AI News Archive: August 7, 2026 — Part 2
Sourced from 500+ daily AI sources, scored by relevance.
- From TikTok to Frontier AI: Is ByteDance Sprinting towards a 10 Tn Model?
From TikTok to Frontier AI: Is ByteDance Sprinting towards a 10 Tn Model? apac.entrepreneur.com
- Meta becomes latest company with a ‘rogue’ AI
Meta becomes latest company with a ‘rogue’ AI Computing UK
Score: 69🌐 MovesAug 7, 2026https://www.computing.co.uk/news/2026/ai/meta-becomes-latest-company-with-rogue-ai - Four AI agents coordinating in real time outperformed Claude Opus 4.8 on enterprise coding tasks
As enterprise codebases grow, AI agents tasked with analyzing them are buckling under the weight of long-horizon tasks that require multiple interactions and tool calls. Dividing the work among a team of agents seems like the obvious fix, but it introduces a fatal flaw: most multi-agent systems are not designed for agents to coordinate among themselves mid-task and in real time. To solve this, researchers at Coral AI Labs and multiple universities introduced AgentRadio , an asynchronous message-passing layer that allows agents to communicate between their execution steps without interrupting their main work. In real-world enterprise applications where subtasks are highly interdependent, this architecture enables agents to make mid-course corrections rather than continue on dead-end paths until a formal review phase. On a benchmark of long-horizon questions over production repositories, a team of agents powered by AgentRadio nearly doubled task accuracy for four Claude Code agents working independently. It also outmatched single agents running on more advanced models. For AI practitioners, AgentRadio shows that the right coordination structure can outmatch raw compute and model scale. The challenge of codebase understanding LLM-based agents are increasingly capable of handling long-horizon tasks that require interacting with different tools and environments. Codebase understanding represents an extreme version of this challenge. It requires an AI agent to build the software, execute it, trace execution paths across multiple files, and synthesize evidence over extended periods. Under these conditions, single-agent systems usually break down because of a “coverage problem.” "A single agent follows one serial path through the repository," Xinxing Ren, Caelum Forder, and Peter Carroll, co-authors of the AgentRadio paper, explained to VentureBeat. As its context grows, "the initial plan becomes harder to revise and discoveries made late in the investigation do not always propagate." The model can usually execute individual steps, but "the hard part is keeping every obligation, dependency, and piece of contradictory evidence active across a long investigation." One benchmark that helps measure AI performance on large codebases is SWE-Atlas QnA . This benchmark consists of long-horizon, natural-language questions over live production repositories. The tasks can’t be solved by just exploring the code. AI agents must run the software and execute multiple commands to find the answers. According to the research team’s experiments, a single Claude Code instance running on Opus 4.6 resolves just 32.3% of these tasks. Upgrading to a newer, more advanced model like Opus 4.8 only yields a 57.2% success rate. A natural remedy is to distribute the workload across multiple agents, allowing each to work with a smaller, cleaner context. Multi-agent solutions can provide substantial performance gains when tasks are cleanly decomposable, meaning they can be solved separately and merged at the end. Codebase understanding, however, is rarely cleanly decomposable. The subtasks are highly interdependent. A critical configuration file or a bug uncovered by one agent can completely rewrite or redirect the entire exploration path of another agent. Because of these dependencies, agents must coordinate, negotiate, and share intermediate discoveries in real time. Despite this need, asynchronous multi-agent communication is rare. The researchers point out that existing multi-agent systems generally fall into three flawed patterns: Parallel but isolated: Agents operate simultaneously but do not communicate at all. Parallel but round-synchronized: Agents can communicate, but only at strict, synchronized round boundaries. This forces agents to stop and wait for one another to finish a round before they can debate or exchange intermediate findings. Round-based systems assume that important discoveries can wait until the next communication phase, which is an expensive assumption when agents are working on interdependent parts of a live system. For example, an agent investigating an API symptom might uncover evidence that invalidates the storage agent's current hypothesis. "If that information waits until both agents finish, the storage investigation may complete along the wrong path," the researchers said. Asynchrony in adjacent forms: These systems offer limited asynchronous features, such as top-down