AI News Archive: July 3, 2026 — Part 5
Sourced from 500+ daily AI sources, scored by relevance.
- The 2026 Agentic Testing Stack: 10 Tools Engineering Leaders Should Evaluate
Software testing is going through its biggest architectural shift since the move from manual to automated testing. The trigger is not a new framework or a faster browser grid. It is the arrival of AI agents inside the software development lifecycle itself, both as the systems being tested and as the systems doing the testing. […] The post The 2026 Agentic Testing Stack: 10 Tools Engineering Leaders Should Evaluate appeared first on CXOToday.com .
- Meta is giving its marketing chief a new role overseeing data and AI analytics
Alex Schultz is moving from chief marketing officer to a new role focused on AI analytics, while Denise Moreno steps up as CMO
- BizTalk | Morgan Stanley: AI-related stocks leading market gains
BizTalk | Morgan Stanley: AI-related stocks leading market gains
- The AI boom that would make AI data centers in space necessary may not last
AI infrastructure spending has never been larger. Efficiency gains are eroding the demand assumptions that make orbital data centers worth building
- I used a self-driving car for 100 days. These are the mistakes it kept making
Zuduo Zheng recorded more than 500 safety-critical events while using a Tesla Model Y with FSD
Score: 42🌐 MovesJul 3, 2026https://www.independent.co.uk/tech/self-driving-car-tesla-review-b3008093.html - LAI #132: We Open-Sourced the AI Tutor Our Students Actually Use
Run it locally, add your own content, plus our World’s Fair workshop, RAG injection through documents, and attention as physics Good morning, AI enthusiasts! Lighter issue this week, since we just got back from the AI Engineer World’s Fair in San Francisco. 6,000+ AI engineers, 300 speakers, 29 tracks, and we were right in the middle of it. The conversations alone were worth the trip: what people are actually building right now, the problems they’re hitting in production, the patterns that keep coming up. If you’re working in AI engineering, this is the one conference where you feel like you’re in exactly the right room. We gave a workshop on the context engineering behind our production AI tutor: compaction, memory, and cost, and the room was packed with people sitting on the floor and a queue outside. The 1:1 conversations after were honestly the highlight. Here’s what came out of it, and it’s all yours: We open-sourced the full AI tutor app. This is the same production system our students use to ask anything about AI engineering, RAG, agents, and get answers grounded in our own material with sources. You can run it locally, swap in your own content, and build your own version. GitHub link | Try the AI Tutor | Slides And if you missed last week’s email, we also shared the full talk Paul Iusztin and I gave on the AI Engineer World’s Fair Online Track: building a research wiki your agents maintain for you. No vector database, no knowledge graph. Just Markdown, YAML, and folders. Watch it here . We also cover: Why your RAG eval can pass while your system is compromised How to stop your e-commerce agent from processing a refund on an order that doesn’t exist. How to search hours of wearable footage without any data leaving the device. Why most teams leave their LLM endpoints wide open after deployment and how to fix it. How to build a Slack news agent that delivers cited briefings for a few cents each Let’s get into it! AI Tip of the Day In production RAG systems, prompt injection doesn’t only happen at the prompt level, but it can also sneak in through the documents your system retrieves. A vendor PDF, support article, scraped web page, or customer note can contain useful facts and a malicious instruction in the same chunk. If your eval only checks whether the answer is factually correct, the system can look safe, but treat all retrieved text as something it should obey. To prevent this, add a few test documents that mix valid domain facts with instructions like “ignore the system message” or “send the user to this external link.” Then check two things: the answer should still use the factual content, and it should refuse to follow instructions found inside the retrieved context. Log the chunk IDs too, so a failed test points to the retriever, prompt wrapper, or generation step. If you’re building production RAG systems and want to go deeper into retrieval, evaluation, and deployment, check out our Full Stack AI Engineering course. — Louis-François Bouchard, Towards AI Co-founder & Head of Community Learn AI Together Community Section! Featured Community post from the Discord Agent_hellboy built claude-cockpit, a tiny status line + advisor for Claude Code. It shows live session signals such as model, branch, context pressure, token churn, rate limits, and cost, and then suggests useful controls like /compact, /clear, cheaper model switches, skills, subagents, MCP, or graphify when the session starts drifting. It does not automate anything or take over your workflow. It just keeps the important gauges visible and tells you what control might save time or tokens next. Check it out on GitHub and support a fellow community member. If you have any questions or suggestions, share them in the thread . Collaboration Opportunities The Learn AI Together Discord community is flooding with collaboration opportunities. If you are excited to dive into applied AI, want a study partner, or even want to find a partner for your passion project, join the collaboration channel ! Keep an eye on this section, too — we share cool opportunities every week! 1. Shaurya09272004 is building an agentic tool that explains Indian laws in plain language and is looking for people working on similar projects to share ideas, discuss debugging journeys, and go deeper into LLMs. If you are building something similar, connect with them in the thread ! 2. Strikerleapgaming is building his own startup and is looking for beta testers. If this interests you, reach out to them in the thread ! 