AI News Archive: May 30, 2026 — Part 3
Sourced from 500+ daily AI sources, scored by relevance.
- Oxford MedTech Society x Google Gemini Clinical Hackathon
Oxford MedTech Society x Google Gemini Clinical Hackathon University of Oxford
Score: 38🌐 MovesMay 30, 2026https://enspire.ox.ac.uk/event/oxford-medtech-society-x-google-gemini-clinical-hackathon - The Billionaire Coding Genius Making the Tough Decisions at OpenAI
After years in the shadow of better-known co-founders, Greg Brockman is stepping into the spotlight. He and his wife, Anna, are also Silicon Valley superdonors.
Score: 38🌐 MovesMay 30, 2026https://www.wsj.com/tech/openai-who-is-greg-brockman-e699816c?mod=rss_Technology - 'Europe is kind of waking up': I went to Mistral's summit in Paris and heard a clear message about AI
'Europe is kind of waking up': I went to Mistral's summit in Paris and heard a clear message about AI Business Insider
Score: 37🌐 MovesMay 30, 2026https://www.businessinsider.com/mistral-ai-summit-europe-ai-future-waking-up-2026-5 - A hidden AI 'landmine' can upend your business. Here are the warning signs and best practices.
The price tag for a lack of accountability can be huge, but there are ways business owners can be on guard.
Score: 37🌐 MovesMay 30, 2026https://www.bizjournals.com/bizjournals/news/2026/05/30/ai-use-dangers-errors-accountability.html?ana=brss_6150 - Sutikshana Technologies Pvt. Ltd Launches Samwaad.ai, India’s First Personal AI Interview Coach
Sutikshana Technologies Pvt. Ltd launches Samwaad.ai, an India-first AI interview coach designed to help candidates prepare smarter for tech roles, US visa interviews, and competitive examinations. Getting an interview is hard. Converting it is harder. In most cases, what separates two equally qualified candidates is not what they know but how they communicate it. The confidence […] The post Sutikshana Technologies Pvt. Ltd Launches Samwaad.ai, India’s First Personal AI Interview Coach appeared first on CXOToday.com .
- Don't want to let an AI agent take over your machine? Here are 3 no-install OpenClaw alternatives you can try today
Don't want to let an AI agent take over your machine? Here are 3 no-install OpenClaw alternatives you can try today Tom's Guide
- Google Health has taken over from Fitbit on my phone, but I still don't trust AI enough to talk to me about my health
The Google Health Coach is an AI-powered assistant ready and waiting to improve your fitness — but is it useful?
- Google built the answer to your AI bill before you knew you had a problem
Google's Gemini 3.5 Flash rivals frontier AI models at a fraction of the cost, as companies burn through annual token budgets within months of the year starting.
- More than 550,000 students reached through Microsoft Stem and artificial intelligence programme
Communication tool for non-verbal students and road-safety system for country roads among projects presented at national showcase
- Ditch the Cloud: Build a Free, Local AI Coding Agent with llama.cpp
How to use local models to perform in your daily work without losing your PC’s performance. Source: Image by William Harrison on Sewe Table of Contents Introduction Download the Open-Weights Model Download and Prepare llama.cpp Create the Execution Script Install Node.js Install and Configure Qwen Code Install and Configure Opencode Time to Code! Conclusion References Introduction We all know the struggle: you want to use advanced AI agents for your daily coding tasks, but online subscriptions are expensive, privacy is a concern, and heavy local models can turn your PC into a space heater. What if I told you that you can run a fully functional, autonomous AI coding agent for FREE? In this guide (inspired by the excellent workflow shared by Nichonauta ), we will set up Qwen 3.5-4B locally. It is 100% free, entirely private, and highly optimized. By combining this lightweight model with the right agentic tools, you can automate complex coding tasks without bottlenecking your computer’s performance. Let’s dive into how to build your own local coding assistant. Download the Open-Weights Model You don’t need to pay for API keys; everything relies on open weights. Go to Hugging Face and search for unsloth/Qwen 