Agentic AI News

Broadcom Unveils AI-Ready Data Foundations in VMware Tanzu Platform to Power Secure Enterprise AI Cloud

Broadcom's VMware Tanzu Platform gets new AI-ready data foundations at VMware Explore 2026 for enterprise agent clouds.

Broadcom announced new AI-ready data foundations for the VMware Tanzu Platform at VMware Explore 2026 in Las Vegas. The update is aimed at supporting secure enterprise AI cloud deployments, with the platform positioned as VMware's official agent platform. For those tracking AI agent infrastructure, this reinforces how cloud vendors are packaging data, security, and orchestration together. The main question is whether data foundations alone are enough to pull agent builders onto Tanzu.

Mureka Introduces Next-Generation AI Music Model V9.5, Advancing Toward More Human-Like Song Creation

Can Mureka's V9.5 make AI songs sound human? The August 31 release promises richer vocals and tighter arrangements.

Mureka said on Aug. 31, 2026, that it has launched V9.5, its latest AI music model. The company says the update brings richer emotional depth, more cohesive arrangements, lifelike vocals, and a stronger musical identity. That push comes as AI music tools race to move from rough demos to tracks creators can actually use in production. For those tracking AI music generation, this signals a push toward songs that feel less synthetic and more like finished performances, with vocals as the key test.

Tealium expands agentic capabilities to bring governed customer context to any AI platform

Tealium's new Configuration MCP gives AI platforms governed customer context and natural-language control today.

Tealium announced Configuration MCP, a natural-language interface in Tealium Studio, plus new platform APIs through its developer portal. The launch is aimed at letting AI tools work with customer data and platform settings without losing governance. For those tracking agentic AI, this signals a move toward controlled tool use inside enterprise data stacks. The key detail is that Tealium is opening both the interface and the APIs at once.

Security Risk Advisors Launches SCALR AI as a Free SOC AI Platform for Security Teams

Security Risk Advisors' free SCALR AI could cut SOC triage time, but one built-in limit is still missing.

Security Risk Advisors launched SCALR AI on August 24 as a free SOC AI platform through Azure Marketplace. The product ships with incident triage, enrichment, and AI-driven quality review built in. For those tracking agentic AI, this signals security teams are getting tools that can do real operational work, not just chat. The big question is what usage limit, if any, SRA left off the free tier.

Amazon paper reveals KV-cache policy influences inference and training of long-context models

Amazon AGI says reinforcement learning can lift long-context performance by 20 points while cutting inference costs.

Amazon AGI researchers report that KV-cache policy and reinforcement learning fine-tuning can improve long-context model performance by up to 20 points. They also say the approach makes inference more efficient, which matters as models handle longer tasks and larger inputs. For those tracking agentic AI, this signals a possible memory and planning boost for systems that need to hold onto context across long workflows. The most interesting part is that a cache policy choice may shape both training and runtime behavior.

rhino.ai Launches rhino Agentic Mainframe Modernization on AWS, Giving Enterprises a Complete Blueprint to Modernize Beyond the Mainframe

rhino.ai launched Agentic Mainframe Modernization on AWS to map legacy logic across connected systems.

rhino.ai launched Agentic Mainframe Modernization, a platform that maps business logic across the mainframe and connected systems. The company says the goal is to turn that estate into one governed model for teams modernizing on AWS. For those tracking agentic AI, this signals a move toward using agents in complex enterprise migration work, not just customer service or chat. The hook is that human teams and AI agents are supposed to build from the same trusted map.

madVR Labs Launches madAssist AI Product Expert - Residential Systems

madVR Labs launches madAssist, a patent-pending AI expert that can analyze and configure products 24/7.

madVR Labs unveiled madAssist as a branded AI Product Expert for complex technical products. The platform answers questions, guides users through support flows, and can directly analyze and configure supported devices when integrated. For those tracking agentic AI, this shows vendors pushing assistants from customer service into hands-on product control. The key detail is which supported products it can actually configure at launch.

GCSA Agent Achieves 91.3% on CyberGym, Ranking Among the World's Leading AI Cybersecurity Agents

GCSA Agent scored 91.3% on CyberGym, putting it among the top cybersecurity agents on a tough real-world benchmark.