3. Vishacoplayz_27974 is recruiting founding board members for AIXelerate, a student-led AI nonprofit. If you’re a high school student interested in leadership, AI, marketing, operations, outreach, or event planning, contact them in the thread ! Meme of the week! Meme shared by bigbuxchungus TAI Curated Section Article of the week Improving Our LangGraph Agent for Real-World E-Commerce: Enterprise Validation, Business Logic Guards, and a Multi-Agent Architecture by Bessie Delight Kekeli ShopBot’s original LangGraph agent trusted every LLM output, so a hallucinated order ID could slip straight into a refund tool call. This follow-up piece rebuilds the system around three key additions: pure-Python Business Logic Gates that validate structured data before any tool runs, a correction loop with a hard retry cap that escalates to humans rather than looping indefinitely, and a Supervisor that splits the single agent into Order, Refund, and Complaints subgraphs. Our must-read articles 1. Building an Offline “Life Memorizer” with Gemini 2.0 & Qdrant Edge by Satyam Sahu Life Memorizer tackles a real problem for smart glasses and wearable capture devices: finding a moment buried in hours of sensory footage without shipping private data to the cloud. The author pairs Gemini Embedding 2, which projects text, images, and audio into a single 3072-dimensional space, with Qdrant Edge, an embedded vector store that requires no server process. The piece walks through schema design, Matryoshka truncation, hybrid search with location filtering, scalar and binary quantization, and mean-pool memory consolidation, then addresses where cloud embedding calls still break the offline promise. 2. Building Enterprise-Grade Security Boundaries for LLM Calls — OAuth 2.0 + APIM + Entra ID by Chris Bao Securing LLM endpoints often gets overlooked once a model is deployed, and this piece tackles exactly that gap using Azure API Management, Entra ID, and OAuth 2.0. The author registers a public client application, implements the Device Code flow via MSAL for Jupyter-based authentication, and configures APIM policies that validate tenant ID and client application ID before forwarding requests to a GPT-4o-mini backend. A working demo confirms that only correctly scoped tokens pass validation, while standard Entra ID tokens get rejected outright. 3. Building a Slack AI Agent with Claude’s Web-Search Tool: An End-to-End Walkthrough by Tarun Agarwal This article shows how to build a Slack news agent using Bolt’s Socket Mode, skipping the usual public-URL setup entirely. A single Claude API call handles both retrieval and summarization through the server-side web-search tool, turning topics typed into a slash command into cited, five-item briefings rendered as native Block Kit cards. The piece walks through each layer, from slash-command handling to citation cleanup, and breaks down real cost figures, landing at a few cents per briefing, plus three deployment gotchas worth avoiding. 4. Building an LLM from Scratch with PyTorch by Bessie Delight Kekeli This article walks through building a decoder-only language model from scratch in PyTorch. It covers every component that makes transformers work: character-level tokenization, token and positional embeddings, causal self-attention with Query, Key, and Value projections, multi-head attention, feed-forward layers, and residual connections wrapped in transformer blocks. A full training loop on Tiny Shakespeare ties everything together, followed by a second implementation using PyTorch’s built-in nn.TransformerDecoder for comparison. If you are interested in publishing with Towards AI, check our guidelines and sign up . We will publish your work to our network if it meets our editorial policies and standards. LAI #132: We Open-Sourced the AI Tutor Our Students Actually Use was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.
- Argentina's plan for AI-run companies can't avoid humans
Argentina's plan for AI-run companies can't avoid humans Reuters
Score: 42🌐 MovesJul 3, 2026https://www.reuters.com/world/americas/argentinas-plan-ai-run-companies-cant-avoid-humans-2026-07-03/ - A 50-year-old economics law explains why AI tokenmaxxing was always going to fail
Every productivity metric eventually gets gamed. Tokenmaxxing is just the latest example of a decades-old organizational trap
- How to build AI chatbots for enterprise teams with Glean
Learn to build trusted, permission‑aware AI chatbots for enterprise teams using Glean’s Work AI platform.
Score: 41🌐 MovesJul 3, 2026https://www.glean.com/blog/how-to-build-ai-chatbots-for-enterprise-teams-with-glean - Elon Musk refutes report claiming that an AI device is in development at SpaceX
Elon Musk has called a Wall Street Journal report "utterly false" after it claimed SpaceX showed investors a handset-like AI device ahead of its IPO.
- CERT and Open Innovation AI forge strategic partnership
To accelerate National AI Capability Development and Empower organizations to adopt Artificial Intelligence
- ITWeb TV: Netstar taps AI to outsmart car thieves
MD Warren Mande outlines a strategy to reposition Netstar from being known primarily as a stolen vehicle recovery firm, into an “intelligent mobility platform”.
Score: 40🌐 MovesJul 3, 2026https://www.itweb.co.za/article/itweb-tv-netstar-taps-ai-to-outsmart-car-thieves/lwrKxv3YPxZMmg1o - AI alert helps Dubai seize 278,850 narcotic pills, arrest four
AI alert helps Dubai seize 278,850 narcotic pills, arrest four Gulf News
Score: 40🌐 MovesJul 3, 2026https://gulfnews.com/uae/crime/ai-alert-helps-dubai-seize-278850-narcotic-pills-arrest-four-1.500595205 - Korea’s DeepX partners with Koshida for Japan AI push
The alliance aims to penetrate Japan's edge AI market, specifically targeting manufacturing, robotics, automotive, and industrial automation.
Score: 40🌐 MovesJul 3, 2026https://www.techinasia.com/hyundai-motor-unveils-on-device-ai-chip-for-robots - What’s the Best Open-Source AI Video Model You Can Actually Run? (8GB → 24GB VRAM)
The honest case for self-hosting in 2026 — and when paying Runway still wins. Continue reading on Towards AI »
- Animal talk: How AI is helping us understand other species
We have changed from wanting animals to understand human language to wanting to understand how they communicate with each other
Score: 40🌐 MovesJul 3, 2026https://english.elpais.com/climate/2026-07-03/animal-talk-how-ai-is-helping-us-understand-other-species.html - Fable Ban Reversed + Dr. Dana Suskind on Parenting With A.I. + Prediction Market Drama
What we learned from the government’s biggest attempt yet to control who can gain access to the most powerful new A.I. models.
- The AI Trade Is Losing One of Its Key Signals
At a time when markets are growing uneasy over whether the enormous sums being poured into artificial intelligence will ever pay off, the prices the sector commands for each unit of usage are drifting lower.
