE Same Claude. Same model. Completely different results. Out of the box, Claude Code builds websites that look… fine. Clean. Functional. Forgettable. But give it the right web-design skills and suddenly it starts building pages that look like a $5,000 agency made them. And here’s the crazy part: Those skills are completely FREE. They’re sitting on GitHub right now, and almost nobody is using them. I found the exact setup that turns Claude Code from an average website builder into a serious design machine. The skills + workflow are in the article below. Once you install them, going back to default Claude feels broken.

MOST PEOPLE USING CLAUDE CODE ARE LEAVING 90% OF ITS DESIGN SKILLS ON THE TABLE Same Claude. Same model. Completely different results. Out of the box, Claude Code builds websites that look… fine. Clean. Functional. Forgettable. But give it the right web-design skills and suddenly it starts building pages that look like a $5,000 agency made them. And here’s the crazy part: Those skills are completely FREE. They’re sitting on GitHub right now, and almost nobody is using them. I found the exact setup that turns Claude Code from an average website builder into a serious design machine. The skills + workflow are in the article below. Once you install them, going back to default Claude feels broken.

Romario

65,986 Aufrufe • vor 1 Monat

🆕 Introducing Etherscan Flow • Turn a maze of transactions into one clear map: how funds moved, findings to back it - every hop verified onchain • Build a flow from scratch or let an agent build it for you with the Etherscan Flow skill • Share your findings on X and tag etherscan.eth!

🆕 Introducing Etherscan Flow • Turn a maze of transactions into one clear map: how funds moved, findings to back it - every hop verified onchain • Build a flow from scratch or let an agent build it for you with the Etherscan Flow skill • Share your findings on X and tag etherscan.eth!

etherscan.eth

115,136 Aufrufe • vor 25 Tagen

New in ChatGPT: a better way to schedule tasks. Scheduled tasks are faster, more reliable, and easier to manage from the new Scheduled page. The new scheduled tasks experience is rolling out to Go, Plus, Pro, Business, and Enterprise users on web and mobile.

New in ChatGPT: a better way to schedule tasks. Scheduled tasks are faster, more reliable, and easier to manage from the new Scheduled page. The new scheduled tasks experience is rolling out to Go, Plus, Pro, Business, and Enterprise users on web and mobile.

ChatGPT

473,682 Aufrufe • vor 3 Monaten

New course: Building Coding Agents with Tool Execution, taught by Tereza Tizkova and Fra from E2B. Most AI agents are limited to predefined function calls. This short course teaches you to build agents that write and execute code to accomplish tasks, accessing entire programming language ecosystems instead of being restricted to a fixed set of tools. You'll learn to run agent-generated code safely in sandboxed cloud environments that protect your systems from harmful operations. Skills you'll gain: - Build agents that write and execute code, manage files, and handle errors autonomously through feedback loops - Run agent code safely in E2B cloud sandboxes and understand tradeoffs between local, containerized, and cloud execution - Create a data analyst agent that explores visualizes data with Pandas - Create a full-stack agent that builds complete Next.js web applications Join and build agents that code their way through complex tasks:

New course: Building Coding Agents with Tool Execution, taught by Tereza Tizkova and Fra from E2B. Most AI agents are limited to predefined function calls. This short course teaches you to build agents that write and execute code to accomplish tasks, accessing entire programming language ecosystems instead of being restricted to a fixed set of tools. You'll learn to run agent-generated code safely in sandboxed cloud environments that protect your systems from harmful operations. Skills you'll gain: - Build agents that write and execute code, manage files, and handle errors autonomously through feedback loops - Run agent code safely in E2B cloud sandboxes and understand tradeoffs between local, containerized, and cloud execution - Create a data analyst agent that explores visualizes data with Pandas - Create a full-stack agent that builds complete Next.js web applications Join and build agents that code their way through complex tasks:

Andrew Ng

204,109 Aufrufe • vor 10 Monaten

I tested Runway Agent with a simple brief: build a launch campaign for a smart notebook that turns handwritten ideas into organized digital notes, tasks and project plans. Agent helped shape the campaign direction, plan the creative, and turn it into usable assets: hero video, social ad variants, product visuals, voiceover, music and captions. Not just a notebook concept... a full campaign you can keep refining from the same conversation. Runway Agent is available now on the Runway web app: Get 30% off 3 months with code RUNWAYAGENT #MadeWithRunway #ad