The Global Cybersecurity Alliance says GCSA Agent reached a 91.3% success rate on the CyberGym benchmark. That places it in CyberGym's Leading Systems Above 90% category, based on a research framework from the University of California, Berkeley. For those tracking agentic AI, this is evidence that autonomous security agents are getting much better at real vulnerability work. The key detail is how the agent performed on vulnerability analysis and proof-of-concept generation.

Auckland scientist joins AI agency debate - Inside Government NZ

Could artificial systems develop their own goals? University of Auckland scientist joins a $7 million study hunting the answer.

A University of Auckland computer scientist, Dr Matthew Egbert, has joined a $7 million international research program on agency and intelligence. The team is studying how goals, meaning, and autonomy arise in living systems and whether machines could ever share those traits. For those tracking agentic AI, this gets at the core question behind reliable autonomy and self-directed action. The big tease is whether the project thinks artificial agents can ever move past task following into real goal formation.

Gnani.ai launches Gnani Artha, an end-to-end sovereign AI stack for Indian enterprises and public institutions

Gnani.ai's 30-billion-parameter Gnani Evon v3.3 powers a sovereign AI stack for India, but which 11 languages?

Gnani.ai has launched Gnani Artha, an end-to-end sovereign AI stack for Indian enterprises and public institutions. The stack combines Gnani Evon v3.3, a 30-billion-parameter open-weights model trained natively in 11 Indian languages, with the company's Gnani Plexus agentic AI platform. It arrives as governments and enterprises push for AI systems they can control, localize, and deploy inside sensitive workflows. For those tracking agentic AI products, this signals a move toward language-specific, state-backed systems built to take real actions in regulated settings.

Anthropic Opens a Research Preview of the Model Hardware Standard (MHS): A Shared Specification for AI Agents to Safely Operate Physical Devices

Anthropic’s Model Hardware Standard could cut device setup from weeks to hours, but one missing piece still matters.

Anthropic has opened a research preview of the Model Hardware Standard, a shared specification for AI agents to discover and operate physical devices. The standard targets a messy lab and factory problem where every vendor ships its own interface and teams build custom translators. For those tracking AI agent infrastructure, this signals a push to make real-world device control more interoperable and less manual. The open question is whether the standard can work across enough hardware to become a common layer instead of another silo.

Why Okta (OKTA) Is Up 23.0% After Raising 2027 Outlook On AI-Agent Security Demand - And What's Next

Okta's $3.22 billion 2027 forecast rides AI-agent security demand, but the real prize is still hidden.

Okta said second-quarter fiscal 2026 revenue hit $805 million and net income reached $116 million, while it lifted full-year fiscal 2027 revenue guidance to about $3.22 billion. The company said demand for its identity security platform is strengthening as more customers look for ways to secure autonomous AI agents. For those tracking the AI agent ecosystem, this signals that identity and access control are becoming a must-have layer for agent deployments. The bigger question is how much of Okta's new AI-focused bookings can scale as the category moves from early interest to actual production use.

SwarmWorld: Stigmergic technological evolution in societies of LLM agents

SwarmWorld explores stigmergic coordination in societies of LLM agents, focusing on how agents leave signals that guide later actions. The work sits in a fast-growing push to build multi-agent systems that can coordinate without tight central control. For those tracking agent platform ecosystems, this points to a different path for orchestration, memory, and emergent teamwork. The key question is whether simple shared traces can produce complex behavior without breaking down.

From India, for the World: VerSe Innovation Unveils Spark Station, Its First Global AI Content Platform

Can VerSe Innovation's Spark Station really run content from idea to final cut at Film Expo 2026?

VerSe Innovation unveiled Spark Station at Film Expo 2026 as its first global AI Content Operating System. The platform is built to carry a project from first idea through production and final cut inside one workflow. For those tracking AI writing and content tools, this signals a push beyond text generation into full creative operations. The big question is whether VerSe can turn that promise into a system creators actually trust for end-to-end production.

VP Radhakrishnan launches 'Gnani Artha' sovereign AI stack

Can GNANI AI's Gnani Artha really connect models to real-world work? VP Radhakrishnan launched it Friday.