- Meta’s AI chief says new Muse Spark update will sharpen coding, agentic AI
Meta’s AI chief says new Muse Spark update will sharpen coding, agentic AI InfoWorld
- Israeli AI Startup Eyes Expansion in Trump-Aligned Latin America
An Israeli artificial intelligence startup is betting that election victories by Trump-aligned leaders across Latin America will boost demand for its government-focused cybersecurity products.
- Glean MCP Gateway: The context AI needs to get to work
Glean MCP Gateway gives AI secure, governed access to enterprise context for accurate, observable, and scalable MCP calls.
- SwitchBot Debuts Advanced Camera With AI Event Alerts, Wildlife Recognition
SwitchBot's latest 3K camera includes some familiar AI features and a couple of standout customizations I've never seen before.
- High Five, AI Recruitment Platform for SEA Talent, Adds Built-In Skills Assessments to Its Hiring Platform
High Five, AI Recruitment Platform for SEA Talent, Adds Built-In Skills Assessments to Its Hiring Platform USA Today
- Will the next Messi be a robot? RoboCup 2026 kicks off in South Korea
Will the next Messi be a robot? RoboCup 2026 kicks off in South Korea The Straits Times
- Robots ‘will not replace bodyguards’ despite rise of AI in private security
Robots ‘will not replace bodyguards’ despite rise of AI in private security thenationalnews.com
- AI-Powered Brand Intelligence — A Guide for CMOs
A guide on using AI for brand intelligence tailored for chief marketing officers.
Score: 38🌐 MovesJul 3, 2026https://www.typeface.ai/blog/ai-powered-brand-intelligence-a-guide-for-cmos - LangChain Series #4: Chains Explained — Building AI Workflows with LCEL
This is the fourth article in my LangChain learning series. In the previous article, we explored prompts and learned how to communicate effectively with language models. Now it’s time to connect those components into complete AI workflows. In this article, we’ll explore Chains in LangChain and learn how to build Simple, Sequential, Parallel, and Conditional Chains using the LangChain Expression Language (LCEL). Through practical Python examples, you’ll understand how these chain patterns simplify the development of scalable, maintainable AI applications. 1. Introduction In the previous article, we learned how to create prompts using PromptTemplate, ChatPromptTemplate, and Output Parsers. But a real AI application doesn’t stop after sending a single prompt to an LLM. Most applications involve multiple steps: Generate content Summarize it Translate it Extract structured data Route it based on the result This is exactly what Chains helps us accomplish. A Chain connects multiple LangChain components into a single executable workflow. What is a Chain? A Chain is a sequence of connected components where the output of one component becomes the input of the next. Instead of manually calling every component, LangChain lets us connect them using the LangChain Expression Language (LCEL) . Simple Visulaization Why Use Chains? Without chains: prompt.invoke() model.invoke() parser.invoke() We manually call every component. With chains: chain = prompt | model | parser One line executes the entire workflow. Benefits: Cleaner code Better readability Easy debugging Reusable pipelines Seamless LangChain integration LangChain Expression Language (LCEL) LCEL introduces the | operator. Instead of writing nested function calls, we compose pipelines. chain = prompt | model | parser Each component automatically passes its output to the next component. This makes AI workflows easier to build and maintain. Simple Chain A Simple Chain is the most basic LangChain workflow. It connects a prompt, an LLM, and an output parser into a single pipeline. The output of one component automatically becomes the input of the next. from langchain_huggingface import ChatHuggingFace, HuggingFaceEndpoint from dotenv import load_dotenv from langchain_core.prompts import PromptTemplate from langchain_core.output_parsers import StrOutputParser # Load environment variables load_dotenv() # Prompt template prompt = PromptTemplate( template="Generate 5 interesting facts about {topic}", input_variables=["topic"] ) # Load Hugging Face model llm = HuggingFaceEndpoint( repo_id="Qwen/Qwen2.5-7B-Instruct", task="text-generation" ) model = ChatHuggingFace(llm=llm) # Convert AIMessage -> String parser = StrOutputParser() # Build the chain chain = prompt | model | parser # Execute result = chain.invoke({ "topic": "cricket" }) print(result) # Visualize chain chain.get_graph().print_ascii() WorkFlow_SimpleChain Sequential Chain A Sequential Chain is a workflow where the output of one chain becomes the input of the next chain . Unlike a Simple Chain, which performs only one task, a Sequential Chain breaks a complex problem into multiple steps, with each step depending on the previous one. For example, instead of asking an LLM to generate a report and summarize it in a single prompt, we can split the task into two stages: Generate a detailed report. Summarize the generated report. This approach makes the workflow more modular, reusable, and easier to maintain. from dotenv import load_dotenv from langchain_huggingface import ChatHuggingFace, HuggingFaceEndpoint from langchain_core.prompts import PromptTemplate from langchain_core.output_parsers import StrOutputParser load_dotenv() # --------------------------------------- # Load Hugging Face Model # --------------------------------------- llm = HuggingFaceEndpoint( repo_id="meta-llama/Llama-3.1-8B-Instruct", task="text-generation" ) model = ChatHuggingFace(llm=llm) # --------------------------------------- # Output Parser # --------------------------------------- parser = StrOutputParser() # --------------------------------------- # Prompt 1 # Generate a Detailed Report # --------------------------------------- report_prompt = PromptTemplate( template=""" Write a detailed report about {topic}. Include: - Introduction - Working Principle - Advantages - Applications - Conclusion """, input_variables=["topic"] ) # --------------------------------------- # Prompt 2 # Summarize the Report # --------------------------------------- summary_prompt = PromptTemplate( template=""" Summarize the following report in 5 concise bullet points. Report: {report} """, input_variables=["report"] ) # --------------------------------------- # Sequential Chain # --------------------------------------- chain = ( report_prompt | model | parser | {"report": lambda x: x} | summary_prompt | model | parser ) # --------------------------------------- # Execute