I tested Runway Agent with a simple brief: build a launch campaign for a smart notebook that turns handwritten ideas into organized digital notes, tasks and project plans. Agent helped shape the campaign direction, plan the creative, and turn it into usable assets: hero video, social ad variants, product visuals, voiceover, music and captions. Not just a notebook concept... a full campaign you can keep refining from the same conversation. Runway Agent is available now on the Runway web app: Get 30% off 3 months with code RUNWAYAGENT #MadeWithRunway #ad

Leila 👩🏻‍💻

49,414 Aufrufe • vor 3 Monaten

LLMs are great for human in the loop applications, but fail at deterministic developer tasks. Interfaze (YC P26) is a new AI model that outperforms general LLMs on high accuracy tasks like: OCR, Object Detection, Web scraping, Speech-to-text, Classification and more. Congrats on the launch, Yoeven and Harsha!

LLMs are great for human in the loop applications, but fail at deterministic developer tasks. Interfaze (YC P26) is a new AI model that outperforms general LLMs on high accuracy tasks like: OCR, Object Detection, Web scraping, Speech-to-text, Classification and more. Congrats on the launch, Yoeven and Harsha!

Y Combinator

71,358 Aufrufe • vor 5 Monaten

Vida open-sourced BrowserBC, a framework that allows users to turn browser sessions into reusable skills for AI agents. > Instead of recalculating navigation on every turn, agents can follow a skill created from earlier task execution. > Vida reports a substantially higher success rate with fewer steps, via the same AI agent. Hotel booking bench? 👀

Vida open-sourced BrowserBC, a framework that allows users to turn browser sessions into reusable skills for AI agents. > Instead of recalculating navigation on every turn, agents can follow a skill created from earlier task execution. > Vida reports a substantially higher success rate with fewer steps, via the same AI agent. Hotel booking bench? 👀

🚨 AI News | TestingCatalog

25,685 Aufrufe • vor 3 Monaten

Today we are announcing Keenable AI, an AI-native index of the best human knowledge we have, starting with the open web. AI seems to know just about everything until you ask it about something you know deeply. The answer isn’t wrong, but it's just very average. We started Keenable to fix this problem: every model and every agent should be able to query, reason over, and continuously learn from the living web. Backed by a $26M Seed from Accel and Conviction. Built by the team that took on Google at Yandex Search, with researchers and engineers from Amazon AGI, X, and Perplexity. The Web Search API and Web Query Language for AI are live now. Together they make the web queryable at AI scale. Both are free until end of September. Seven days of benchmarks, stories, and launches ahead. Let your agents search.

Today we are announcing Keenable AI, an AI-native index of the best human knowledge we have, starting with the open web. AI seems to know just about everything until you ask it about something you know deeply. The answer isn’t wrong, but it's just very average. We started Keenable to fix this problem: every model and every agent should be able to query, reason over, and continuously learn from the living web. Backed by a $26M Seed from Accel and Conviction. Built by the team that took on Google at Yandex Search, with researchers and engineers from Amazon AGI, X, and Perplexity. The Web Search API and Web Query Language for AI are live now. Together they make the web queryable at AI scale. Both are free until end of September. Seven days of benchmarks, stories, and launches ahead. Let your agents search.

Andrey Styskin

553,177 Aufrufe • vor 1 Monat

Built a Skool clone with Claude Code (Opus 4.5), with auth, db + storage with the new web app builder. (Launching next week). We are building the easiest way to use the most powerful coding model (Claude Code) in the world to build mobile apps and web apps. No performative hour long agent runs that waste tokens. Just raw Claude Code running in a sandbox, pre built with skills that make frontend + backend just work. Let me know if you want to beta test.

Built a Skool clone with Claude Code (Opus 4.5), with auth, db + storage with the new web app builder. (Launching next week). We are building the easiest way to use the most powerful coding model (Claude Code) in the world to build mobile apps and web apps. No performative hour long agent runs that waste tokens. Just raw Claude Code running in a sandbox, pre built with skills that make frontend + backend just work. Let me know if you want to beta test.

