IT
Agent frameworks, SDKs and CLIs, open source releases, local inference and coding tools.

codex-acp standard and preview entries move to version 2.9.0
The standard and preview entries previously listed different versions in the Agent Client Protocol registry.
Last 7 daysNov 17 – Nov 21
Douzone Bizon demos AI agent for internal agency networks
Douzone Bizon described a design that keeps the language model inside the agency, while Seoul’s city government showed how AI could book a meeting room.
AI illustrationPraxi-Bot sets out a design for approving model tool actions
Once a shell and browser are connected, desktop agent developers still have to decide which actions proposed by the model can run.
Codex makes its terminal workspace a default part of the CLI
Developers can search document previews and view images and subagent transcripts alongside the main conversation.
AI illustrationBasic International unveils AI SyncOps Studio for businesses
Basic International’s AI centre uses design specifications to divide work among AI agents.
Web course covers AI agent creation and control
The course presents experiments with real agent interactions as a way to learn permissions, security, monitoring and collaboration.
Google publishes website access guidance for Google-Agent
The help document covers access for Google-hosted agents that browse websites and perform tasks at users’ request.Coinbase presents AI agent trading as a revenue source
The trading volume forecast sits alongside restoring trust as a stated condition for Coinbase's expansion.browser-use makes 15-times token pricing the default
Developers must explicitly pass llm='bu-fast' to retain the previous model and pricing.
AI agent control extends to tools, files and task execution
Tool use and file changes shift the discussion from what AI agents can do to which actions they are permitted to take.
AI booking agents face access limits beyond user approval
Final user approval for bookings and payments is a separate condition from the target service allowing an AI agent to connect.
Korean workshop targets AI agent standardisation
Korean industry, university and research experts will discuss how AI agents change technology and how to respond to international standardisation.
Fasoo AI puts agent sourcing at centre of AI strategy
Fasoo AI frames the choice between building and buying agents as part of AI transformation strategy for businesses and institutions.
Commercial TTS design extends to request volume and latency
Commercial speech synthesis design extends beyond voice quality to the volume of requests a service handles and the delay in serving them.
InfoQ opens five-week cohorts for senior practitioners
Participants test approaches in their own work and compare results with peers.
Qwen Code unifies terminal and subagent execution
A shared agent loop becomes the default, while reasoning effort choices are limited to levels supported by the active model.
Worksign launches AI service with staff approval in Korea
The service puts companies in charge of task goals and scope, with staff reviewing results and approving the work required.
Agent memory test highlights risks from stored secrets
A fake-key test raises two linked issues: removing secrets before storage and their later reuse through agent memory.
STORM Agents shares work knowledge across AI agents
The service connects chat requests to business systems and requires staff approval for actions that affect external parties.
Autotorino more than doubles automated call capacity
The increase was measured against the previous system, but the case study disclosed neither the actual hang-up rate nor the measurement period.
Google introduces AdMob agent skills for ad coding
AdMob’s coding instructions are available now, while the server connecting developer environments to account data remains in early testing.Naver adds AI Cafe answers based on membership rank
The service turns community experience into answers while preserving each member's access rights.Trusted, connected AI agents need distinct security
A DEV post challenged the security approach of treating capable, connected AI agents as ordinary software.
IBK builds AI coding agent on internal infrastructure
The Korean bank plans to widen staff participation in improving work processes through a development tool built for financial-sector security constraints.Brave Search API allows agent use but limits result storage
Monthly credits are separate from a free plan, while permission to use the API with agents does not extend to storing search results.Nota supplies models and servers for Dube's coding agent
The supply covers model optimisation and server configuration for a coding agent designed to run on financial firms' closed networks.1 week agoNov 10 – Nov 16
Databricks SQL commands target coding agents
A DEV article examines how coding agents query Databricks SQL when table data and the command-line tool are already available.
Deva design separates agent interfaces from memory
A design that separates the user interface from memory processing connects personal agent use with collective workflows.