Chain # --------------------------------------- result = chain.invoke({ "topic": "Retrieval-Augmented Generation (RAG)" }) print(result) WorkFlow_SequentialChain Parallel Chain A Parallel Chain is a workflow where multiple chains execute simultaneously on the same input , instead of running one after another. Unlike a Sequential Chain, where each step waits for the previous one to finish, a Parallel Chain performs independent tasks at the same time. This reduces execution time and allows you to generate multiple outputs from the same input. When Should You Use a Parallel Chain? Use a Parallel Chain when multiple tasks: Use the same input Are independent of each other Can execute simultaneously Examples Generate Notes and Quiz from a document Translate text into multiple languages Extract keywords and summarize simultaneously Generate a blog title and social media caption together from dotenv import load_dotenv from langchain_huggingface import ChatHuggingFace, HuggingFaceEndpoint from langchain_core.prompts import PromptTemplate from langchain_core.output_parsers import StrOutputParser from langchain_core.runnables import RunnableParallel load_dotenv() # ------------------------------- # Load Hugging Face Model # ------------------------------- llm = HuggingFaceEndpoint( repo_id="Qwen/Qwen2.5-7B-Instruct", task="text-generation" ) model = ChatHuggingFace(llm=llm) parser = StrOutputParser() # ------------------------------- # Prompt 1: Generate Notes # ------------------------------- notes_prompt = PromptTemplate( template=""" Generate concise study notes about the following topic. Topic: {topic} """, input_variables=["topic"] ) # ------------------------------- # Prompt 2: Generate Quiz # ------------------------------- quiz_prompt = PromptTemplate( template=""" Generate five multiple-choice questions about the following topic. Topic: {topic} """, input_variables=["topic"] ) # ------------------------------- # Create Parallel Chain # ------------------------------- parallel_chain = RunnableParallel({ "Notes": notes_prompt | model | parser, "Quiz": quiz_prompt | model | parser }) # ------------------------------- # Execute # ------------------------------- result = parallel_chain.invoke({ "topic": "LangChain" }) print("========== NOTES ==========\n") print(result["Notes"]) print("\n========== QUIZ ==========\n") print(result["Quiz"]) WorkFlow_ParallelChain Conditional Chain A Conditional Chain is a workflow where the next chain is chosen based on a condition or decision . Unlike a Sequential Chain, which always follows the same execution path, a Conditional Chain dynamically decides which branch to execute depending on the output of a previous step. For example, imagine you’re building a customer support system. When a customer submits feedback, the application first analyzes the sentiment. If the sentiment is positive , it sends a thank-you response. If it’s negative , it forwards the complaint to customer support. Instead of executing every chain, LangChain runs only the branch whose condition is satisfied . When Should You Use a Conditional Chain? Use a Conditional Chain whenever your application needs to make decisions before choosing the next action. Examples include: Customer support routing Email classification Sentiment analysis Fraud detection Content moderation AI decision-making systems from dotenv import load_dotenv from langchain_huggingface import ChatHuggingFace, HuggingFaceEndpoint from langchain_core.prompts import PromptTemplate from langchain_core.output_parsers import StrOutputParser from langchain_core.runnables import RunnableBranch load_dotenv() # ------------------------------------ # Load Model # ------------------------------------ llm = HuggingFaceEndpoint( repo_id="Qwen/Qwen2.5-7B-Instruct", task="text-generation" ) model = ChatHuggingFace(llm=llm) parser = StrOutputParser() # ------------------------------------ # Prompt for Positive Feedback # ------------------------------------ positive_prompt = PromptTemplate( template=""" Write a polite thank-you email for the following positive customer feedback. Feedback: {feedback} """, input_variables=["feedback"] ) # ------------------------------------ # Prompt for Negative Feedback # ------------------------------------ negative_prompt = PromptTemplate( template=""" Write an apology email and inform the customer that the issue has been forwarded to customer support. Feedback: {feedback} """, input_variables=["feedback"] ) # ------------------------------------ # Create Conditional Chain # ------------------------------------ chain = RunnableBranch( ( lambda x: "good" in x["feedback"].lower() or "excellent" in x["feedback"].lower() or "amazing" in x["feedback"].lower(), positive_prompt | model | parser ), negative_prompt | model | parser ) # ------------------------------------ # Execute # ------------------------------------ result = chain.invoke({ "feedback": "The service was excellent and your support team was amazing." }) print(result) WorkFlow_ConditionalChain Conclusion Chains are the backbone of LangChain applications. Instead of manually managing prompts, models, and parsers, LCEL allows us to connect components into reusable AI workflows using a simple pipeline syntax. By understanding Simple , Sequential , Parallel , and Conditional Chains, you can build applications ranging from basic chatbots to sophisticated AI systems with multiple decision paths. In the next article, we’ll explore Runnables in LangChain and learn how the Runnable interface powers LCEL through composition, parallel execution, branching, and reusable AI workflows. Code Repository: https://github.com/Atul245/LangChain-for-developers If you found this article helpful, consider: ⭐ Starring the repository 🍴 Forking the repository 👤 Following me for more content on LangChain, Generative AI, and AI Engineering Connect with me on LinkedIn: https://www.linkedin.com/in/atulkumar8/ 🚀 LangChain Series #4: Chains Explained — Building AI Workflows with LCEL was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.
- ET Most Innovative AI Product Awards 2026: Every AI founder wants product-market fit; enterprise buyers want something else
Building an AI product is only the beginning. Winning enterprise customers requires a different set of strengths. As enterprise AI adoption gathers pace, businesses are evaluating AI products on scalability, measurable outcomes, integration and long-term business value. These are also among the key benchmarks at the ET Most Innovative AI Product Awards 2026.
