Assess practical performance and occupational competence
As of 2026-10-09, AI does this work at L3 Conditional automation. 59 updates bear on it. Strongest evidence: Confirmed by several sources.
- Level
- L3Conditional automation
- Updates
- 59
- Companies
- 9Google · Anthropic · Alibaba
- Strongest evidence
- T2Confirmed by several sources
Held by a non-technical condition: Physical Presence. The level above is what the evidence supports. Whether the work may be handed over is a separate question.
What moved it
Every update that bears on this work, newest first.
- 2026-10-08GoogleAMIE matches doctor diagnoses in 90% of 100 patient interactions with 0 safety stops
Google Research reports that in a Lancet-published evaluation with BIDMC Medicine, AMIE had 0 safety stops and matched doctor diagnoses in 90% of cases across 100 patient interactions in a real clinic.
- 2026-10-08AnthropicAstrophysicist uses Claude Science to build first complete ultraviolet sky map
Anthropic says astrophysicist Brice Ménard worked with Claude Science to find existing datasets, combine them and fill gaps with statistical inference, producing the first complete ultraviolet map of the sky in a few days.
- 2026-10-08AlibabaIgnite Vision builds AI creator matching on Qwen, cutting screening from hours to minutes
Alibaba Cloud reported that Ignite Vision built AI creator matching on Qwen, reducing KOL screening from hours of manual work to minutes across millions of profiles.
- 2026-10-07MicrosoftClaude Haiku 5.5 generally available in GitHub Copilot
GitHub announces Anthropic's Claude Haiku 5.5 is now generally available in GitHub Copilot, a lightweight model for fast, high-volume work like subagents, quick edits and terminal tasks; early testing showed it matched Claude Sonnet 5 on many coding tasks with fewer tokens and steps.
- 2026-10-07GoogleGoogle Research shares RCT findings on AI and professional expertise
Google Research announces new findings from Visiting Fellow David Autor on how AI impacts how professionals build expertise, based on a three-month randomized controlled trial with practicing patent attorneys testing short-term productivity and longer-term skill building.
- 2026-10-07GoogleGemini in Google Docs can convert briefs into presentation decks
Gemini in Google Docs can now convert a written brief into an on-brand presentation deck grounded in the user's work files, emails, and chats.
- 2026-10-06MicrosoftGitHub releases ReviewBench for AI code review agents
GitHub introduces ReviewBench, an open benchmark for AI code review agents, built from analysis of 103.9M pull requests with 219 PRs across 19 languages.
- 2026-10-02MetaMeta shares six papers from mathematicians working with Muse Spark
Meta reports that mathematicians used Muse Spark 1.1 and 1.2 in Thinking Mode to find solutions to open problems, producing six papers.
- 2026-10-02ByteDancePCG uses Seedance 2.5 for Saudi Arabia Railways film
PCG created an AI-powered film for Saudi Arabia Railways using ByteDance's Seedance 2.5 to recreate historical moments.
- 2026-10-01MetaMeta releases Immersive Web SDK version 1.0
Meta releases Immersive Web SDK version 1.0, enabling agents to play, profile, and fix apps on device until frame rate targets are met.
- 2026-09-30GoogleGoogle science AI model ranks highest in CDC flu forecasting
The CDC announces that of 39 eligible models, Google's science AI model ranked highest for forecasting flu-related hospital admissions during the 2025-26 flu season.
- 2026-09-25AnthropicAnthropic Science Blog: Claude can do Nine Loops
Anthropic's Science Blog discusses whether Claude can compute nine-loop scattering amplitudes, which are notoriously hard to calculate.
- 2026-09-25GoogleRemento uses Google Workspace with Gemini for customer insights
Google highlights how Remento uses Google Workspace with Gemini to turn weekly customer conversations into actionable insights.
- 2026-09-24OpenAIHarvey uses GPT-6 Astra for structured legal drafts
OpenAI states that Harvey uses GPT-6 Astra to turn stacks of documents into structured legal drafts.
- 2026-09-24GoogleStitch supports brand kit and product shots in prompts
Google's Stitch now lets users drag brand kits and product shots into prompts, with Gemini 3.8 Flash extracting design systems and generating cohesive layouts around brand rules.
- 2026-09-21GoogleGemini in Sheets builds spreadsheets from prompts
Google Workspace Intelligence brings Gemini in Sheets the ability to create fully formatted, editable spreadsheets with native formulas and charts from a prompt.
- 2026-09-19AlibabaAlibaba releases Qwen3.8-LiveTranslate real-time interpretation model
Qwen releases Qwen3.8-LiveTranslate, a next-generation real-time simultaneous interpretation model built on an Interleave architecture, reducing average lagging (LAAL) from 2.8s to 2.3s across 60 languages.
- 2026-09-18GoogleGoogle releases Android Bench 2.0 for long-horizon engineering tasks
Google introduced Android Bench 2.0, an updated benchmark evaluating long-horizon tasks such as building apps from scratch, migrating cross-platform codebases to Android, and complex architectural transitions.