Riley Brown

51,044 Aufrufe • vor 9 Monaten

Claude Code is not just a coding tool. It can also be used as a general agent in a "workspace" Let's create an "Agent Workspace" for Claude Code, and we're going to create 5 skills including: > Notion for context > Typefully for posting on X > Nano Banana for generating graphics We're going to do all this in a safe web environment on , the easiest way to use Claude Code on the browser or your phone. (Web app builder launching in 24hrs) TIMESTAMPS: 00:00 Intro 00:59 The Agent Workspace on 01:28 Adding a notion skill + Integration 07:23 Testing Notion Integration 08:17 Creating Tasks in the dashboard 09:19 Testing Tasks in the UI 13:40 Testing Multiple tasks at once 14:20 Creating Twitter Posts Panel (Typefully API) 15:44 Prompting for the nano banana skill 16:59 Testing Typefully and nano banana tasks 18:22 Let's check if typefully and nano banana work 19:54 Almost had a heart attack lol 20:59 figuring out typefully inssues 21:26 Creating "Twitter-Post-Skill" 22:23 All the Skills a Breakdown 23:37 It Works!!!!!

Claude Code is not just a coding tool. It can also be used as a general agent in a "workspace" Let's create an "Agent Workspace" for Claude Code, and we're going to create 5 skills including: > Notion for context > Typefully for posting on X > Nano Banana for generating graphics We're going to do all this in a safe web environment on , the easiest way to use Claude Code on the browser or your phone. (Web app builder launching in 24hrs) TIMESTAMPS: 00:00 Intro 00:59 The Agent Workspace on 01:28 Adding a notion skill + Integration 07:23 Testing Notion Integration 08:17 Creating Tasks in the dashboard 09:19 Testing Tasks in the UI 13:40 Testing Multiple tasks at once 14:20 Creating Twitter Posts Panel (Typefully API) 15:44 Prompting for the nano banana skill 16:59 Testing Typefully and nano banana tasks 18:22 Let's check if typefully and nano banana work 19:54 Almost had a heart attack lol 20:59 figuring out typefully inssues 21:26 Creating "Twitter-Post-Skill" 22:23 All the Skills a Breakdown 23:37 It Works!!!!!

Riley Brown

77,582 Aufrufe • vor 8 Monaten

Meet the new Kimi Browser Extension, formerly Kimi WebBridge. From your browser sidebar, you can chat with Kimi to navigate websites, fill out forms, and get things done. For repetitive tasks, record your steps once and save them as a skill. Kimi can take it from there next time. Available now on and the Chrome Web Store.

Meet the new Kimi Browser Extension, formerly Kimi WebBridge. From your browser sidebar, you can chat with Kimi to navigate websites, fill out forms, and get things done. For repetitive tasks, record your steps once and save them as a skill. Kimi can take it from there next time. Available now on and the Chrome Web Store.

Kimi.ai

74,218 Aufrufe • vor 11 Tagen

New skill: self-managed-context (make the agent's context an editable file) It explains how to build agents that decide what to keep, update, or remove from the information they use to do their work. It can archive a long log while keeping the exact error, update its progress notes, or remove outdated information. Those edits then change what the model sees on its next turn. 1- Keep the system instructions and original task protected, outside the editable file. 2- Write the remaining conversation to a file, with labels for each message. 3- Let the agent edit that file using its usual code tools. 4- After each command, read the file back and use the updated messages for the next model call. Loading the skill ( alone into a fixed harness won't create live context editing, but it can help an agent build and then operate a harness that supports it. I gave the skill to a coding agent and had it build the harness itself. The task is a long stream of server logs that doesn't fit in the window. The agent reads it in 18 chunks, about 10k tokens in total, with a 5.5k budget. It has to report one incident ticket exactly and the final value of every config key. Same model & budget, three setups: 1- Model manages its own context 2- Harness forces a summary at 75% full 3- Keeps everything The video shows a real GPT-5.4 run. - Self-managed solved it 3 out of 3. - Keep-everything overflowed 3 out of 3. - Forced summary also solved it 3 out of 3. On GPT-5.4 the self-managed agent re-processed about 24% fewer prompt tokens than the forced summary. When it edited, it cut hard, so little was left after the edit to re-process (one edit took 5,537 tokens down to 771). On GPT-4.1 it saved nothing. It edited near the top of its context but kept most of what was below, and every edit forces everything after it to be re-processed. This is a small test at about 2x context pressure. The paper goes up to 24x, but imho the video below and the skill are a good way to start understanding the technique.