Vice-President CP Radhakrishnan on Friday launched Gnani Artha, a sovereign AI stack from GNANI AI at Uprashtrapati Bhavan. The stack combines Evon 3.3, a large language model, with Plexus, a platform built to connect AI intelligence to institutions and work. For those tracking agentic AI, this signals another push to turn models into systems that can act inside real workflows, not just answer prompts. The launch points to India's ambition in sovereign AI, but the key question is how much real-world action Plexus will actually handle.

Gnani unveils sovereign AI stack 'Artha' with open-weight model, enterprise agents

Gnani

Agtonomy Highlights Expanding Data Capabilities and New Autonomous Maneuvers Across Its Physical AI Platform

Agtonomy adds autonomous multi-point turning to its platform, pushing field rigs through tighter jobs without a driver.

Agtonomy said it has expanded data collection across its autonomy stack and added autonomous multi-point turning to its units. The update builds on its push to bring factory-fit automation to off-road equipment through OEM partnerships. For those tracking agentic AI, this signals more systems that can sense, plan, and finish real tasks in messy physical environments. The standout detail is how the platform now logs passive data at scale while executing maneuvers that usually need a human operator.

Gnani AI launches Artha sovereign AI stack with 30-billion-parameter Evon 3.3

Gnani AI said Friday Artha lets Indian firms run AI inside their own systems, backed by 30-billion-parameter Evon 3.3.

Gnani AI launched Artha, a sovereign AI stack for Indian companies and public institutions. The stack includes the Evon 3.3 language model and the Plexus agentic platform, with models that can run inside a customer's own infrastructure. For those tracking agentic AI product launches, this is a clear push toward local, controlled deployments instead of cloud-only tools. The standout detail is that Artha pairs open-weights Indic-language tech with an agentic platform built for in-country use.

This Week In Electric Vehicles - Einride Unveils Flip AI To Streamline Electric Fleet Operations

Einride AB's Flip AI will handle charger problems and delay alerts for electric fleets, shippers, and operators.

Einride AB introduced Flip AI, an agent built to streamline operations for electric fleets, shippers, and charging infrastructure operators. The system was developed by Flipturn, which Einride recently acquired, and it automates tasks that usually need manual intervention. For those tracking AI agents in business operations, this is a concrete fleet workflow moving toward automation. The interesting part is how far it can go before a human still has to step in.

Tutti * VM Launches in Early Access: The Google Docs Moment for Cross-Agent Collaboration

Can Tutti VM make Claude Code and Codex work together in one cloud room while keeping identities local?

Tutti opened Early Access for Tutti VM on Aug. 27, calling it a real-time collaboration space for multiple agents and users. The product puts separate coding agents into a shared cloud room while keeping local agents and identity on the user's machine. For those tracking agentic AI, this points to a new layer of coordination between tools instead of a single assistant acting alone. The catch is how much of that collaboration stays private when multiple agents are working in parallel.

Somansa Launches AI Agent Security Tool That Blocks Forbidden File Access

Somansa's Privacy-i AIDR now blocks AI agents from forbidden files before they can read sensitive data.

Somansa launched Privacy-i AIDR on August 26, adding AI Detection and Response to its endpoint security platform. The module enforces OS-level permissions on shadow AI agents and controls what files, programs, and system resources they can touch. For those tracking agentic AI product launches, this shows security vendors are moving from detection to real-time control. The big question is whether this kind of enforcement becomes the standard line between safe agents and risky ones.

HighByte Releases Agentic Configuration for Industrial DataOps

HighByte's Intelligence Hub 4.5 adds agentic configuration and a new data lineage graph for industrial data teams.

HighByte released Intelligence Hub 4.5 with agentic configuration tools for contextualizing industrial data, plus a new Data Lineage Graph. The update also improves AVEVA PI System integration, expands i3X support, and strengthens enterprise deployment features. For those tracking agentic AI product launches, this shows vendors pushing autonomy into messy industrial workflows, not just chat interfaces. The most interesting piece is how the release ties agent behavior to traceable configuration changes in production data systems.