3.5-4B-GGUF ( or any unsloth/model ). Download the Q4_K_XL (or MXFP4) quantization. This specific format is crucial because it maintains the best response quality while compressing the model down to a tiny 2.91 GB. (Optional but recommended) Download the Vision Encoder file in Files and versions (mmproj in BF16 format). This allows your AI to "see" images or analyze web browser screenshots later on. Download and Prepare llama.cpp llama.cpp is the magic engine that runs our model efficiently, utilizing either your GPU or CPU (RAM) without devouring system resources. Head over to the official llama.cpp GitHub repository . Download the compiled release that matches your hardware. If you have an Nvidia card, grab the Windows 64-bit CUDA version ( e.g., CUDA 12 or 13) and the CUDA 1x.x DLLs . Unzip the downloaded .zip files into a dedicated folder on your computer. Move your downloaded model files (the .gguf and the mmproj files) into one same folder to keep everything organized. Create the Execution Script llama.cpp file example To avoid typing a massive command line every time you want to work, create a simple .bat file (for Windows) to boot up the model as a local API server. Here are the key parameters you want to include in your script: Context Window: Set it high (e.g., 262144 native tokens) so the agent remembers the whole project. KV Cache at 8-bits: This significantly saves VRAM. Chain of Thought (CoT): Keep it active. This boosts the model’s reasoning capabilities, which is essential for programming. Sampling Parameters: Set Temperature to 1.0, Top P to 0.95, and Min P to 0.01 to prevent repetitive loops. .\llama-b8873-bin-win-cuda-13.1-x64\llama-server.exe ^ -m gguf\Qwen3.5-4B\Qwen3.5-4B-UD-Q4_K_XL.gguf ^ -c 262144 ^ -mg 1 ^ -sm none ^ --cache-type-k q8_0 ^ --cache-type-v q8_0 ^ --image-min-tokens 1024 ^ --reasoning on ^ --temp 0.6 ^ --top-p 0.95 ^ --top-k 20 ^ --min-p 0.00 ^ --presence-penalty 0.0 ^ --repeat-penalty 1.0 ^ --dry-multiplier 0.1 ^ --dry-base 1.05 ^ --dry-allowed-length 12 ^ --dry-penalty-last-n 128 ^ -a Qwen3.5-4B Install Node.js Our AI needs a visual interface and an agentic environment to work inside your projects. For this, we need a JavaScript runtime environment. Go to the official Node.js website . Download the installer and run through the standard setup (default settings are fine). Install and Configure Qwen Code We will use Qwen Code as our agent interface. It is visually appealing, easy to use, and natively designed for agentic software development. Open your terminal (PowerShell or CMD) as Administrator. Paste the installation command from the Qwen Code GitHub repository (usually an npm install -g @qwen-code/qwen-code@latestcommand). Once installed, run the app once and close it. This generates a hidden configuration folder. Navigate to C:\Users\YourUser\.qwen\ and open the settings.json file. Connect it to your local llama.cpp server by updating these fields: Provider: Set it to an OpenAI-compatible API. URL: Point it to your localhost (usually http://localhost:8080/v1). Model Name: Enter the exact name you specified in your .bat file. Context Window: Match the context length from your script (e.g., 262144) so the agent knows when to auto-compact its memory. { "modelProviders": { "openai": [ { "id": "llama.cpp", "name": "llama.cpp", "envKey": "LLAMA_CPP_API_KEY", "baseUrl": "http://127.0.0.1:8080/v1", "generationConfig": { "contextWindowSize": 262144 } } ] }, "security": { "auth": { "selectedType": "openai" } }, "model": { "name": "llama.cpp" }, "env": { "LLAMA_CPP_API_KEY": "llama.cpp" }, "$version": 3, "tools": { "approvalMode": "plan" }, "general": { "language": "en", "outputLanguage": "Español" }, "mcpServers": { "playwright": { "command": "npx", "args": [ "@playwright/mcp@latest" ] }, "brave-search": { "command": "npx", "args": [ "-y", "@brave/brave-search-mcp-server" ], "env": { "BRAVE_API_KEY": "BRAVE_API_KEY" } }, "context7": { "httpUrl": "https://mcp.context7.com/mcp/" } } } Install and Configure Opencode While Qwen Code is fantastic for straightforward tasks, you might want an environment that is even more robust and feature-rich. OpenCode is an extremely comprehensive agentic platform that gives you absolute