New skill: self-managed-context (make the agent's context an editable file) It explains how to build agents that decide what to keep, update, or remove from the information they use to do their work. It can archive a long log while keeping the exact error, update its progress notes, or remove outdated information. Those edits then change what the model sees on its next turn. 1- Keep the system instructions and original task protected, outside the editable file. 2- Write the remaining conversation to a file, with labels for each message. 3- Let the agent edit that file using its usual code tools. 4- After each command, read the file back and use the updated messages for the next model call. Loading the skill ( alone into a fixed harness won't create live context editing, but it can help an agent build and then operate a harness that supports it. I gave the skill to a coding agent and had it build the harness itself. The task is a long stream of server logs that doesn't fit in the window. The agent reads it in 18 chunks, about 10k tokens in total, with a 5.5k budget. It has to report one incident ticket exactly and the final value of every config key. Same model & budget, three setups: 1- Model manages its own context 2- Harness forces a summary at 75% full 3- Keeps everything The video shows a real GPT-5.4 run. - Self-managed solved it 3 out of 3. - Keep-everything overflowed 3 out of 3. - Forced summary also solved it 3 out of 3. On GPT-5.4 the self-managed agent re-processed about 24% fewer prompt tokens than the forced summary. When it edited, it cut hard, so little was left after the edit to re-process (one edit took 5,537 tokens down to 771). On GPT-4.1 it saved nothing. It edited near the top of its context but kept most of what was below, and every edit forces everything after it to be re-processed. This is a small test at about 2x context pressure. The paper goes up to 24x, but imho the video below and the skill are a good way to start understanding the technique.

Muratcan Koylan

14,523 Aufrufe • vor 2 Tagen

holy sh*t. this is f**king insane. langchain just documented the missing memory layer for agents that are supposed to improve after every run. instead of stuffing everything into one giant prompt, deep agents can keep long-term memory, reusable skills and lessons from previous tasks. [it takes a few minutes to understand the loop] 1/ run the task 2/ save what actually worked 3/ load that lesson into the next run that’s how an agent stops starting from zero every single time.

holy sh*t. this is f**king insane. langchain just documented the missing memory layer for agents that are supposed to improve after every run. instead of stuffing everything into one giant prompt, deep agents can keep long-term memory, reusable skills and lessons from previous tasks. [it takes a few minutes to understand the loop] 1/ run the task 2/ save what actually worked 3/ load that lesson into the next run that’s how an agent stops starting from zero every single time.

Gipp 🦅

33,704 Aufrufe • vor 25 Tagen

Introducing Vals-Smith: turn your code base into a customized benchmark. Public benchmarks tell you which model is strongest overall, not which model is the best on your code. Vals-Smith turns your merged pull requests into real coding tasks and measures the percentage a model can actually resolve. New models ship every week. Vals-Smith tells you which one to trust with your code.

Introducing Vals-Smith: turn your code base into a customized benchmark. Public benchmarks tell you which model is strongest overall, not which model is the best on your code. Vals-Smith turns your merged pull requests into real coding tasks and measures the percentage a model can actually resolve. New models ship every week. Vals-Smith tells you which one to trust with your code.

Vals AI

46,698 Aufrufe • vor 2 Monaten

Run your coding agent on a local model, with QVAC. OpenCode now plugs into a model on your own machine through our official plugin. That changes two things. Privacy by default. Your repo, your prompts, and every command the agent runs stay on your device. No third-party model ever sees your code. For an agent that reads your codebase and runs commands, that is the point, not a nice-to-have. New workflows. With no meter and no rate limit, the agent does not have to stop. Let it loop on a task, refactor a whole repo overnight, or run a watch-build-fix loop that keeps going. Setup takes a few minutes: one line in opencode.json and OpenCode runs on a local QVAC model. In the video demo, it handles three real developer tasks back to back, fully offline.

Run your coding agent on a local model, with QVAC. OpenCode now plugs into a model on your own machine through our official plugin. That changes two things. Privacy by default. Your repo, your prompts, and every command the agent runs stay on your device. No third-party model ever sees your code. For an agent that reads your codebase and runs commands, that is the point, not a nice-to-have. New workflows. With no meter and no rate limit, the agent does not have to stop. Let it loop on a task, refactor a whole repo overnight, or run a watch-build-fix loop that keeps going. Setup takes a few minutes: one line in opencode.json and OpenCode runs on a local QVAC model. In the video demo, it handles three real developer tasks back to back, fully offline.