Postman drill examines where agent-handled tokens end up
A staged drill using a fake token shifts attention from whether an agent completes a replacement to where secrets remain.
Wapzen targets AI assistance on existing WhatsApp numbers
A developer describes Wapzen as an AI agent for small businesses that want to keep the number their customers already know.
Deleted checks matter more than an AI coding agent's summary
The merge scenario in a DEV post turns on what a deleted check protected, rather than how much the agent wrote.
AI test answers can vary with the same code and prompt
A single AI test response gives quality teams little basis for judging consistency across repeated runs.
HyperS2C launches two AI agents for verification
The agents form part of a wider toolchain and platform launch, whose product count does not measure community response.
dev-diagnosis examines coding agents' unnecessary reading
The tool examines repository reading scope and token costs, while adoption requires a separate assessment of availability and commercial terms.
BPA Platform 2027 adds distributed task processing
The release spreads task execution across servers and expands ERP connectors and managed hosting options.
Meta Muse request share does not measure web market share
One security firm's observations show AI agent activity, but do not establish user numbers or purchase outcomes.
StareBrain details screen reading and action dispatch
The development series moves from user input and plan validation to how the Android agent reads screens and sends interface actions.
Guarapay’s AI card raises spending authority questions
Guarapay’s freeze, approve, refreeze proposal separates an agent’s ability to execute payments from its permission to spend.
Copilot CLI adds regex rules for enterprise shell access
The update expands command permission rules, fixes the order of follow-up instructions and preserves saved settings when configuration files cannot be read.
ElevenLabs unveils ElevenAI meeting app for teams
ElevenAI can use meeting records for questions and email drafts, with preview access requiring an early-access application.
n8n Labs releases Nod personal agent as a research preview
Nod is presented as a personal agent with its own computer and browser, but adoption depends on its terms of use and where it runs.
Google releases AQuA to diagnose AI agent failures
AQuA links recurring failures to deployed code, with session replay used to check prompt changes.
Stopping the model can leave an AI app running
Tasks that can persist after model shutdown show why containment testing needs to cover the whole AI app.
agent-browser targets checks on AI-generated pages
A DEV Community article presents browser interaction and result checking after code generation as the tool's purpose.
Block offers AI agents as teammates for businesses
Offering internal tools to businesses makes it necessary to distinguish agents’ collaborative roles from users’ rights to use them.
Spider Networks unveils AI agent rules hub
KUKURI lets users select rules for their work instead of writing complex AI instructions from scratch.
Google launches Hydrogen for faster coding and agent tasks
The reported task-speed gain uses an earlier lightweight model as its baseline, while the overall performance comparison uses a higher-tier model.
Codex adds terminal chat tabs and restart recovery
Users can reopen saved side chats and search older tasks by name, prompt preview or working directory.
ETRI develops Trinity to assess user-aware task planning
Trinity assesses whether a planned action suits the person receiving the service, beyond whether it completes the instruction, AI Times reported.
AI memory test checks what survives a correction
An evaluation proposal checks whether correcting one fact preserves other memories and carries through to a new conversation.
AWS introduces UAEF to unify AI agent evaluation
UAEF converts execution records from different frameworks into a common format so teams can compare results using shared metrics.
Replit users bear responsibility for Agent-generated code
A security checklist described Replit’s division of responsibility between users who own generated code and the company that protects the platform.
Google web guidance adds code audits for AI coding agents
The expansion of a free skill set should be distinguished from evidence that its checks work.
Meshy DCC Core fixes agent disconnections in `0.2.4`
The update prevents retries when a request's result is unknown and checks app readiness through the path used to send requests.
QCon AI New York programme covers agents and code review
Engineers’ accounts of running AI systems frame a programme covering agent permissions, code review and team adoption.
Anthropic urges review of prompts in Fable 5.2 guide
The guide addresses development problems including early stops during long tasks and the scope of code changes.