control over your local AI. Here is how to set it up to talk with your local llama.cpp server: Install OpenCode: Open your terminal and install the tool globally (usually via npm with a command like npm install -g opencode-ai, or by downloading their desktop release ). Connect to Your Local API: Navigate to the OpenCode settings; it should be C:\Users\YourUser\.qwen\. Open the opencode.json file and paste this settings: { "$schema": "https://opencode.ai/config.json", "provider": { "llamacpp-local": { "name": "LlamaCPP local", "npm": "@ai-sdk/openai-compatible", "models":{ "Qwen3.5-4B": { "name": "Qwen3.5-4B" } }, "options": { "baseURL": "http://127.0.0.1:8080/v1" } } } } Time to Code! You are fully set up. Here is your new daily workflow: Double-click your .bat file to start llama.cpp quietly in the background. Navigate to your project folder using your terminal. Type qwen or opencode and hit enter. Depending on the complexity of your task, you have three distinct ways to interact with your local model. The Guided Assistant: Qwen Code Qwen Code is fantastic when you want an interactive, visually appealing interface to build features step-by-step. It acts as a collaborative partner. How to use it: Open your terminal in your project folder, type qwen, and switch to Yolo Mode . Example Prompt: “Using uv for dependencies, write a Python script using LangChain and FAISS to set up a basic Retrieval-Augmented Generation (RAG) pipeline. Ensure the code is modular and well-commented.” The Autonomous Developer: OpenCode When you need an agent to take the wheel, manage multiple files, and handle complex workspace architecture, OpenCode is your heavy-duty tool. How to use it: Launch opencode in your terminal and let it read your environment. Example Prompt: “Analyze my current workspace. Refactor the existing agentic chatbot scripts into a cleaner folder structure, update the relative paths, and run a quick test to ensure no dependencies are broken.” The Quick Sandbox: llama.cpp Web UI Sometimes you don’t need a coding agent; you just need a quick chat or want to test the model’s raw logic and generation speed. llama.cpp comes with a lightweight, built-in chat interface. How to use it: Open your web browser and navigate directly to http://127.0.0.1:8080/. Example Prompt: “Explain the technical advantages of integrating the Model Context Protocol (MCP) into a local LLM architecture, focusing on tool-calling capabilities.” Conclusion Building a local AI coding agent is no longer a luxury reserved for massive server farms or expensive monthly subscriptions. As we have seen, combining lightweight open-weight models like Qwen 3.5-4B with the highly optimized engine of llama.cpp allows you to transform your personal computer into a private, autonomous development powerhouse. Whether you prefer the collaborative, step-by-step nature of Qwen Code, the comprehensive architectural control of OpenCode, or just a quick sandbox chat, you now have a complete toolkit at your fingertips. You can build, refactor, and test complex applications while keeping your codebase entirely private and your wallet full. Welcome to the era of local, uncompromised AI development. Happy coding! References unsloth/Qwen3.5–4B-GGUF · Hugging Face . (2026, May 18). Huggingface.Co. https://huggingface.co/unsloth/Qwen3.5-4B-GGUF Nichonauta. (18 de abril de 2026). [TUTORIAL] El Fin de la IA de Pago . YouTube. https://www.youtube.com/watch?v=ewuJcBoKhA4 . Ditch the Cloud: Build a Free, Local AI Coding Agent with llama.cpp was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.
Score: 35🌐 MovesMay 30, 2026https://pub.towardsai.net/build-local-ai-coding-agent-llama-cpp-826a8357b510?source=rss----98111c9905da---4 - Hermes Agent Ships Tool Search for MCP: Anthropic Evals Show 49% to 74% Accuracy Gain on Opus 4
Hermes Agent Ships Tool Search for MCP: Anthropic Evals Show 49% to 74% Accuracy Gain on Opus 4 MarkTechPost
- Anthropic Cuts Unauthorized Platform List by Half After Pushback
Anthropic PBC updated its warning about secondary markets for its shares, cutting the number of unauthorized platforms by half, after the notice caused panic among investors and a sharp rebuttal from one of the companies named.