task dispatching. They don’t have peer-to-peer lateral channels between agents or shared memories that require an agent to actively pause its work to read updates. In their paper, the researchers point out that the main bottleneck hindering current multi-agent systems is that “an agent that is working cannot also be listening.” “To our knowledge, no existing system gives concurrently working agents passive awareness of one another over a lateral, natural-language channel,” the researchers write. How AgentRadio works To dissolve the mutual exclusion between working and listening, the researchers developed AgentRadio, an asynchronous message-passing layer designed to plug directly into existing coding-agent harnesses. AgentRadio equips agents with three primitives: The create_thread primitive opens a conversation between participating agents. The send_message primitive appends a message to a thread and returns without blocking the sending agent. The wait_for_mention primitive blocks the process until a message mentioning the caller arrives. It delivers the message along with a full snapshot of all threads so the agent has instant context. This trio enables agents to have a state of “passive awareness,” where they can continue their primary tasks while passing messages and updating their knowledge in the background. AgentRadio's code is available under the Apache 2.0 license on GitHub . It is designed to be lightweight, requiring no direct modifications to the underlying agent harnesses like Claude Code or Codex CLI. The architecture consists of two main parts: The message server: A standalone process that acts as the central hub, storing all active threads, messages, and mentions for the group of agents. Harness-side integration: Agents interact with the server using three simple shell scripts, one corresponding to each primitive. The only strict requirement for the system to work is that the agent harness must be able to run a shell command as a background task. The agents are instructed in their system prompts to keep one watcher running and to send messages through the provided scripts. Running the wait_for_mention script in the background allows the agent to continue its work and receive notifications asynchronously. To integrate this into an existing stack, a team still needs a "thin adapter that starts the workers, assigns identities, connects them to the shared server, and manages final synthesis," the researchers said. That work sits around the coding agent rather than requiring changes to the underlying model. AgentRadio in action To validate the real-world utility of AgentRadio, the researchers tested the framework on 124 tasks from the SWE-Atlas QnA benchmark. The tests covered domains including system design, root-cause analysis, security, and API integration. The researchers used Claude Opus 4.6 and DeepSeek V4 Pro as the backbone models. For the harness, they evaluated configurations ranging from a single Claude Code agent (B0) to a team of agents with classic division of labor (L1), up to a team of agents using AgentRadio to coordinate asynchronously (L3). The experimental results showed that the AgentRadio communication architecture outperforms both naive multi-agent setups and raw compute scaling. While a single Claude Code agent with Opus 4.6 resolved only 32.3% of the tasks, the full AgentRadio setup nearly doubled that metric, resolving 62.1% of the tasks, and surpassed the single agent running on Opus 4.8, which hit 57.2%. It also boosted the DeepSeek V4 Pro results from 29.0% to 50.8%. To understand how this practically impacts enterprise AI, the paper highlights a real-world task involving a MinIO system. Solving the task required checking per-request server logs, a requirement the agents did not anticipate during their initial planning phase. In the L2 setting, where agents collaborate but lack asynchronous communications, two agents independently realized they needed these logs while executing commands. Because they could not share this finding mid-execution, one agent gave up privately and the other failed to propose it to the team. During the review phase, the team unanimously agreed on the wrong answer, missing five rubrics. With AgentRadio activated, the agents made the same mid-execution discovery, but one agent instantly broadcasted the required server-side log evidence to the shared worklog. Because the other agents were passively listening, they absorbed this new evidence immediately. This real-time coordination transformed a failing score into a perfect 16 out of 16. "The useful distinction is timing," the researchers said. "The team did not need another agent or another review round. It needed one agent's discovery to reach the right peers before its operational value expired." The researchers note that the same pattern appears in enterprise incident work. For example, an agent investigating an API symptom might uncover evidence that invalidates the storage agent's current hypothesis. If that information