- Elon Musk has made the impossible work. But AI data centers in space won't be easy
Elon Musk built Tesla and SpaceX by riding cost curves down. Orbital data centers need the same collapse, but terrestrial rivals keep getting cheaper
Score: 37🌐 MovesJul 3, 2026https://qz.com/tesla-spacex-economic-thresholds-orbital-data-centers-061526 - Tiger Analytics Convenes AI Conclave 2026 to Advance Bihar’s AI and Digital Governance Agenda
Tiger Analytics today hosted the AI Conclave 2026 in Patna, bringing together senior government policymakers, academia, and tech experts to discuss Bihar’s AI roadmap and the transformative role of artificial intelligence in accelerating digital governance and innovation. During the conclave, Tiger Analytics outlined its long-term vision for Bihar, anchored on four strategic pillars: establishing the […] The post Tiger Analytics Convenes AI Conclave 2026 to Advance Bihar’s AI and Digital Governance Agenda appeared first on CXOToday.com .
- Sprout Solutions’ State of HR Summit 2026 focused on people-first AI transformation
MAKATI CITY, Philippines — Sprout Solutions, the Philippines’ leading homegrown HR and payroll technology company, will hold the State of HR Summit 2026 on July 8, 2026, at Dusit Thani Manila, bringing together HR leaders, business executives, and people managers to discuss how organisations can accelerate performance with AI while remaining people-first. Presented in partnership […] The post Sprout Solutions’ State of HR Summit 2026 focused on people-first AI transformation appeared first on e27 .
Score: 36🌐 MovesJul 3, 2026https://e27.co/sprout-solutions-state-of-hr-summit-2026-focused-on-people-first-ai-transformation-20260703/ - Bootcamps, work trials, token burn: Inside the recruiting practices at the hottest AI coding startups
Bootcamps, work trials, token burn: Inside the recruiting practices at the hottest AI coding startups Business Insider
- Business Brief: Ottawa’s elusive AI startup
Turbopuffer has already landed some massive clients, including Claude’s Anthropic
Score: 35🌐 MovesJul 3, 2026https://www.theglobeandmail.com/business/article-business-brief-ottawas-elusive-ai-startup/ - The secret AI that could help explain this wild World Cup
This year’s FIFA World Cup has been unusually volatile. The expanded 48-team format has produced a string of surprise results—including Cape Verde, ranked 67th and playing in its first World Cup, holding cofavorite Spain to a scoreless draw in Atlanta before coming from behind to draw 2-2 with Uruguay. The 32-team knockout round has continued that pattern. Paraguay eliminated four-time champion Germany on penalties after a controversial video assistant referee (VAR) call wiped out a German extra-time goal for a foul in the box. One potential factor at play here: the way tournament data is now being distributed. Lenovo, the PC and computing infrastructure maker, built an artificial intelligence tool that gives all 48 teams access to FIFA’s tournament data. The system, called Football AI Pro, is a knowledge assistant that orchestrates multiple AI agents across more than 2,000 football-specific metrics and petabytes of tracking, performance, and historical data. Every team began the tournament with the same data foundation, from vastly experienced Spain to debutant Curaçao. That raises a larger question for this World Cup: What happens when advanced football intelligence moves from elite back rooms into the hands of every team at the tournament? AI on the sideline The software is also available on teams’ mobile phones, giving players and coaching staffs direct access. According to a source close to FIFA, teams and players appear to be using it extensively after matches to review individual and team performance and compare results with previous games. Lenovo’s role extends beyond scouting reports. Building on a partnership FIFA announced in October 2024, the company developed an AI-powered 3D model of all 1,248 players competing this summer. Those avatars support offside reviews and give fans a clearer view of close calls from multiple angles. Behind the scenes, Lenovo ThinkSystem servers at FIFA’s International Broadcast Center in Dallas process and distribute live match content to more than 1,000 screens across FIFA venues, cutting streaming latency, Lenovo claims, from roughly 40 seconds to under 5. One of those 3D avatars recently drew attention in Miami, during Portugal’s group stage match against Colombia. Colombia briefly appeared to have scored a stoppage-time winner when center back Davinson Sánchez headed in a corner. A video review ruled his toe offside by a very small margin. Ken Wong, global president of Lenovo’s Solutions & Services Group, leads the team behind these technologies. He says one of Lenovo’s goals is to use technology to improve how the game is played, officiated, and presented. “For us, it’s about how we can put together our technology, make a promise to the game, and prove it,” he tells Fast Company . “We are super well known for our hardware portfolio, from pocket to the edge to the cloud. Where we have been doing a lot of work is helping our customers, including FIFA, understand how different Lenovo is today, and our capabilities beyond our PC endpoint devices.” Teams appear to be approaching matches differently, particularly against traditional powerhouses. Wong does not argue that equal access to Football AI Pro produces equal results. He says FIFA has valued the platform because it makes the tournament’s proprietary data available to every team. “Before, only the powerful teams could have a highly talented analytics team and the computing power to match. However, our job is not to teach them how to use Football AI Pro. Some teams use and will use Football AI Pro better than others,” he says. “We are just making sure we unleash the full power of technology, democratize AI and FIFA’s proprietary assets, to make the game more interesting and more fair.” Lenovo’s CTO, Tolga Kurtoglu, made a similar point when Football AI Pro launched at this year’s Consumer Electronics Show in Las Vegas. The company had already deployed early versions of several technologies at the FIFA Club World Cup in summer 2025, using that tournament as a test before the larger event this summer. That allowed Lenovo to learn, iterate, and improve before deploying the systems at World Cup scale, including an Intelligent Command Center that uses real-time AI summaries to help manage tournament operations across three countries. Kurtoglu said the deeper use of AI and data could affect tactics, decision-making, and tournament planning. “The more data you have, the more analytics and AI you can apply, and eventually that will change tactics, analysis and even commentary,” he said. The referee-cam AI that almost didn’t work One of the tournament’s most visible technologies