- I followed Sam Altman's tip and signed into OpenClaw with my ChatGPT account — here’s what happened
I followed Sam Altman's tip and signed into OpenClaw with my ChatGPT account — here’s what happened Tom's Guide
- Why Toronto's Rosedale neighbourhood could be ground zero for AI-powered security in Canada
Why Toronto's Rosedale neighbourhood could be ground zero for AI-powered security in Canada Toronto Star
- MapmyIndia replaces Google Maps in Amazon Now, launches AI address verification for banks: Q4FY26
Amazon Now, Geo-Verify, delayed government contracts, and a growing order book dominated MapmyIndia's Q4FY26 earnings call. The post MapmyIndia replaces Google Maps in Amazon Now, launches AI address verification for banks: Q4FY26 appeared first on MEDIANAMA .
Score: 34🌐 MovesMay 30, 2026https://www.medianama.com/2026/05/223-mapmyindia-q4fy26-amazon-now-integrates-mappls-apis/ - Philippines banks can quickly catch upin AI race - Citi
Philippine banks and financial institutions may still be taking cautious steps toward artificial intelligence, but the gap with more advanced markets can narrow quickly as firms move from experimentation to practical use cases, according to global banking giant Citi.
Score: 33🌐 MovesMay 30, 2026https://www.philstar.com/business/2026/05/31/2531668/philippines-banks-can-quickly-catch-upin-ai-race-citi - A technorealistic approach to AI literacy in Estonian schools
As European governments discuss the ethical use of AI in education and plan budgets to facilitate AI literacy among students, the Estonian AI Leap approach stands out as highly pragmatic and thoughtful.
Score: 33🌐 MovesMay 30, 2026http://www.euronews.com/next/2026/05/30/a-technorealistic-approach-to-ai-literacy-in-estonian-schools - ‘Most likely, you won’t see it on a Leica M camera’: Leica hints that generative AI tools like Gemini Omni are at odds with its photography heritage, but says they ‘make perfect sense’ for phones like the Xiaomi 17T Pro
At a roundtable attended by TechRadar, Leica shared its thoughts on the relationship between photography and generative AI.
- OMODA & JAECOO brings advanced AI-powered driverless parking solution to the UAE
Bringing its advanced AI-powered driverless parking system ‑ VPD (Valet Parking Driver) ‑ to select models in the UAE market
- Multi-Agent Workflow Runtime: How to Build Agent Teams That Don’t Turn Into AI Meetings
A useful multi-agent system looks less like a meeting and more like a runtime: roles, state, routing, gates, and telemetry. Multi-agent AI sounds powerful until the first demo becomes five chatbots politely forwarding work to each other. One agent asks another agent to research. The researcher asks a planner to clarify. The planner asks a reviewer to validate. Ten turns later, the system has spent money, produced logs, and still has no reliable output. The problem is rarely that the models are weak. The problem is that the team built a conversation instead of a runtime. In 2026, developers have better tools than a pile of prompts. Frameworks such as Google’s Agent Development Kit , LangGraph , and the OpenAI Agents SDK point in the same direction: agents need state, routing, handoffs, tools, tracing, and evaluation. The winners will not be the teams with the most agents. They will be the teams with the clearest runtime contract. This guide is for developers, founders, AI engineers, and technical leads who want to build multi-agent workflows that survive real users. We will cover when multi-agent architecture is worth it, how to design the runtime, what to log, where to put humans, and how to avoid the expensive anti-pattern I call the AI meeting. The Search Intent Behind Multi-Agent Workflows Has Changed A year ago, many developers searched for agent frameworks because they wanted to know which tool was hot. Today the more useful question is narrower: how do you make multiple agents coordinate without losing control? Recent developer discussions around agent frameworks, coding agents, Model Context Protocol servers, and AI workflow engines keep circling the same pain points: Who owns the final answer when multiple agents contribute? How do you stop agents from repeating each other? Where should shared state live? How do you debug a bad handoff? When should an agent call a tool directly instead of asking another agent? How do you keep cost and latency from exploding? Those questions are not solved by naming agents “planner,” “critic,” and “executor.” They are runtime design questions. A multi-agent system is not a group chat. It is a distributed workflow where each agent needs a contract, budget, state boundary, and exit condition. That shift creates a practical SEO opportunity too. Broad keywords such as “AI agents” and “multi-agent systems” are crowded. But long-tail searches such as “multi-agent workflow runtime,” “multi-agent