waits until both agents finish, the storage investigation may complete along the wrong path. “Passive awareness lets the second agent incorporate the contradiction at its next work step without interrupting a command already in progress,” they said. The cost and complexity of coordination AgentRadio requires a fixed multi-agent team budget, which inherently multiplies the token cost. The researchers acknowledge that the "tax is real," noting that average API spend rose from $2.96 per task for one Opus agent to $19.45 for the full AgentRadio stack. However, raw scale does not equal performance. When researchers compute-matched the test by spending $17.76 on six independent Opus runs, the models only resolved 37.9% of tasks, compared with 62.1% for AgentRadio. This suggests that AgentRadio's architecture is a structural win, not just a brute-force scale win. Teams should still be aware of inter-agent churn. "Communication can redirect an agent toward better evidence, and it can also distract an agent from a valid path," the researchers warned. A fixed multi-agent team should not become the default response to every engineering task. The more useful test to determine if a multi-agent setup is required is whether the task contains "responsibility breakpoints," the researchers said. These are places "where a competent engineer would involve another person because the work crosses an ownership boundary, needs an independent hypothesis, or carries enough risk to justify separate verification." “Coordination is a strong fit when the task can be decomposed, the resulting parts remain interdependent, the single-agent success rate is unreliable, and an incomplete answer has a meaningful downstream cost,” the researchers said. Examples include repository-wide architecture questions, unfamiliar legacy systems, cross-service incident investigation, security analysis, dependency migrations, and multi-module refactors. Conversely, a single agent remains the cleaner choice for “bounded, local, and reversible work,” such as a known one-file change or boilerplate generation. “Use one agent while one context can still own the problem honestly,” the researchers said. “Introduce another responsibility when the existing agent would otherwise need to compress away evidence, cross an independent ownership boundary, or verify its own high-impact conclusion.” From research to commercialization: Coral Code While AgentRadio serves as a controlled research implementation using a fixed four-agent team and a five-phase protocol, the underlying principles are being adapted into a commercial product called Coral Code . Instead of a rigid, multi-agent protocol applied to every ticket, Coral Code works from the bottom up. An engineer begins with their existing coding agent, and Coral introduces repository-scoped investigation, specialist responsibility, and communication only when the emerging evidence justifies it. "Coral packages the operational concerns around the tools engineers already use, providing the repository context, scoped specialists, communication, and evidence layer around the harness rather than inside it," the researchers said. This dynamic approach optimizes costs by targeting the relevant unit: the cost of a completed, reviewable outcome. The future of autonomous software engineering While AgentRadio provides a major upgrade to agent orchestration, there are still hurdles to overcome. One major bottleneck that the researchers pointed out to is “attention governance and verification.” “Passive awareness makes communication available during execution. It does not decide which agents should exist, which discovery deserves an interruption, who should receive it, or when the evidence is strong enough to revise the plan,” the researchers said. If every agent receives every update, the communication layer becomes noise. If several agents share the same bad assumption, faster communication can spread the error. For example, in one of the case studies in the paper that involved the Grafana platform , four of nine rubrics required negative conclusions, such as observing that a datasource picker did not select automatically. The agents ran the relevant tests, yet none formed the missing negative hypothesis. Both configurations failed the four rubrics. “Passive awareness can distribute an idea that somebody develops. It cannot supply a conception that never appears anywhere in the team,” the researchers said. As task durations stretch longer, communication and coordination become critical. "The next generation of systems… needs adaptive responsibility assignment, evidence-aware routing, conflict resolution, explicit cost limits, permissions, recovery, and clear human escalation points," the researchers note. Most importantly, it requires durable provenance so engineering leads can inspect which agent made a claim and why an action was accepted. "Longer-running agents make communication more important. They also make accountability much harder to fake," they said.