- 2026-09-17OpenAIChatGPT now available in Microsoft Word
OpenAI adds ChatGPT to Microsoft Word, enabling users to turn rough notes into drafts, proofread, get suggested edits, and spot formatting issues without leaving the document.
- 2026-09-15OpenAICognition's Devin uses GPT-6 Astra for test generation
OpenAI Devs states that GPT-6 Astra helps Cognition's Devin back up 'it works' with tests before the team ships.
- 2026-09-15ByteDanceFortal Studio uses BytePlus for generative short-drama production
Fortal Studio worked with BytePlus to generate consistent shots across 15 to 40 episode short dramas.
- 2026-09-15MiniMaxMiniMax H3 demonstrates pencil drawing capability
MiniMax H3 is shown to be capable with pencil drawing.
- 2026-09-15AlibabaVenice AI uses Wan 3.0 in real creative workflow
Jordan Urus from Venice AI and Johnny Mai from Alibaba Cloud discuss how Wan 3.0 is used in a real creative workflow with iterative prototyping from 360p to 1080p.
- 2026-09-14MiniMaxMiniMax H3 achieves over 2x real-time denoising on 8x B200
MiniMax H3 with sgl_project and VDN-H3 achieves beyond 2x real-time denoising on 8x B200, generating 14.4s of 768p video in 9.0s end-to-end after warmup with no measured quality regression.
- 2026-09-14MicrosoftGitHub legal operations manager builds internal tool with Copilot CLI
Victoria, an operations manager on GitHub's legal team, used GitHub Copilot CLI to build a centralized platform for operational tools without engineering resources.
- 2026-09-14OpenAIPerplexity engineer uses GPT-6 Astra in Codex for end-to-end testing
An engineer at Perplexity reports using GPT-6 Astra in Codex to build test harnesses and mock third-party API responses for end-to-end testing.
- 2026-09-11OpenAIYelp uses GPT-Live-1 for natural restaurant reservation calls
Yelp is using GPT-Live-1, which listens while it speaks, to make restaurant reservation calls feel more natural by allowing callers to interrupt and change direction mid-conversation.
- 2026-09-11GoogleGemini in Google Workspace can extract data from Drive folders into Sheets
Google Workspace users can now select sources in Google Drive and prompt Gemini to extract data, calculate metrics, and build a Sheet without leaving their workflow.
- 2026-09-11GoogleGoogle DeepMind uses AI to reconstruct memory for documentary 'Love, Rendered'
Google DeepMind paired restored archival photos with pose control models to capture mannerisms and micro-expressions for the documentary 'Love, Rendered'.
- 2026-09-11MiniMaxHailuo AI game trailer teaser
MiniMax's Hailuo AI posts a teaser comparing output to a real game trailer, aimed at game developers, without naming a product.
- 2026-09-10MetaMeta highlights Muse Code and Immersive Web SDK for VR development
Meta for Developers showcased building a VR rhythm game with Muse Code and the Immersive Web SDK (IWSDK), reporting $1.75 in model usage and about four hours of agent runtime.
- 2026-09-10OpenAIOpenAI publishes GPT-Live-1 production voice agent benchmark results
OpenAI shared benchmark results for GPT-Live-1 covering task completion, multi-turn conversation and turn-taking, response latency, and tool use for production voice agents.
- 2026-09-09OpenAIOpenAI shares cyber defense architecture and playbook
OpenAI describes mobilizing 250+ people to strengthen defenses across hundreds of systems, with cyber models finding and fixing vulnerabilities, and shares the architecture and a practical playbook.
- 2026-09-09xAIGrok adds Coinbase connector for trading and portfolio management
Grok can now trade and manage a portfolio on Coinbase via a connector, letting users check balances, analyze data, and place or cancel trades from chat.
- 2026-09-09MicrosoftGitHub Copilot can approve pull requests after review
GitHub announced that users can now get an approval as soon as Copilot reviews their pull request, instead of waiting for a human sign-off.
- 2026-09-08MetaMuse Spark 1.3 now available in Cursor
Meta announces that Muse Spark 1.3 is now available in Cursor, built for long-horizon coding tasks with strong agentic performance on CursorBench.
- 2026-09-06GoogleFOX Sports used AI for World Cup broadcast coverage
FOX Sports used AI on Google Cloud to handle its World Cup coverage across 16 cities, 3 countries, 48 teams and 104 games.
- 2026-09-05GoogleVirgin Media O2 deploys multi-agent system on Google Cloud
Virgin Media O2 built a multi-agent system using Gemini models, BigQuery, Pub/Sub, and Gemini Enterprise Agent Platform, achieving a 9.2% reduction in call handling time and a 26% boost in Net Promoter Score.