orchestration patterns,” “AI agent handoff design,” and “how to debug multi-agent workflows” still have room for technical, implementation-first content. First Decide Whether You Actually Need Multiple Agents Most failed multi-agent projects start one step too late. The team asks, “Which agents should we create?” before asking, “Why is one agent not enough?” A single agent with a strong prompt, good tools, retrieval, structured output, and evals is often simpler, cheaper, and easier to debug. Add more agents only when the work has real separation of concerns. Use multiple agents when the workflow has different modes of judgment Multi-agent design starts to make sense when the same task requires meaningfully different reasoning styles. For example, a procurement assistant may need one agent to parse policy, one to compare vendor risk, and one to draft a purchase recommendation. A software maintenance workflow may need one agent to inspect an issue, one to make a code change, and one to review the diff against architectural rules. The key is not the job title. The key is whether each role has different context, tools, evaluation criteria, or authority. Avoid multiple agents when you only want better answers If your only reason is “more agents might be smarter,” pause. You may need a better task brief, retrieval setup, model choice, or evaluation loop. Adding a critic agent to a vague workflow often creates confident disagreement instead of quality. Use this quick filter: If agents need different tool permissions, separation may help. If agents need different private context, separation may help. If agents produce different artifacts, separation may help. If agents only restate the same instruction, keep one agent. If agents cannot be evaluated independently, keep the design simpler. The Runtime Contract: The Missing Layer A runtime contract defines how agents enter the workflow, what they receive, what they may do, what they must return, and when the system stops. Without that contract, your agents negotiate the workflow while doing the work. That is where cost, latency, and ambiguity creep in. A practical multi-agent runtime contract has seven parts. 1. Task envelope Every agent should receive a task envelope, not a loose chat history. The envelope should include the goal, constraints, available tools, required output schema, budget, deadline, and any prior state that the agent is allowed to see. { "task_id": "support-refund-0421", "role": "policy_checker", "goal": "Determine whether this refund request is allowed", "input_refs": ["ticket", "order", "refund_policy"], "allowed_tools": ["policy_search", "order_lookup"], "blocked_tools": ["send_email", "issue_refund"], "max_tool_calls": 5, "required_output": "policy_decision_v1" } This structure keeps the agent focused. It also gives your logs something meaningful to index later. 2. Role boundary A role boundary says what the agent is responsible for and what it must not decide. For example, a policy checker can classify a refund request but cannot issue the refund. A code reviewer can flag architectural risk but cannot rewrite the diff unless the workflow explicitly routes back to an implementation agent. Role boundaries reduce drift. They also make it easier to replace one agent without rewriting the whole system. 3. Shared state model Shared state is where many multi-agent systems break. If every agent sees the entire conversation, context grows fast and important facts get buried. If agents see too little, they repeat work or contradict each other. Use a state model with typed fields. For example: user request, task plan, retrieved evidence, tool outputs, decisions, rejected options, open questions, final artifact, and audit notes. Agents should read only the fields they need and write only the fields they own. 4. Routing rules Routing decides which agent runs next. This can be deterministic, model-driven, or hybrid. Deterministic routing is easier to test. Model-driven routing is useful when input variety is high. Hybrid routing is often best: use rules for known paths, and use an LLM router only for classification or ambiguity. 5. Handoff schema A handoff is not “please take over.” A handoff should be a structured object: what was done, what evidence was used, what assumptions remain, what the next agent must decide, and what failure mode to watch for. 6. Stop conditions Every runtime needs a clean stop condition. Stop when the required artifact passes validation, when the budget is exhausted, when an approval is needed, when the system detects repeated low-confidence loops, or when the user needs to clarify intent. 