Research Solutions at 17th Annual Midwest IDEAS Conference: ai push lifts margins By Investing.com

Research Solutions is shifting from low-margin document delivery to higher-margin AI software, even as transactional revenue slows.

Research Solutions used its August 27 presentation at the Midwest IDEAS Conference to say its business is moving toward AI software. The company said the shift is lifting margins, but transactional revenue is under pressure and sales cycles are getting longer. For those tracking AI in business operations, this shows where enterprise workflow tools can improve profitability while adoption friction still bites. The key question is how much of the old business the company can replace before the slowdown catches up.

Microsoft 2.5: Superintelligence leader Ali Farhadi points company toward AI self-sufficiency

Microsoft built its AI business on OpenAI, but Ali Farhadi is pushing toward self-sufficient frontier models.

Microsoft is profiling the people behind its “Microsoft 2.5” era, with Ali Farhadi leading a push inside Microsoft AI. The company first gained its AI foothold through OpenAI, but now it wants to build its own frontier models. For those tracking the agent platform ecosystem, this signals a possible shift in who controls the core model layer. The key question is how far Microsoft will go before its dependence on OpenAI starts to fade.

The Trade Desk launches Kokai Zuma with agentic AI and new measurement tools

The Trade Desk's Kokai Zuma adds agentic AI to ad buying, aiming to automate campaign changes and cut routine work.

The Trade Desk launched Kokai Zuma, the latest version of its Kokai platform, with agentic AI capabilities, simpler measurement tools, and a redesigned interface. The company says the release helps advertisers focus on campaign outcomes and automate changes instead of spending time on routine workflows. For those tracking agentic AI, this signals that ad-tech vendors are turning autonomous features into core product upgrades. The biggest question is how much control advertisers will hand over when the system starts making campaign moves on its own.

YuVerse Launches Instant-Build Voice & Video AI Platforms for Enterprises

YuVerse rolled out four enterprise AI tools at Yunite, and one targets voice mistakes before customers hear them.

YuVerse announced four new products at its Yunite event in Mumbai on Aug. 27, 2026. The lineup includes tools for AI agent building, video generation, voice quality assurance, and customer engagement campaign management. Vendors are moving fast to package agentic AI as enterprise software, not just experiments. For those tracking agentic AI product launches, the standout is the push to make agents and voice tools easier to deploy at scale.

Skyflow launches data-control layer for Glean enterprise AI search

Skyflow is adding data controls for Glean as enterprise AI search spreads across SaaS apps, wikis, and data lakes.

Skyflow launched Skyflow for Glean, a data-control layer aimed at enterprise AI search and agent workflows. The product is designed to protect sensitive information as it moves through sources like CRM systems, SaaS apps, data lakes, and internal wikis. As agents touch more business data, security and governance are becoming part of the product stack, not an afterthought. For those tracking agentic AI product launches, this matters because data access often decides whether a deployment can move forward at all.

AvenuesAI launches agentic payments playform 'PayCentral'

AvenuesAI, formerly Infibeam Avenues, launched its PayCentral platform through subsidiary PhroneticAI. The system lets AI agents transact directly with one another without a person manually starting every payment. For those tracking agentic AI, this shows vendors pushing from chatbots toward tools that can execute real financial actions. The missing question is how PayCentral handles control, trust, and limits when agents start paying each other.

AccuKnox Launches AgentZ to Help Enterprises Build, Run, and Govern AI Agents at Scale

AccuKnox announced AgentZ on Aug. 27, 2026 as a platform for building, running, and governing AI agents. The company says it brings the agent, execution environment, tools, workflows, permissions, and governance into one stack. Enterprises are under pressure to move agents from experiments to production without stitching together too many parts. For those tracking agentic AI product launches, this is a clear bet on the tooling needed to run agents safely at scale.

An LLM-agent interface for end-to-end probabilistic seismic hazard and risk analysis - npj Natural Hazards

Can one LLM agent really run seismic risk analysis end to end, and trust the output enough for building codes?

Researchers in npj Natural Hazards presented an LLM-agent interface for end-to-end probabilistic seismic hazard and risk analysis. The work targets a field that shapes building codes, insurance pricing, and disaster planning, but has mostly stayed in expert hands. By automating the pipeline, the system tries to make advanced analysis easier to access. For those tracking agentic AI, the surprise is that a research-grade agent is moving into a domain where mistakes carry real-world cost.