TTSE separates agent experience into facts and tips
TTSE proposes a way to classify agent experience, while reported gains remain the development team's own results.
Web update breaks nightly tests for an unchanged agent
Moved buttons and renamed fields raise compatibility concerns for evaluations that rely on reading and operating a screen.
IBM and Microsoft unveil enterprise AI agent platform
Imagine Enterprise is designed to connect with existing business systems, with deployment and operation through IBM Consulting alongside self-management options.
IBM and Microsoft unveil enterprise AI agent platform
Imagine Enterprise centres on adapting pre-built assets to company workflows, while its customer example focuses on technology exploration.
Qwen Code preview removes separate Node.js install
Packages for major operating systems support session resumption, while preview interfaces, settings and behaviour remain subject to change.
Unreal Engine 5.9 draws divided reactions over AI
For developers weighing AI agents in game production, the central issue is which tasks those agents would handle.
Kora Systems previews an agent-centred PC operating system
Kora Systems plans an operating system designed around AI agents, with a full release scheduled for next year.
Support AI launch puts action permissions under scrutiny
A favourable penetration test report needs to be read alongside the agent’s permission to change customer records and issue refunds.
UiPath launches Cartographer to map workplace decisions
AI Times described how structured maps of workplace knowledge support AI agent design and operation.
Alibaba adds AI agents to Qianwen Office team workspaces
Shared spaces linked to chats and documents and event-triggered tasks are the two strands of the Qianwen Office 2.0 update.
Google expands eighth-generation TPU range for enterprises
The report presents the new TPU systems as intended for enterprise agentic AI workloads.
Developer automates app language additions and deployment
The setup checks a language queue daily and runs a coding agent every three days, linking language additions to deployment, checks and email reports.
DeepSeek Harness reaches 1.54 million daily active users
Company-reported usage supports DeepSeek's entry into the market, but does not establish how its product performs against rivals.
Sendbird launches agent to unify customer service
Sendbird aims to bring customer service tasks together and let business staff build and improve agents.
Kora puts AI models and agents into Linux for Snapdragon PCs
For game developers, Kora raises the question of how AI fits into the environment that runs their development tools.
Claude Fable 5.2 divides development work among AI agents
The reported speed gain compares the new models with version 5.1, while Fable adds request analysis and task allocation for complex development work.
Guide details checks on WhatsApp AI agents at UAE businesses
A dev.to guide argues that the first clear, correct reply usually wins the next step, but its opening gives no percentage or count.
AWS details seven-layer reference architecture for AI agents
The framework lets teams swap the technology within each layer, but the announcement does not name a licence for its sample repository.
Cloudflare chief compares AI agent optimisation to a new SEO
OpenAI's payment for page-change signals and possible lower charges for agents that explain their visits were cited as examples.
Space and Time Enterprise lets AI agents verify private data
Space and Time Enterprise uses Proof of SQL to prove calculations without exposing records, but usage pricing is not in the public material.
5GB Gemini app-swap claim tells little about app connections
Google's Connected Apps documentation sets out the conditions for linking apps to Gemini, while no public document holds measurements behind the 12 apps and 5GB.
qwen-code v0.27.1 adds opt-in Mem0 memory for managed agents
The CLI also warns of Wasm memory spikes on Node 24 and 25, while the release notes give no commercial or redistribution terms
Ant's Ling-3.2-flash tops Claude Opus 5 in visual benchmarks
Ling-3.2-flash's lead over Claude Opus 5 is limited to some evaluations, while Ant Group is pitching low usage fees and a context of 1 million tokens.
AWS adds always-on Kiro Bots to Kiro Crew to organise agents
AWS points to a lighter load of agent supervision, but no measurement shows how far that load falls.
Meta open-sources Astryx, a design system made for AI agents
The system reportedly aims to reduce the inconsistencies that arise when AI composes screens arbitrarily, but the report omits the licence terms outside development teams need to assess.