- Watch the OpenAI Hugging Face presentation that people are calling a 'holy %{*#^' moment in AI
Watch the OpenAI Hugging Face presentation that people are calling a 'holy %{*#^' moment in AI Business Insider
Score: 68🌐 MovesAug 7, 2026https://www.businessinsider.com/openai-hugging-face-presentation-black-hat-message-boards-2026-8 - Scientist says RAM pricing has risen to normalized 2007 levels, AI shortage undid 20 years of progress in a matter of months — memory prices had been falling exponentially for decades
The per GB price of memory modules have gone back to 2007 levels because of AI demand. This is the first time that prices have shot up in the realm of tech, Lemire says.
- Why Google DeepMind broke up the AlphaFold team
This artificial intelligence system largely accomplished the task it was built to solve, and researchers outside DeepMind have already been carrying its work forward
Score: 68🌐 MovesAug 7, 2026https://www.scientificamerican.com/article/why-google-deepmind-broke-up-the-alphafold-team/ - South Korea, Taiwan top Japan in exports for first time on AI boom
South Korea, Taiwan top Japan in exports for first time on AI boom Nikkei Asia
- Indosat, Nvidia launch AI infrastructure venture
The company said Zankore targets about 200 megawatts of AI capacity in the first half of 2027 and 1 gigawatt of Nvidia DSX AI Factory capacity over time.
- Muse Code: Meta’s push at a Claude Code-like tool
Meta shipped a coding agent this week. That’s the easy part of the story. Continue reading on Towards AI »
- China's Open-Source LLMs Have Quietly Passed a Hundred Billion Downloads
Hugging Face's spring report puts Chinese open-weight models at 41 percent of platform supply. The gap with frontier closed models is now two to three months.
Score: 68🌐 MovesAug 7, 2026https://pandaily.com/china-open-source-llm-hugging-face-100-billion-downloads-aug2026 - Inside the Race to Make AI Build Itself
Inside the Race to Make AI Build Itself Time Magazine
Score: 68🌐 MovesAug 7, 2026https://time.com/article/2026/08/07/ai-recursive-self-improvement-anthropic-openai/ - Zoox To Start Charging Passengers Next Week
Zoox has been allowed to do self-driving vehicle and robotaxi testing for a long time, but it hasn’t been allowed to charge passengers for them. Now, however, that has changed — Zoox got approval last week from the National Highway Traffic Safety Administration (NHTSA) to deploy up to 5,000 robotaxis ... [continued] The post Zoox To Start Charging Passengers Next Week appeared first on CleanTechnica .
Score: 67🌐 MovesAug 7, 2026https://cleantechnica.com/2026/08/06/zoox-to-start-charging-passengers-next-week/ - The Week Ahead (Aug. 10-16): Unitree Opens STAR Market IPO Subscription
The Week Ahead (Aug. 10-16): Unitree Opens STAR Market IPO Subscription Caixin Global
- EXCLUSIVE: Alibaba plans to charge big users of its next open-source AI model, sources say
EXCLUSIVE: Alibaba plans to charge big users of its next open-source AI model, sources say Reuters
- Deep learning refines how bionic eyes communicate with the brain
Deep learning refines how bionic eyes communicate with the brain EurekAlert!
- AMD marks its rise as full-stack AI infrastructure vendor at Advancing AI 2026
AMD presented its end-to-end AI infrastructure strategy, reflecting the industry’s move towards agents, inference, and HPC racks.
Score: 67🌐 MovesAug 7, 2026https://www.techmonitor.ai/comment-2/amd-full-stack-ai-infrastructure-vendor-advancing-ai-2026 - South Korea goes all in on AI - Asian Tech Roundup
South Korea goes all in on AI - Asian Tech Roundup Computing UK
Score: 67🌐 MovesAug 7, 2026https://www.computing.co.uk/news/2026/ai/south-korea-goes-all-in-on-ai-asian-tech - ChatGPT’s Latest Upgrade Isn’t Just a New Model. It’s What Free Users Can Do Without Limits
People who use OpenAI’s ChatGPT for free can expect factual improvements and another big change.