QVAC

11,382,936 Aufrufe • vor 3 Monaten

Introducing Agent Mode: Agentic AI is now measured in the Arena. Agent Mode can do deep research, create reports, generate images, build websites, debug code, and more. It completes more complex tasks by using tools like web search, bash in a sandbox environment, image generation, file writing, and asking follow-up questions. Frontier models are waiting for you in Agent Mode to take on real-world tasks. GPT-5.5, Claude Opus 4.7, Gemini 3.1 Pro, and top open models. Test them yourself.

Introducing Agent Mode: Agentic AI is now measured in the Arena. Agent Mode can do deep research, create reports, generate images, build websites, debug code, and more. It completes more complex tasks by using tools like web search, bash in a sandbox environment, image generation, file writing, and asking follow-up questions. Frontier models are waiting for you in Agent Mode to take on real-world tasks. GPT-5.5, Claude Opus 4.7, Gemini 3.1 Pro, and top open models. Test them yourself.

Arena.ai

539,715 Aufrufe • vor 4 Monaten

Narrative violation from Dylan on Dwarkesh: H100s are worth more today than they were 3 years ago. There's a sentiment that data center buildouts are priced into the risk of rapidly depreciating GPUs. But the models want to learn. Token prices are falling so fast that you can get more intelligence out of GPT-5.4 on a 3-year old H100 than GPT-4 on a brand new H100. Dylan Patel dwarkeshpatel

Sarah Wang

159,395 Aufrufe • vor 6 Monaten

Achieve SOTA on multimodal, complex tasks with minimal supervision! Introducing In-Context Abstraction Learning (ICAL): a method that extracts insights from visual demos and human feedback to continually teach multimodal LLM agents 🚀🤖🧠 ICAL learns from sub-optimal demonstrations and human feedback by transforming noisy trajectories into optimized, annotated examples that improve multimodal decision-making in new domains. ICAL is the first to show state-of-the-art performance across three challenging domains: we obtain a 34% relative improvement on TEACh household robot instruction following from dialogue, a 59% relative improvement on Visual Web Arena for autonomous visual web agents, and beat few-shot GPT-4V and supervised baselines on Ego4D action forecasting.

Achieve SOTA on multimodal, complex tasks with minimal supervision! Introducing In-Context Abstraction Learning (ICAL): a method that extracts insights from visual demos and human feedback to continually teach multimodal LLM agents 🚀🤖🧠 ICAL learns from sub-optimal demonstrations and human feedback by transforming noisy trajectories into optimized, annotated examples that improve multimodal decision-making in new domains. ICAL is the first to show state-of-the-art performance across three challenging domains: we obtain a 34% relative improvement on TEACh household robot instruction following from dialogue, a 59% relative improvement on Visual Web Arena for autonomous visual web agents, and beat few-shot GPT-4V and supervised baselines on Ego4D action forecasting.

Gabriel Sarch

19,340 Aufrufe • vor 2 Jahren

this work from Google engineers is pure f*cking treasure they built WikiSkill: a system that turns an agent's execution history into persistent knowledge, then compiles it into reusable skills this is basically an experience compiler for agents the loop: run tasks ➜ preserve raw traces ➜ consolidate recurring failures + successful strategies into a wiki ➜ propose one atomic skill update ➜ validate ➜ keep or rollback the clever part: skills can roll back the wiki never does successful strategies, recurring failures, rejected edits and skill impact history survive into the next iteration the result is f*cking amazing my takeaway: the next agent memory layer should distill experience into procedures an agent can execute, validate and transfer raw traces ➜ persistent knowledge ➜ validated skills ➜ a better next run save this, then select your gpt-6 astra use case ⭣

this work from Google engineers is pure fcking treasure they built WikiSkill: a system that turns an agent's execution history into persistent knowledge, then compiles it into reusable skills this is basically an experience compiler for agents the loop: run tasks ➜ preserve raw traces ➜ consolidate recurring failures + successful strategies into a wiki ➜ propose one atomic skill update ➜ validate ➜ keep or rollback the clever part: skills can roll back the wiki never does successful strategies, recurring failures, rejected edits and skill impact history survive into the next iteration the result is fcking amazing my takeaway: the next agent memory layer should distill experience into procedures an agent can execute, validate and transfer raw traces ➜ persistent knowledge ➜ validated skills ➜ a better next run save this, then select your gpt-6 astra use case ⭣

beamnxw ./

38,578 Aufrufe • vor 18 Tagen