Sequel Launches Agentic Engagement Cloud to Move the Webinar Category Beyond the Webinar

Sequel.io's Aug. 26 launch could turn webinar attendance into pipeline, but only if agents can use website behavior.

Sequel.io launched the Sequel Agentic Engagement Cloud at Human Moments: Agentic Momentum on August 26, 2026. The platform blends webinar video with first-party web behavior so AI agents can create content, personalize programs, and automate engagement workflows. For those tracking agentic AI product launches, this signals a push beyond chatbots into systems that act on audience data. The key question is which marketing tasks Sequel can now automate that competitors still leave to humans.

Business News | ROI Spectrum Launches AI-Powered Search Technology Suite. As It Expands Its Full-Funnel Digital Growth Platform

ROI Spectrum launched three AI search tools in New Delhi on August 26, chasing stronger digital authority.

ROI Spectrum said on August 26 that it is expanding its AI-powered digital growth platform with three new search capabilities. The company says the tools are built to help businesses strengthen authority and compete in changing search and discovery channels. For those tracking AI for sales and marketing, this points to more vendors packaging search visibility as a growth service. The catch is the announcement stays broad, with no clear details on how the tools actually change outbound or campaign execution.

AI Agents Use 5 Times More LLM Tokens Than Humans

Andreessen Horowitz says AI agents burn five times more tokens than humans, and cached inputs may be part of why.

Andreessen Horowitz used OpenRouter data to estimate that autonomous AI workloads consume about five times as many LLM tokens as human-driven usage. Much of that traffic comes from cached inputs, which can still carry costs. For those tracking AI agent platforms, the finding raises the price of scale and the strain on inference infrastructure. The interesting part is that the biggest cost driver may be hidden in the parts users do not notice.

QueryStory Launches a New Class of Agentic Data Platform to Close the Gap Between Company Data and Decisions

QueryStory says Fortune 500 firms are seeing strong results from its new platform, but its decision engine stays hidden.

QueryStory launched its agentic data platform on August 26, 2026, after emerging from stealth in San Francisco. The company says its founders come from Google, Chronicle, EvolutionIQ, and Accenture, and it is already seeing strong results with Fortune 500 clients in regulated sectors. For those tracking agentic AI product launches, this signals a push to move AI from analysis into actual business decisions. The key question is how QueryStory plans to bridge data and action for enterprises that cannot afford mistakes.

Workato Launches Otto for Everyone, AI Superagents That Run Whole Projects Start to Finish

Workato’s Otto now runs across 1,400 apps, with a free launch in Slack, web, and SMS.

Workato has launched Otto for Everyone, making its AI superagents available to individuals and teams through Slack, the web, and SMS. The service can take action across Gmail, Salesforce, GitLab, and more than 1,400 business applications. For those tracking agentic AI product launches, this is a strong signal that workflow agents are moving from enterprise back offices to everyday use. The catch is how much of a whole project Otto can finish before a human steps in.

Runable secures $21M to enhance AI agent growth capabilities

Runable says 60% to 70% of its trillion-token load came from paying customers in just 90 days.

Runable says it processed more than one trillion tokens in the last 90 days and that 60% to 70% of that usage came from paying customers. The platform lets users build websites, slides, reports, videos, and audio through natural-language prompts. The broader signal is that agent platforms are being judged less by demos and more by real customer usage. For those tracking AI agent ecosystems, this is a strong demand signal for a young product with paid traction.

Multilogin releases open-source agent so AI can run social accounts on real Android cloud phones

Multilogin's open-source agent now runs TikTok, Instagram, and Facebook on real Android cloud phones where browser bots fail.

Multilogin released an open-source agent that connects AI systems to real Android cloud phones. The tool lets agents view screens, tap, type, and manage social profiles on TikTok, Instagram, and Facebook. The broader push is toward agents that can operate on full mobile devices instead of being limited to browsers. For those tracking agent tooling, this shows a practical way to close a gap browser-based automation has not solved.