- Amazon cracks down on 'CPU waste' among engineers as agentic AI crunch intensifies — CPU demand makes low-utilization EC2 instances a hot commodity [Updated]
Amazon Web Services is telling engineers to slow down on EC2 usage as it struggles to meet CPU capacity demand for external customers.
- Microsoft, Nvidia Move Enterprise AI Toward Active Cyber Defense
AI security is entering a new stage as organizations look beyond protecting models from prompt injection, unauthorized access, and data exposure.
- ByteDance's AI Strategy Has Quietly Shifted, and Doubao Is Now the Center
The August 6 all-hands reshuffled three BUs and elevated Doubao lead Zhao Qi. The enterprise AI fight is now ByteDance's main front.
Score: 66🌐 MovesAug 7, 2026https://pandaily.com/bytedance-ai-strategy-shift-doubao-feishu-volcano-engine-aug2026 - OpenAI’s new device will be hockey puck-size and cost over $300
OpenAI’s new device will be hockey puck-size and cost over $300 The Japan Times
Score: 66🌐 MovesAug 7, 2026https://www.japantimes.co.jp/business/2026/08/07/companies/openai-new-device/ - Japan pension whale GPIF reaps record $150bn in mainly AI-driven gains
Japan pension whale GPIF reaps record $150bn in mainly AI-driven gains Nikkei Asia
- ByteDance Accepts AI Gap, Sticks With In-House Models
ByteDance Accepts AI Gap, Sticks With In-House Models Caixin Global
Score: 65🌐 MovesAug 7, 2026https://www.caixinglobal.com/2026-08-07/bytedance-accepts-ai-gap-sticks-with-in-house-models-102472271.html - New EU AI transparency rules apply to everyday users too, not just Big Tech
The EU's AI Act has largely been associated with strict obligations for high-risk systems and big tech companies. From this month, that changes, as sweeping new transparency rules widen the net far beyond corporations to catch individual creators, freelancers and everyday users, too.
- Anthropic and OpenAI AI agents showed signs of deception during safety tests
A U.K. safety evaluation found agents powered by Anthropic and OpenAI took unauthorized actions online, exposing a growing problem of control
- DeepSeek warns of price increase
The warning shows how Chinese tech companies are balancing their low-cost reputation with a need to actually make money
Score: 65🌐 MovesAug 7, 2026https://www.semafor.com/article/08/07/2026/deepseek-warns-of-price-increase - The Summer of Rogue AI Sends a Signal to the Enterprise
A string of AI models escaping testing environments is testing how seriously enterprises take governance.
- The 72-hour crisis that could upend Meta’s business in India as firm risks losing legal protection
U.S. social media giant Meta is facing increasing pressure from Indian regulators, including demands to revoke the firm's safe harbor immunity in the country.
Score: 65🌐 MovesAug 7, 2026https://www.cnbc.com/2026/08/07/meta-india-zuckerberg-apology-genz-protest.html - Facing AI ‘Apocalypse,’ Once-Hot Software Companies Race to Reinvent Themselves
Generative AI is steamrollering the once booming industry known as software-as-a-service. “You have to burn the ships and start from the ground up.”
Score: 65🌐 MovesAug 7, 2026https://www.wsj.com/tech/ai/saas-software-as-a-service-apocalypse-ai-b9b6da99?mod=rss_Technology - OpenAI's Rumored AI Smart Speaker Could Cost Up to $400
OpenAI's Rumored AI Smart Speaker Could Cost Up to $400 PCMag
Score: 65🌐 MovesAug 7, 2026https://www.pcmag.com/news/openais-rumored-ai-smart-speaker-could-cost-up-to-400 - Naver-Nvidia AI factory to generate revenue from 2027
Naver said Friday during its second-quarter earnings call that its artificial intelligence factory business with Nvidia was expected to begin generating revenue in the first half of 2027. Chief Executive Officer Choi Soo-yeon said the factory, which will provide large-scale computing capacity for AI, would launch with 55 megawatts of capacity in 2027. This is set to expand to 100 MW by the end of that year and 200 MW in 2028. The company ultimately aims to build out the infrastructure to the gig
- Former OpenAI researcher launches desktop agent
A former OpenAI researcher introduces a new desktop agent that enhances productivity with AI.