- 2026-09-04AnthropicClaude completes first formalized proof of Fermat's Last Theorem
Anthropic said Claude completed the first formalized proof of Fermat's Last Theorem, converting the reasoning into a form Lean can verify.
- 2026-09-04MicrosoftMicrosoft reports HydraFusion results on Terminal-Bench 2.1
Microsoft said Project HydraFusion achieved 4.9 percentage points higher verified task quality and 67% lower estimated cost versus Claude Opus 5 on Terminal-Bench 2.1 in controlled offline evaluations.
- 2026-09-04GoogleCognizant uses Google Workspace and Gemini Enterprise to cut dev timelines
Cognizant's Senior Director describes how Cognizant uses Google Workspace, Gemini Enterprise and AI agents to reduce software development timelines from 16-18 weeks to a few days.
- 2026-09-04AlibabaWan and GMI Cloud AI filmmaking challenge in Taipei
Wan and GMI Cloud hosted an AI filmmaking challenge in Taipei with over 60 professionals across 19 teams, producing 17 short films using Wan3.0.
- 2026-09-03OpenAIOpenAI introduces GPT-6 Astra
OpenAI announced GPT-6 Astra, a model for computer use, software engineering and visual understanding, with the ability to perform computer tasks on the user's behalf.
- 2026-09-03MicrosoftGemini 3.8 Flash available in GitHub Copilot
Google's Gemini 3.8 Flash became available in GitHub Copilot, with early testing showing strong performance on complex terminal-based coding tasks.
- 2026-09-03AlibabaQwen introduces E-Commerce Bench benchmark
Alibaba's Qwen team presents E-Commerce Bench, a benchmark for long-horizon autonomous business operations where agents run online stores for 365 days with ¥100,000 starting capital.
- 2026-09-02AlibabaQwen3.8-Max-0902 ranks #1 overall on Code Arena
Qwen3.8-Max-0902 takes the #1 overall spot on Code Arena and tops the Pareto frontier at $5/MToken.
- 2026-09-02AlibabaQwen3.8-Max-0902 tops CodeArena WebDev at 1691
Qwen3.8-Max-0902 ranks #1 on the CodeArena WebDev leaderboard, rising from 1669 to 1691 and setting a record for agentic coding workflows.
- 2026-09-01xAILatchBio evaluates Grok 4.6 on biosecurity tasks
LatchBio evaluated Grok's performance on biosecurity monitoring and adversarial biological tasks, finding Grok 4.6 detects and refuses dangerous queries while allowing beneficial scientific queries.
- 2026-09-01AlibabaQwen3.8-Max tops CommerceAgentBench among open-weight models
Alibaba reports that Qwen3.8-Max delivers the strongest overall performance among open-weight models on CommerceAgentBench, a benchmark built from real commercial demand.
- 2026-08-27AlibabaWan3.0 ranks #1 on Artificial Analysis video leaderboards
According to the latest Artificial Analysis leaderboards, Wan3.0 is ranked #1 in Video Editing (with Audio) and #1 in Text-to-Video (with Audio).
- 2026-08-26OpenAIDeveloper ships npm library built with Codex
OpenAI highlighted a developer who used Codex to ship her own npm library for loading experiences in AI apps.
- 2026-08-26GoogleTuring uses Gemini to automate 60k+ HR tickets
Turing's VP of Talent shares how the company uses Gemini to automate over 60,000 manual HR tickets and improve human connections.
- 2026-08-26AlibabaQwen3.8-27B ranks #1 among open models on Image-to-WebDev Arena
Alibaba's Qwen3.8-27B placed first among open models and seventh overall on the Image-to-WebDev Arena with 27B parameters.
- 2026-08-25MiniMaxMiniMax-M3 cited as cheapest model on agent benchmark
MiniMax highlights a real-world agent benchmark where MiniMax-M3 costs $0.018 to spin up an inbox and write and send a business email end-to-end, the cheapest model on the board.
- 2026-08-25AlibabaQwen3.8-27B ranks 9th overall on Code Arena
Qwen3.8-27B places 9th overall on Code Arena, the only model in its size class in the top 10.
- UndatedGoogleCooley builds confidential redaction agent on Gemini Enterprise for Legal
Cooley is building a confidential redaction agent on Gemini Enterprise for Legal that automates redactions of personally identifiable information before filings go public.
- UndatedGoogleHitachi deploys agents to frontline workers via HMAX for 200+ use cases
Hitachi is deploying agents directly to frontline workers through HMAX; field technicians have built agents for over 200 use cases, boosting productivity by up to 30%.
- UndatedGoogleOn migrates 24 core enterprise services to Google Cloud with agent-led strategy
On transitioned 24 core enterprise services to Google Cloud using an agent-led strategy, cutting migration time from 3 months to 2 weeks.
- UndatedGoogleGemini ML skills write and test PySpark code in notebooks
Equipped with ML skills, Gemini writes and tests PySpark code in notebooks, trains models, and fixes broken pipelines inside isolated, governed cloud sandboxes.