7. Trace and replay The runtime should make agent coordination observable: task queue, router, workers, shared state, tools, evals, retries, and human approval. If you cannot replay a bad run, you cannot improve the system reliably. Trace the task envelope, model calls, tool calls, state changes, routing decisions, validation results, and human approvals. OpenTelemetry-style tracing and framework-level tracing are no longer nice-to-have for agent workflows. They are how you debug the runtime instead of blaming the model. A Practical Multi-Agent Runtime Architecture You do not need a huge platform to start. A strong architecture can be simple. Think in layers: The interface layer receives the user request and turns it into a task. The router decides whether this is a single-agent or multi-agent path. The state store holds structured workflow state. Worker agents perform bounded jobs with scoped tools. The validation layer checks outputs with schemas, tests, rules, and evals. The policy layer enforces permissions, secrets, network access, and approval gates. The observability layer records traces, costs, latency, and outcomes. That architecture works whether you use ADK, LangGraph, the OpenAI Agents SDK, Semantic Kernel, a custom queue, or a traditional workflow engine. The framework matters, but the runtime contract matters more. Example: Customer Support Resolution Imagine a support assistant that handles refund requests. A naive multi-agent version might let a triage agent, policy agent, order agent, and reply agent talk until they agree. A runtime-first version is cleaner. Triage classifies the request and extracts required fields. Policy checker reads the refund policy and returns an allowed, denied, or needs-review decision. Order checker verifies dates, payment status, item category, and prior refund history. Decision composer combines evidence into a structured recommendation. Human approval is required if the value is high, the policy is ambiguous, or the customer is escalated. Reply drafter writes the customer message only after the decision is approved. Notice the authority boundary. The reply agent does not decide policy. The policy agent does not email the customer. The order checker does not improvise exceptions. Each agent has a job, and the runtime owns the flow. Example: AI Code Maintenance A coding workflow can follow the same pattern: Issue analyst summarizes the bug, constraints, and likely files. Implementation agent edits code within a sandbox and records assumptions. Test agent runs focused tests and reports failures. Architecture reviewer checks whether the fix violates design boundaries. Human reviewer receives a compact diff summary, test results, and unresolved risks. This is different from asking three agents to “solve the bug.” The runtime forces the work into observable stages. How to Choose Between Graphs, Supervisors, and Queues Most multi-agent systems use one of three patterns: graph workflow, supervisor-worker, or queue-based orchestration. Graph workflow A graph workflow is best when the steps are known. Each node performs a task. Edges define where control goes next. Conditional edges handle validation failures, retries, and approval gates. This is a natural fit for LangGraph-style designs and many ADK workflows. Use a graph when you can draw the process before building it. Supervisor-worker A supervisor-worker design gives one agent the authority to delegate to specialized workers. This is useful when the task shape varies, but it is risky if the supervisor becomes a vague manager. The supervisor should route, decompose, and synthesize. It should not endlessly debate with workers. Use a supervisor when inputs are unpredictable but roles are stable. Queue-based orchestration A queue-based design treats agent tasks like jobs. Agents pick up typed work items, write outputs, and emit events. This pattern is useful for long-running, asynchronous, or high-volume systems. It also works well when some tasks are better handled by non-LLM services. Use queues when the workflow needs scale, retries, isolation, or human-in-the-loop pauses. The AI Meeting Anti-Pattern When agents coordinate through loose messages, work loops. When they coordinate through contracts, state, and validators, work moves. The easiest way to spot a weak multi-agent design is to inspect the transcript. If the agents spend most of the run restating goals, requesting clarification from each other, debating vague quality standards, or asking another agent to check something they could check directly, you have an AI meeting. AI meetings usually have five causes: No owner for the final artifact. No typed handoff between agents. No shared state outside the chat transcript. No budget on agent turns or tool calls. No validator that can end the loop. The fix is not to add a stronger manager prompt. The fix is to move coordination out of the conversation and into the runtime. Evaluation: Test the Workflow, Not Just the Agent Agent evaluation often starts at the wrong level. Teams test whether