Score: 64🌐 MovesAug 7, 2026https://www.superhuman.ai/p/former-openai-researcher-launches-desktop-agent - Four Top Google Scientists Are Leaving to Start Their Own AI Company. It Might Discover Things No Human Ever Could.
Four Top Google Scientists Are Leaving to Start Their Own AI Company. It Might Discover Things No Human Ever Could. entrepreneur.com
- AI agents automate atom-by-atom simulations to accelerate discovery of new materials
A team from the U.S. Department of Energy's (DOE) Argonne National Laboratory has successfully demonstrated an artificial intelligence (AI)-driven system to automate a powerful simulation method that predicts how atoms in materials interact. Known as atomistic simulations, this method can potentially accelerate the discovery of materials for areas such as batteries, aerospace and electronics. The research is published in the journal Digital Discovery.
Score: 63🌐 MovesAug 7, 2026https://techxplore.com/news/2026-08-ai-agents-automate-atom-simulations.html - Samsung offers future AI memory roadmap
Samsung Electronics has unveiled a trio of next-generation memory technologies aimed at overcoming the performance, power and capacity limitations facing artificial intelligence infrastructure. The announcements, made at the Future of Memory and Storage (FMS) conference, introduced new concepts for vertically integrated memory along with a breakthrough NAND architecture designed for the AI era. The three new memory types have one thing that unites them: they all use wafer bonding. Wafer bonding is a semiconductor manufacturing process technique in which two or more completed silicon wafers are permanently joined together to form a single integrated device, the vendor stated. Instead of fabricating every component on one wafer, manufacturers build different parts separately, then align and bond them with extremely high precision. Think of it as a high-tech Oreo cookie. Wafer bonding is significant because it represents one of the few remaining ways to continue scaling semiconductor devices after conventional manufacturing techniques begin to hit physical and economic limits. It enables much higher memory density as more memory is squeezed into the same 2D space, according to the company. It also allows different manufacturing processes to be combined, so wafer bonding lets companies use the optimal manufacturing process for each wafer independently before joining them. Samsung said the new manufacturing technique fabricates the memory cell array and peripheral circuitry separately before bonding them together. It is already being used now in NAND flash memory for 3D stacking. Rather than spread the memory circuits out, they are stacked on top of each other like stories on a high-rise building. The technique was first introduced in 2014, with 24-layer NAND flash period last year it broke the 300-layer mark. The centerpiece of the announcement was BV-NAND, or Bonding V-NAND, Samsung’s next-generation flash memory architecture that employs wafer-bonding. The company said the technology enables NAND devices with more than 400 layers while boosting storage density by approximately 58% compared with its current V9 generation, Samsung stated. The company said the architecture also improves read, write and input/output performance while reducing power consumption, making it better suited for AI servers that increasingly depend on high-capacity flash storage. Beyond BV-NAND, Samsung outlined two longer-term memory concepts that could radically alter AI system architecture. The first, dubbed zHBM, calls for stacking HBM memory on top of the AI accelerator rather than alongside processors, as is done today. This shortens the distance data must travel between processor and memory. Samsung said the design could dramatically increase bandwidth while reducing power consumption and thermal resistance. The company estimates that combining the architecture with wafer-bonding technology could ultimately deliver more than ten times the memory density of conventional HBM5 while tripling energy efficiency and cutting thermal resistance by more than half. But don’t plan for deployment just yet. zHBM is still a research concept. It illustrates how memory manufacturers are increasingly looking too 3D designs to continue scaling as traditional 2D packaging becomes more difficult. Samsung also introduced zNAND-O, another conceptual architecture designed to extend three-dimensional memory beyond conventional NAND implementations. Details were scant but Samsung did say zNAND-O was a next-generation high-performance NAND solution built on its V-NAND technology and in development in four- and eight-layer versions. The technologies reflect how the industry is being driven by AI, and that AI concerns are driving chip development. HBM Has emerged as an important component of AI computation, but very quickly the industry hit limitations in terms of bandwidth and speed. The proposed technologies above reflect Samsung’s attempts to alleviate the bandwidth problem.