Algorand Unveils AC2 Protocol to Secure Next-Gen AI Agents - Crypto Economy

Algorand

Auxia Launches Agent Studio

Auxia's Agent Studio goes live today, giving marketers a control plane for AI agents, and one workflow still won't run itself.

Auxia launched Auxia Agent Studio today for marketers who want AI agents to plan, build, and launch campaigns. The product gives teams shared business context and moves manual execution into agent workflows. For those tracking agentic AI, this is a clear sign that marketing tools are shifting from chat to coordinated action. The sharpest detail is that it is meant to replace back-and-forth reviews and approval chasing.

Olas launches Pearl Connect for Claude coding agents

Claude coding agents now get self-custody wallets on Gnosis and Polygon, opening on-chain actions inside sessions.

Olas launched Pearl Connect for users of Claude Code and Claude Desktop in beta. The product is available on the Gnosis and Polygon networks and gives coding agents a self-custodial crypto wallet. For those tracking agentic AI, this signals agents are moving from code helpers to tools that can prepare real transactions. The notable twist is that users can research markets and queue on-chain trades without leaving the session.

Temporal Technologies: Temporal Releases 'The 2026 State of Development Report: AI Agents,' Revealing a 70.8% Leap in AI Agent Use Among Engineers

Temporal says 550-plus engineers boosted daily AI agent use 70.8%, and 9 in 10 saw productivity gains.

Temporal released The State of Development Report: AI Agents, based on a survey of more than 550 engineers and engineering leaders in the US and UK. The report says daily AI agent use rose 33 points in a year, with most respondents calling agents helpful or transformative. For those tracking agent infrastructure, this is a sign that usage is moving from experimentation to routine development work. The most interesting detail is how fast daily use is spreading inside real engineering teams.

Crescendo Launches the AI-Native Customer Experience Platform: One System That Runs and Continuously Improves the Entire CX Operation

Crescendo says its new CXP can replace six CX systems and go live in 30 days with AI agents running them all.

Crescendo launched its Customer Experience Platform, or CXP, as an AI-native system for contact centers, ticketing, workforce management, quality, and knowledge. The company says specialized AI agents run the stack end to end and keep improving it over time. For those tracking agentic AI, this shows how vendors are bundling automation into full operating platforms. The big question is whether one system can really absorb so many categories at once.

ClassNK Launches Survey Compass AI Assistant For Shipowners

Can ClassNK's Survey Compass cut ship survey prep time for shipowners, or just answer questions faster?

ClassNK has launched Survey Compass, an AI assistant for shipowners and ship management companies handling survey-related work. The tool answers natural-language questions and shows supporting sources to help users check classification survey matters faster. For those tracking agentic AI product launches, this shows vendors are pushing into narrow work tasks where accuracy and traceability matter. The key question is whether this becomes a real workflow helper or stays a smarter FAQ bot.

Prompt injection searches jump 141 percent as LLM attack research leaves the lab - Technology Org

Why did prompt injection searches jump 141 percent as LLM attack research moved from papers into public view?

Searches for prompt injection rose 141 percent year over year to about 43,400, according to Conflict International. The topic, once mostly limited to academic and security circles, is now showing up in mainstream search patterns. For those tracking agentic AI, this matters because prompt injection is a direct threat to tool-using systems. The telling detail is that attention is rising almost as fast as the risk.

RoboColiseum Launches Standardized Simulation Platform for Embodied AI Evaluation

Can RoboColiseum show where embodied AI still fails, after its August 24 launch in Shanghai?

RoboColiseum officially launched in Shanghai, China, on August 24, 2026 as a standardized simulation platform for embodied AI. The system is meant to close a gap in evaluation standards as robot and embodied intelligence models improve fast. For those tracking agentic AI, this signals a new way to test whether systems can actually complete physical tasks, not just score well in demos. The key question is which real-world robot failures this benchmark will expose first.

EXCLUSIVE: As Home Sales Slump, RealReports Launches AI Tool to Help Agents Win More Business

Six months after closing, why is RealReports' Aiden turning one sale into 87 new prospects?