2 weeks agoNov 03 – Nov 09
Aircheck gathers all posts agents queued today on one screen
The title of the Show HN post does not give a licence or say whether the tool runs locally.
Google ships Agent Substrate 1.0 on GKE and gives it to CNCF
Agent Substrate moves to the CNCF alongside gVisor and OpenRL, but public documents give no test conditions for its density figures and no list of breaking changes.
AWS adds Gateway policy layer to Okta agent delegation setup
Per-user permission decisions stay on the tool servers, and the AWS post gives no measurements at the scale of hundreds of agents.
Taxbit details tax AI design that keeps identity from models
The assistant for a regulated sector keeps permission scope outside the model, and Taxbit did not name the model that runs inference.
OpenAI Codex 0.167.0 gains backend switching in the terminal
Saved chats stay visible after a model provider change, and system proxy support arrives as an experimental setting.
Amazon Quick lets a separate model approve its agent actions
Smart Approvals is on by default, so AWS advises administrators to check first which actions stay under human approval.
U Mobile runs its AiDaH helpdesk on Amazon Bedrock AgentCore
AWS says U Mobile's helpdesk settles about 70% of requests without a model call, but the passage reporting a ticket cut of up to 84.5% gives no measurement period.
CWE-bench v2 opens 100 security tasks to AI model developers
Even the leading model failed 19% of the tasks on every attempt, and the announcement gives no licence or commercial terms.
3 weeks agoOct 26 – Oct 31
NVIDIA post ties NemoClaw agents to policy-enforcing runtime
NVIDIA presents a policy-enforcing runtime as a requirement for more secure agent deployment, but the post's introduction gives no version number or licence terms.
Ollama adds Liquid AI's LFM3 models, on Apple hardware first
The release notes do not name a licence or give a date for Linux and Windows support.
AWS shows a semantic layer for AI agents inside EDB Postgres
The same agent code returns different results depending on the database role that runs it.
Oct 2026Oct 12 – Oct 23
Codex 0.164.0 changes command input and authentication flows
The release notes list no breaking changes, but resumed sessions now restore different settings.
Microsoft posts ThinkingBox-Training agent RL tool on GitHub
Company figures show more tasks passed on every attempt and fewer solved at least once, and the licence name is not confirmed.
Apache-2.0 coding model Flint runs first on macOS via Ollama
The `Apache-2.0` model has 40B parameters, but the release notes give no minimum VRAM or training data sources.
Agents SDK adds Pi 1.4 support but deployments cannot revert
A session storage schema change blocks any return to Pi 1.3, and the announcement does not say how long the migration takes.
NVIDIA says AI agent controls must operate outside the agent
NVIDIA said monitoring and enforcement need to operate beyond an AI agent's reach because an agent that completes its task can still cross a security boundary.
Cursor makes its Agent SDK generally available to developers
Bots defined as code folders can be evaluated and deployed to Cursor or a user's own infrastructure, but the changelog omits the licence and deployment pricing.
Cursor cloud environments now support up to 300 repositories
An Environments API beta and link sharing for canvases arrived with the update, but no general availability date has been published.
Meta launches Muse Spark Open, its 816B agentic coding model
The model lists 42B active parameters and a 1M-token context window, but the introduction to NVIDIA's blog post gives no licence terms or local running requirements.
NVIDIA details NeMo Gym method for tuning chip design agents
NVIDIA says the risk of plausible but wrong outputs spreading through multistage design workflows is a major barrier to production use.
XBOX creates XBOX LABS, plans to share tools with developers
XBOX says the new unit's tools will go to developers across the industry, but its message names neither the tools nor the licence terms.
LiquidAI pitches LFM2.5-24B-A2B for private on-device agents
LiquidAI's post title puts fast execution and privacy first.
Devin Desktop 3.12.28 brings per-session MCP server controls
The release also fixes local-session and Remote SSH errors, but the changelog lists no vulnerability numbers for its security fixes.
104 in totalWritten by Jiyul SeoRSS feed