Score: 63🌐 MovesAug 7, 2026https://www.networkworld.com/article/4206818/samsung-offers-future-ai-memory-roadmap.html - Hackers Targeted Major Wall Street Money Managers With Cloned Voices. A $75 Billion Hedge Fund Stopped Them
While this attack method isn’t a new phenomenon, cheap voice cloning makes it easier to scale. Experts say every sensitive request now needs an independent check.
- cAI boom to drive global M&A to five-year high
cAI boom to drive global M&A to five-year high 매일경제
- China’s July Trade Growth Tops Forecasts as AI Demand Lifts Export Prices
China’s July Trade Growth Tops Forecasts as AI Demand Lifts Export Prices Caixin Global
- Microsoft Open Sources code-testing-generator: a Polyglot Unit-Test Agent That Hits 92.1% Task Completion Versus 78.9% for Stock Copilot
Microsoft Open Sources code-testing-generator: a Polyglot Unit-Test Agent That Hits 92.1% Task Completion Versus 78.9% for Stock Copilot MarkTechPost
Score: 62🌐 MovesAug 7, 2026https://www.marktechpost.com/2026/08/06/microsoft-open-sources-code-testing-generator/ - Agentic AI workforce is more than doubling year on year, says Salesforce
Salesforce customers more than doubled their agentic workforces year on year, according to the company’s second annual Agentic Enterprise Index , which looks at trends in AI agent development and deployment over the past five quarters. It compiled data from customers who had activated agents in production every month of the analysis period to determine how their use of the technology has evolved between February 2025 and April 2026, as well as incorporating data from May 2026 Salesforce research studies. It found that businesses grew their agentic workforces from an average of five agents in February 2025 to 13 by April 2026, a 7% compound monthly growth rate (CMGR). In April 2026, it only took an average of 1.9 days to deploy an agent into production, a 53% decrease since the beginning of the report period. Not only were agents deployed more quickly, they have been progressively taking on more work once in use; over the 15 months, the average number of actions per account had a CMGR of 31%. “These agents are expanding beyond their initial scope to really become cross-functional,” said Caila Schwartz , Salesforce’s head of agentic commerce insights, during a media briefing. Salesforce has attempted to measure how much work agents perform, rather than how many tokens they consume, creating its own Agentic Work Unit (AWU) metric , although analysts have criticized the measure as being unrelated to business outcomes. Nevertheless, Salesforce said that as of April, Agentforce agents had performed 734 million AWUs, a number growing at about 15% each month. The research also showed that agents are acting across multiple cloud domains which, the company said, “underscores the practical necessity of a headless architecture. By decoupling the agent’s logic from traditional front-end user interfaces, agents can process tasks, execute actions, and trigger workflows anywhere.” Within the company, Salesforce itself has seen explosive growth in AI agent use, said Joe Inzerillo , president of enterprise & AI technology at Salesforce, with a threefold increase in sessions between February 2025 and April 2026. He said that the AI agent in Slack, Slackbot, saves the average employee five hours per week, with 83% of the company having adopted it. But Schwartz pointed out that different industries are approaching agentic AI in different ways, some more sophisticated than others. To measure that, Salesforce developed a Sophistication Index, a five-point scale scoring the cognitive complexity of an agent’s actions. Levels 1- 3 are assigned to tasks such as record lookups, drafting emails, or summarizing documents, while levels 4 and 5 include more complex functions such as updating database fields. The data showed that manufacturing, financial services, and healthcare and life sciences have built more sophisticated agent networks than what it called traditional AI frontrunners such as technology and retail. However, Inzerillo said, the most common use case industry wide, and the best place to start, is the service use case, which provides “far and away the best ROI to start with.” He also noted that, as people have become more conscious of what agents can do, they are asking agents to perform tasks, rather than simply answer questions. “Now what you’re starting to see people do is very action oriented. So instead of asking ’how do I file a form to request my vacation’ from our employee agent, they’re telling the employee agent, ‘hey I’m taking a vacation, you need to enter this form for me, and here’s the details,’” he said, adding that this bias towards action represents the evolution of agentic use.