RealReports is launching Aiden, an AI tool that turns a past home sale into a new lead-generation campaign. In a demo for Realtor.com, the system identified 87 prospects nearby, drafted direct-mail copy, and priced the campaign under $100. For those tracking agentic AI product launches, this is a clear example of software taking several sales steps on its own while still leaving the final approval to a human. The most interesting part is where Aiden stopped, because that boundary may decide how fast agents get adopted in real estate.

Caddi Launches an AI Agent That Builds and Governs a Firm's Back-Office Agents

Caddi's new agent builds and governs other agents for wealth managers, law firms, and insurers, with full audit trails.

Caddi launched an AI agent in Seattle today that discovers repetitive work, learns each process like a new hire, and then builds agents to handle it. The company is aiming first at wealth management firms, law firms, and insurance companies with large back offices. For those tracking agentic AI, this pushes beyond task automation into agent creation and governance. The most interesting detail is that every agent it builds is supposed to stay under full audit control.

human.tech Launches Wallet Protocol on Ika Network for Decentralized Apps and AI agents

human.tech launched its Wallet Protocol on Ika Network today, letting AI agents move across EVM, Sui, and Solana.

human.tech says its Wallet Protocol is live today on the Ika decentralized network. The launch lets developers build multi-chain apps across EVM, Sui, and Solana, with more chains coming later. For those tracking agentic AI, this signals wallet access becoming a core layer for autonomous apps that need to act across networks. The unresolved question is how far its chain-agnostic design goes beyond the three chains live at launch.

LexisNexis Unveils 'Legal Intelligence Engine,' Rebuilding Protégé Around Dynamic Agentic Orchestration

LexisNexis Legal & Professional says Protégé can turn ideas into review-ready drafts, but the new engine still hides how.

LexisNexis Legal & Professional today unveiled the next generation of Lexis+ with Protégé, its generative AI legal assistant. The new release centers on a dynamic orchestration layer that coordinates models, agents, skills, and tools. For those tracking agentic AI product launches, this signals a push from chat-style help toward systems that can draft work ready for human review. The unanswered question is how much legal work the new engine can actually complete before a lawyer steps in.

Boosted.ai Introduces a Multi-Model AI Investment Committee for Select Institutional Partners

Boosted.ai's Alfa Prime combines signal scans and model debate for select partners, but one piece is still private.

Boosted.ai announced Alfa Prime for select institutional partners in New York and Toronto. The system combines large-scale signal detection, research agents, and structured debate across independent models to pressure-test investment ideas. For those tracking agentic AI, this is a clear example of AI taking on multi-step judgment inside a real workflow. The catch is that Boosted.ai has not fully disclosed which models or partner firms are already using it.

Galaxea AI Demonstrated Full-Stack Capabilities and Embodied AI Generalization at its WRC 2026 Productivity Awakening Exhibition

Galaxea AI showed over ten demos at WRC 2026, but its dark-store robot loop is the real first.

Galaxea AI closed its Productivity Awakening exhibition at the 2026 World Robot Conference in Beijing on Aug. 24. The company said it used more than ten multi-scenario demos to show embodied AI moving from research into industrial use. For those tracking agentic AI, this signals a shift from robots that move well to systems that complete real workflows. The biggest draw was a closed-loop dark-store fulfillment demo that the company says no one has shown before.

Microsoft introduces ThinkingBox to assess AI agent reliability

Microsoft's ThinkingBox exposes agents that pass once but fail repeatedly, a problem for business use by August 19, 2026.

Microsoft has released ThinkingBox, an open-source sandbox for testing whether AI agents can do real work reliably. The tool, described by Liang-Chun Tsai in a Microsoft Command Line blog post on Aug. 19, 2026, is meant to catch agents that look smart in one-off tests but break in repeated use. For those tracking AI agent tooling, this matters because reliability is becoming a buying criterion, not a nice-to-have. The hard part is that the framework tests consistency, not just whether an agent can finish once.

Microsoft introduces Agent Lightning v1.0, a framework for training AI agents without breaking their production setup

Microsoft's Agent Lightning v1.0 lets agents train in production setups unchanged, but one key tradeoff remains.