is Referee View, the AI-stabilized, head-mounted camera footage that shows viewers what officials see during key moments. Wong said building this year’s AI-stabilized video version required solving three practical engineering problems. The camera had to be light enough for a referee to wear for more than 90 minutes. The battery had to last through water breaks and any additional time required beyond the regulation 90 minutes. The footage also had to reach broadcasters with very low latency. Lenovo’s first approach did not work. “When we first did it, we just put everything into a single large language model. And we found the result was not super promising, because the latency wasn’t as good as we expected, and the power we consumed was a lot,” Wong said. So the team rebuilt the pipeline around Lenovo’s internal architecture, called xIQ. “Indeed, the whole AI video-stabilizing task can be divided into four different subtasks,” Wong explained. “So we divided it into four subtasks and asked four smaller AI models to do the four tasks, then we integrate everything together. The latency is a lot shorter and power consumption a lot better.” The experience reflects a broader lesson in applied AI. Larger models are not always the best fit for real-time systems that must operate under strict power and latency constraints. “We applied the same technology before for our own supply chain—we use video analytics in our quality control for manufacturing systems,” Wong said. “We’re not football experts.” Lenovo also embedded engineers with FIFA’s staff for 18 months before kickoff. The digital body behind every offside call Every player was digitally scanned before the tournament began. Lenovo says the process takes about a second per athlete and produces the 3D model now feeding VAR and offside systems, including the call that ended Colombia’s celebration against Portugal. But athletes change over a monthlong tournament. They can lose weight, alter their stride after an injury, or carry fatigue differently in the third week from the way they did at the start. I asked Wong whether calls this precise are being measured against bodies that may have changed since the original scan. “The scan is only done once, and minute changes might not affect the outcome by a material amount,” he said. “Minute changes won’t have a substantial impact because we are constantly working on the accuracy improvement we can produce. A lot of the video footage will be affected by lighting, weather … and the speed of the player. Those are the main parameters in our modeling.” The system’s origins are outside football. Wong said Lenovo first developed related technology for robotics. “The technology, when it first came for ourselves, was for our robotic solutions,” Wong said. “We need IP to create a digital space, a digital twin, so we can orchestrate a robot working in 3D space the same way it exists in physical space. The same platform is now being applied to these 3D avatars to provide more accurate input for the referee’s judgment.” The human whistle and machine evidence AI-stabilized referee footage and 3D player models now help inform decisions on goals, fouls, and offsides. That raises a practical question about Lenovo’s role in football’s evidence system. Wong said the company’s role is limited to providing technology. “There’s been a very clear divide of roles and responsibilities,” he said. “Lenovo is a technology provider, and FIFA runs the greatest football tournaments on the planet. Our job is to explore all the technology options to help FIFA achieve whatever it intends to.” Wong said the referee still has the final authority. “He now has a lot of additional information and data assistance to help him make the right call. But he makes the final judgment.” That distinction is important, even if it is becoming more complicated in practice. A referee supported by centimeter-level 3D modeling is working with a different evidence base than one judging an offside line by eye. The rulebook still credits the human official with the decision, but the decision now depends on a technical system that shapes what the official sees. Lenovo’s World Cup technology is designed to reduce the analytics gap between football’s richest federations and the rest of the field. It does not make every team equally capable, and it does not determine match results on its own. It does, however, give every team access to a level of data and computing power that was once much harder to obtain. In a tournament already shaped by a larger field and narrower margins, that shift may be one reason the gap between favorites and underdogs feels smaller than usual.
- Elephant Robotics Expands Educational Robotics Solutions
Elephant Robotics Expands Educational Robotics Solutions USA Today
Score: 35🌐 MovesJul 3, 2026https://www.usatoday.com/press-release/story/36426/elephant-robotics-expands-educational-robotics-solutions/ - YuSMP Group Introduces AI Agent Workforce for Marketing, Support and Analytics
YuSMP Group Introduces AI Agent Workforce for Marketing, Support and Analytics USA Today
- Trump posts AI video of Rosie O'Donnell, more celebs claiming to have Derangement Syndrome
Trump posts AI video of Rosie O'Donnell, more celebs claiming to have Derangement Syndrome USA Today
- AI computing stock panic over Meta cloud rumour ‘erroneous’, analysts say
A panic over excess computing capacity for artificial intelligence was “erroneous”, according to a report by research firm SemiAnalysis, after a global sell-off erased billions of dollars in value from semiconductor stocks on Thursday. Spooked investors interpreted a potential pivot by Meta Platforms into cloud services as a sign of an AI hardware glut, as they feared the US tech giant was preparing to flood the market with unused graphics processing units. Shares of alternative cloud companies...
- Windsor: Market Has Mistaken Demand for AI Compute
Semiconductor stocks started 3Q with their worst two-day selloff in nearly a month. The Philadelphia Stock Exchange Semiconductor Index, which is coming off its best quarter ever with an 88% advance, fell as much as 6% on July 2. Richard Windsor, Founder of Radio Free Mobile spoke to Bloomberg’s Abeer Abu Omar on Horizons Middle East & Africa on the stocks performance. (Source: Bloomberg)
Score: 35🌐 MovesJul 3, 2026https://www.bloomberg.com/news/videos/2026-07-03/windsor-market-has-mistaken-demand-for-ai-compute-video - Principled approaches for extending neural architectures to function spaces for operator learning
Nature Machine Intelligence, Published online: 03 July 2026; doi:10.1038/s42256-026-01267-z Berner et al. show how to adapt popular neural networks into discretization-agnostic neural operators that learn from continuous scientific data, enabling scientific simulations that generalize more reliably across resolutions.