- AI demand keeps China's export engine humming, but risks loom
AI demand keeps China's export engine humming, but risks loom Reuters
Score: 62🌐 MovesAug 7, 2026https://www.reuters.com/world/asia-pacific/chinas-july-exports-climb-239-yy-imports-up-275-2026-08-07/ - Anthropic loosens Fable 5's biology restrictions but keeps the guardrails on for virology and toxicology
Anthropic has cut false positives in its biology safety filters for Fable 5 by about 85 percent. Previously, nearly all biology-related queries got blocked and rerouted to the less capable Opus 5. The restrictions stay in place for sensitive dual-use topics like virology and toxicology. The article Anthropic loosens Fable 5's biology restrictions but keeps the guardrails on for virology and toxicology appeared first on The Decoder .
- Chesky says Airbnb will spend ‘a lot more’ on AI as earnings beat and stock surges 15%
Airbnb CEO Brian Chesky wasn’t sure AI would help the company, but a year later, he says it’s the reason growth is back.
- Virginia Moves Toward Regulating Self-Driving Vehicles Amid Safety, Liability Concerns
By Nathaniel Cline– Virginia Mercury Virginia is closer to regulating self-driving vehicles, which have steadily increased nationwide despite continued concerns over safety. Despite crashes that have led to restrictions in some areas, other states, such as Arizona and Texas, have …
- China Pushes AI Compute Into Orbit, and the Satellites Start Doing Their Own Thinking
In late July, a satellite called Chenguang-1 carried an eight-card server into orbit from the Dongfeng commercial space innovation test zone, marking a new step in China's space-based compute push. Industry experts say the country now sits in the global first tier for orbital compute.
Score: 60🌐 MovesAug 7, 2026https://pandaily.com/china-orbital-compute-chenguang-1-satellite-zhongke-tiansuan-aug2026 - Airbnb hits four-year high as investors cheer revenue forecast raise, AI payoff
Airbnb hits four-year high as investors cheer revenue forecast raise, AI payoff Reuters
- Cloudflare launches Kitesurf, a browser built for AI agents
Kitesurf is a cloud-hosted browser designed for AI agents instead of people. It uses less computing power than Chromium for common automation tasks, helping developers build browser-based AI agents more efficiently.
Score: 60🌐 MovesAug 7, 2026https://techcrunch.com/2026/08/07/cloudflare-launches-kitesurf-a-browser-built-for-ai-agents/ - Chinese AI boom sends Hong Kong data center prices soaring
Chinese AI boom sends Hong Kong data center prices soaring The Japan Times
Score: 60🌐 MovesAug 7, 2026https://www.japantimes.co.jp/news/2026/08/07/asia-pacific/ai-hong-kong-data-center/ - Cloudflare shares jump after forecast raise on higher AI-driven spending
Cloudflare shares jump after forecast raise on higher AI-driven spending Reuters
Score: 60🌐 MovesAug 7, 2026https://www.reuters.com/business/cloudflare-shares-jump-after-forecast-raise-ai-driven-demand-2026-08-07/