Microsoft released Agent Lightning v1.0 on August 17 as an open-source reinforcement learning framework for AI agents. The tool is built to let developers train agents without changing their existing code, tools, or environments. For those tracking agentic AI product launches, this signals a push to make training less disruptive and easier to ship. The catch is how much control teams really keep when learning happens inside the same setup they use in production.

Nuggets Launches Authority Control Plane for Autonomous AI Agents

Nuggets' Authority Control Plane now sits between agents and enterprise systems, but the real test is which actions it can block.

Nuggets has launched the Authority Control Plane, or ACP, a governance layer for autonomous AI agents in enterprise systems. The company says the product is available now as more firms move agents from demos into real workflows. For those tracking agentic AI, this signals a push to control what agents are allowed to do, not just who they are. Nuggets says identity tools can log agent actions, but ACP is meant to sit in the execution path and block risky ones.

Google introduces EnvHarness to enhance agent training environments

Google Cloud AI Research has introduced EnvHarness, a programmable wrapper for training environments that can adapt to agent weaknesses in real time. The tool was developed with researchers at Washington University in St. Louis and UNC Chapel Hill. For those tracking agent tooling, this matters because it gives developers a way to probe weak spots without rewriting benchmark code. The most interesting detail is that the environment can shift mid-test, changing what the agent must learn next.

HelloGov Launches GovSchema, an Open Standard for How AI Agents Interact with Government Services

HelloGov's GovSchema launches with 643 schemas across 124 jurisdictions, but the hardest government workflows are still ahead.

HelloGov AI, Inc. launched GovSchema, an open-source standard for how AI agents read and complete government processes. The registry starts with 643 published schemas across 124 jurisdictions, aiming to give agents a common way to handle public-service tasks. For those tracking agentic AI, this signals a push beyond chat into standardized action across messy real-world systems. The big question is which agencies and workflows will adopt it first.

Generalist AI GEN-1.5 Learns New Robot Tasks From Single Demo, No Retraining

Generalist AI's GEN-1.5 learned a new robot task from one 10-second demo after eight months of pretraining.

Generalist AI published GEN-1.5 on August 19, 2026, and showed a robotic arm learning to unzip a pencil pouch from one short human demonstration. The company says eight months of continuous pretraining produced in-context learning it did not explicitly design. For those tracking agentic AI, this signals robots may start adapting to new tasks without retraining or task-specific setup. The surprising part is that the model appears to transfer learning from a single demo into real action on its own.

India's ShepHertz Launches Sovereign AI to Help Enterprises Own Their Intelligence - The Tribune

ShepHertz launches AgentAnywhere in Gurugram, letting banks and hospitals keep AI agents on their own infrastructure.

ShepHertz Technologies launched AgentAnywhere on Aug. 21, 2026, in Gurugram as a sovereign agentic AI platform. The system is built for banks, insurers, hospitals, governments, and BPOs that want agents to run inside their own infrastructure and jurisdiction. For those tracking agentic AI product launches, this signals growing demand for controlled deployment rather than shared cloud access. The key detail is that governance is enforced on every model call, which could matter for regulated buyers.

Tekcapital firms sign robotics deal and unveil AI eyewear features

Tekcapital's Guident will send WatchBot robots to 40 Florida, Georgia and South Carolina sites, but the real test is autonomy.

Tekcapital said portfolio company Guident signed a deal with Coastal Waste & Recycling to provide WatchBot robots, AI software and support. The robots will patrol and inspect more than 40 sites across Florida, Georgia and South Carolina. That puts autonomous systems into a messy, real-world setting where reliability matters more than demos. For those tracking agentic AI, this signals a shift from software agents to physical ones. The unanswered piece is how much decision-making those robots will handle on their own.

China's Baidu Launches Free Task Engine 2.0 for Ernie Bot With End-to-End Autonomous Execution Capabilities

Baidu's free Task Engine 2.0 now promises end-to-end task execution for Ernie Bot, not just conversation.

Baidu unveiled Task Engine 2.0 on August 21 as a free upgrade for Ernie Bot with autonomous task execution. The system covers information collection, data processing, content creation and final output in one closed loop. For those tracking agentic AI product launches, this is a clear move from chat to action inside a mainstream assistant. The biggest detail is that Baidu is giving the upgrade away for free.