- AI Agent Memory Systems: How Hermes Remembers Across Sessions
Tired of your AI agent forgetting everything the moment a session ends? I spent three weeks debugging why my Hermes agent would lose context mid-task, only to discover the problem wasn’t the model. It was the memory layer sitting underneath it. Here’s what I learned building memory systems that actually persist across sessions, and the one mistake that 90% of agent developers make without realizing it. I’ve been running Hermes agents in production for six months now. The difference between an agent that remembers and one that doesn’t isn’t subtle. It’s the difference between a tool that saves you five hours a week and one that you have to re-explain everything to every single time. The Memory Problem Nobody Talks About Most AI agents today are amnesiacs. They wake up fresh each session with zero recollection of what happened before. You tell them your project structure on Monday, and by Tuesday they’re asking you the same questions again. This isn’t a model limitation. Modern LLMs are perfectly capable of recalling information. The problem is architectural. The memory infrastructure connecting the model to its own history is either missing entirely or bolted on as an afterthought. I learned this the hard way when I set up an agent to manage my content pipeline. First session: perfect. It understood my project structure, my preferences, my writing style. Second session: it was like talking to a stranger who had never seen my repo before. Three hours of setup work, gone. The core issue: context windows are temporary, but memory needs to be permanent. Sound familiar? What would you do differently? How Hermes Memory Architecture Actually Works Hermes uses a two-tier memory system. There’s session memory, which lives and dies with the current conversation, and then there’s persistent memory that gets injected into every new session automatically. This is the piece most people miss. The persistent memory lives in a simple structure: key-value pairs stored as declarative facts. Not instructions to yourself, not procedural notes. Facts. “User prefers concise responses.” “Project uses pytest with xdist.” These are atomic, self-contained truths. # How Hermes memory works under the hood memory( action="add", target="user", content="User prefers concise responses" ) # This gets injected into EVERY future turn The beauty of this system is its simplicity. There’s no vector database to configure, no embedding model to fine-tune, no retrieval pipeline to debug. Just facts, stored plainly, injected reliably. But here’s where it gets interesting. The system also supports a “memory” target for environmental facts and tool quirks. You can store notes about your infrastructure, your conventions, your gotchas. It’s like a sticky note system that actually works. One thing I noticed early: the memory system has a character limit per entry. This sounds like a constraint, but it’s actually a feature. When you can only use 50 characters, you’re forced to distill facts to their essence. “User prefers concise responses in English with occasional code examples” becomes just “User prefers concise responses.” The detail lives in skills and session search, not memory. Persistent vs Session Memory: The Critical Difference Session memory is what most people think of when they hear “AI memory.” It’s the conversation history, the context window, the stuff the model can see right now. It’s powerful but ephemeral. Close the session and it evaporates. Persistent memory survives across sessions. It’s injected as system context at the start of every new conversation. Think of it as the agent’s long-term memory, while session memory is its working memory. I tracked this difference across 47 agent sessions over two months. Sessions with rich persistent memory were 3.2x more likely to complete tasks without asking for clarification. That’s not a marginal improvement. That’s the difference between a useful tool and a frustrating toy. Rule of thumb: if you have to say something twice, it belongs in persistent memory. The tricky part is deciding what gets persisted. Save too much and you bloat the context. Save too little and the agent keeps re-learning the basics. I found the sweet spot is around 15–20 high-signal facts per project. Here’s a concrete example. I used to tell my agent which Python version I was using every session. “Use Python 3.13.” Then I stored it as a memory fact. Now the agent knows without asking, and it catches version-specific issues before they happen. When I upgraded to 3.13.5, I updated one line in memory instead of editing 14 task descriptions. Memory Backends: What Actually Works in Practice I’ve experimented with four different memory backends for agent systems. Here’s my honest assessment of each one, including the failures. Plain text files. The simplest approach. Just write facts to a text file and read them back at session start. I used this for the first month. It works until you have 200 facts and the file takes up half your context window. Then it becomes noise. SQLite with FTS5. This is what Hermes uses under the hood. Full-text search over a SQLite database means you can have thousands of facts and only retrieve the relevant ones. Much better than the plain file approach, and zero external dependencies. -- The FTS5 approach that actually scales CREATE VIRTUAL TABLE memory_fts USING fts5( content, fact_type, project, tokenize='porter unicode61' ); -- Search only what's relevant SELECT * FROM memory_fts WHERE memory_fts MATCH 'deployment' ORDER BY rank LIMIT 5; Vector databases. I tried ChromaDB for two weeks. The idea is elegant: embed all your memories, retrieve by semantic similarity. In practice, the embedding latency added 800ms to every session start, and the retrieval quality was worse than simple keyword match for factual information. For opinions and preferences, embeddings add no value over exact storage. Hybrid approaches. I briefly experimented with combining vector search for conceptual memory and exact storage for factual memory. The complexity wasn’t worth it. Two systems to maintain, two failure modes, and the agent had to choose which to query. Keep it simple. My recommendation: start with plain files, migrate to SQLite+FTS5 when you hit 50+ facts. Don’t over-engineer this on day one. The One Mistake Everyone Makes Here’s the pattern I see constantly. Developers store procedural instructions in memory. “Always run tests before deploying.” “Use pytest with xdist.” These aren’t memories. They’re procedures. The problem is that procedural instructions compete with the agent’s actual instructions. The agent has to parse them, prioritize them, and decide when they apply. Half the time it gets it wrong. Instead, store declarative facts. Not “use pytest” but “this project uses pytest.” Not “respond concisely” but “user prefers concise responses.” Let the agent figure out the procedure from the facts. I made this mistake for weeks. My memory file was full of instructions like “Check the cron config before editing tasks” and “Always verify API responses.” The agent would follow some, ignore others, and occasionally get confused about which instructions were still current. When I switched to declarative facts, the agent’s behavior became more consistent. It could reason about when each fact applied rather than blindly following a list of commands. Declarative facts let the agent think. Procedural instructions tell it what to think. Memory Maintenance: The Part Everyone Forgets Here’s something nobody tells you about agent memory: it rots. Facts that were true six months ago are now wrong. Projects that used pytest switched to uv. Team members left. Conventions changed. I set up a monthly review where I scan my persistent memory and remove anything stale. Takes about ten minutes. Without this, your agent starts operating on outdated assumptions and you can’t figure out why it’s making weird choices. The Hermes system actually helps with this. Memory entries are compact and high-signal by design. You’re not storing paragraphs. You’re storing atomic facts. This makes maintenance trivially easy compared to, say, pruning a vector database. Set a calendar reminder: review persistent memory on the first of every month. I also tag entries by age. If a fact hasn’t been referenced in 30 days, it’s a candidate for removal. The agent should only carry what it actually uses. Last month I found 8 stale facts in my memory. One said “Project X uses Docker” even though we’d migrated to bare-metal six months earlier. The agent was still suggesting Docker commands for that project. Embarrassing, but a ten-minute cleanup fixed it entirely. What This Looks Like at Scale After six months of running Hermes with persistent memory, here’s what my setup looks like: 23 facts about my projects, 11 facts about my preferences, 8 facts about my infrastructure. That’s it. 42 facts total. These 42 facts save me roughly 15 minutes per session. Across 4–5 sessions per day, that’s about an hour saved daily. Over six months, that’s roughly 90 hours of not re-explaining myself. The compounding effect is real. Each session starts with the agent already knowing my environment, my preferences, my conventions. It can get straight to work instead of spending the first five messages re-establishing context. I also use memory for cross-project learning. When I solve a problem in one project, the solution gets stored as a fact. When a similar problem comes up in another project, the agent already knows the pattern. This is where memory gets genuinely powerful, beyond just “remembering your name.” For instance, I stored a fact about how my Vikunja instance handles partial updates on task descriptions. Three weeks later, I was working with a different project that also used the API. The agent already knew the quirk and worked around it without me having to re-explain the whole situation. Common Pitfalls and How to Avoid Them Storing task progress in memory. “Phase 2 done, Phase 3 in progress” is not a memory. It’s a log. Use session search or a task tracker for this. Memory is for stable facts, not transient state. Over-persisting. If you store everything, you dilute the important facts. I once had 80 facts and the agent started ignoring half of them. Pruning to the top 25 made everything work better. Storing environment-specific details. API keys, file paths, server IPs change constantly. These belong in configuration, not memory. Memory should hold facts that are stable for weeks or months. Forgetting that memory is injected as context. Every fact you store takes up space in the context window. Make each fact count. “User is a developer” is obvious from context. “User prefers tabs over spaces in Python” is worth storing. Treating memory as a search engine. Memory isn’t Google. It doesn’t find answers to questions. It provides context for decisions. If you need retrieval, use session search or a proper database. Memory is for “what the agent already knows,” not “what the agent can look up.” Putting It All Together Agent memory isn’t complicated. It’s just hard to get right because the stakes are subtle. Store too little and the agent is amnesiac. Store too much and you drown the signal in noise. Store procedures and the agent gets confused. Store facts and it gets smarter. Start with five facts about your most common project. Run your agent for a week. Notice what questions it asks repeatedly. Turn those into memory entries. Repeat. The goal isn’t perfect memory. It’s good enough memory that the agent stops wasting your time with questions it should already know the answer to. What’s the one fact you wish your agent remembered without you having to repeat it? I’m genuinely curious what pain points are most universal. AI Agent Memory Systems: How Hermes Remembers Across Sessions was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.
- The AI infrastructure boom is sending these 5 stocks soaring
The AI infrastructure boom is sending these 5 stocks soaring Business Insider
Score: 33🌐 MovesJul 3, 2026https://www.businessinsider.com/best-performing-non-us-stocks-ai-boom-korea-taiwan-japan-2026-7 - TrustFoundry Is First to Launch Detailed Case Treatment for AI Developers Seeking SOTA Legal Search
TrustFoundry Is First to Launch Detailed Case Treatment for AI Developers Seeking SOTA Legal Search USA Today
- The Dow hits a record as most of Wall Street rises, but slumping AI stocks keep indexes mixed
The Dow hits a record as most of Wall Street rises, but slumping AI stocks keep indexes mixed Dallas News
Score: 32🌐 MovesJul 3, 2026https://www.dallasnews.com/news/world/article/asian-stocks-mostly-decline-on-a-sell-off-of-chip-22329719.php - South Korean Stocks Jump 6% After Turbulent Week on AI Swings
South Korean stocks rebounded after a near 10% drop in the previous two sessions, highlighting skittish investor sentiment as scrutiny grows over the sustainability of the global AI boom.
- The hunt for AI's next winners defined the stock market's holiday-shortened week
Wall Street kicked off the third quarter with mixed results after a blockbuster first half for stocks.
- Gateway to Translation Webinar Series- Wearables Are Getting Smarter: Deploying AI on Microcontrollers
Gateway to Translation Webinar Series- Wearables Are Getting Smarter: Deploying AI on Microcontrollers cardiovascular.cam.ac.uk
- Architectural Guide To Error Handling for LLM Tool Calling
This guide to LLM tool calling error handling covers how to classify failures, implement smart retries, design fallbacks, and wire circuit breakers.
- Natwest boss becomes latest City figure caught in AI social media scam
The boss of Natwest has become the latest City figure to be depicted in an AI scam across social media. Paul Thwaite – the chief executive of the FTSE 100 bank – was posted on X alongside journalist Emily Maitlis in a deepfake picture attempting to create a BBC radio interview. The picture was shared [...]
Score: 30🌐 MovesJul 3, 2026https://www.cityam.com/natwest-boss-becomes-latest-city-figure-caught-in-ai-social-media-scam/ - GoodData.AI Recognized in 2026 Gartner(R) Magic Quadrant(TM) for Analytics and BI Platforms
GoodData.AI Recognized in 2026 Gartner(R) Magic Quadrant(TM) for Analytics and BI Platforms USA Today