one agent gives a good answer, then assume the whole workflow is reliable. Multi-agent systems need workflow-level evals. Track these metrics: Task success rate: did the workflow produce the required artifact? Handoff quality: did the next agent receive enough structured context? Loop rate: how often did the workflow revisit the same state? Tool-call efficiency: how many tool calls were needed per successful task? Human escalation precision: did approvals trigger for the right cases? Cost per resolved task: not cost per token, but cost per useful outcome. Latency by stage: which agent or tool dominates time-to-answer? Use golden tasks for predictable workflows. Use adversarial tasks for edge cases. Use replay for real failures. Keep eval cases versioned with your prompts and schemas so changes are reviewable. Security and Governance Still Matter Multi-agent workflows multiply authority. If one agent can read secrets, another can write code, and a third can send messages, the runtime must enforce boundaries. Do not rely on the agents to remember the security policy. Use scoped tools, least-privilege credentials, sandboxed execution, approval gates for irreversible actions, and audit logs for sensitive decisions. Anthropic has written publicly about containment as an engineering problem in its Claude containment work , and the lesson applies broadly: agent capability needs system-level controls, not just instruction-level controls. A helpful default: agents may propose high-risk actions, but the runtime performs them only after policy checks and explicit approval. A Minimal Implementation Pattern Here is a small framework-agnostic sketch. It is not a full production system, but it shows the separation of responsibilities. async function runWorkflow(input) { const state = await createState({ request: input, evidence: [], decisions: [], risks: [], artifact: null }); const route = await routerAgent({ goal: "Choose the workflow path", state: pick(state, ["request"]), outputSchema: "route_decision_v1" }); for (const step of route.steps) { const envelope = buildTaskEnvelope(step, state); const result = await runBoundedAgent(envelope); await validateSchema(result, step.outputSchema); await appendTrace(step.name, envelope, result); await mergeState(state, step.writes, result); const gate = await validateWorkflowState(state); if (gate.status === "needs_human") return pauseForApproval(state, gate); if (gate.status === "failed") return failWorkflow(state, gate); if (gate.status === "complete") return state.artifact; } return finalize(state); } The important idea is not the syntax. The runtime builds envelopes, enforces schemas, merges state, validates progress, and decides whether the workflow continues. Agents do bounded work inside that structure. What to Build First If you are starting from scratch, do not begin with six agents. Build the smallest runtime that can support one reliable multi-step workflow. Choose one workflow with clear success criteria. Write the final artifact schema before writing agent prompts. Split agents only where context, tools, or authority differ. Create typed task envelopes and handoff schemas. Add tracing before you run pilot users. Add evals for happy path, edge cases, and known failure modes. Set budgets for turns, tool calls, tokens, and runtime duration. Add human approval gates for irreversible, expensive, or sensitive actions. Once that one workflow is boringly reliable, add complexity. Reliability scales from contracts, not from more personalities. Common Mistakes to Avoid Mistake 1: Letting agents write unstructured notes to each other Natural language handoffs are fine for humans, but brittle for machines. Use structured handoffs with explicit assumptions, evidence, confidence, and next action. Mistake 2: Giving every agent every tool Tool access should follow role boundaries. If a reviewer can deploy, a failed review can become a production incident. Mistake 3: Hiding state in the transcript Chat history is not a database. Store workflow state separately and pass only relevant slices to each agent. Mistake 4: Evaluating only final answers Final-answer evals miss bad handoffs, wasted tool calls, risky approvals, and fragile loops. Evaluate intermediate steps too. Mistake 5: Treating the framework as the architecture A framework gives you primitives. Architecture is your contract: state, routing, permissions, validation, observability, and escalation. Final Takeaway Multi-agent workflows are becoming a normal part of AI application development, but the useful version is not a cast of clever agents talking in circles. It is a runtime with typed work, clear authority, shared state, validation, tracing, and human control where it matters. Before you add another agent, ask a harder question: what contract will make this agent useful, measurable, and safe? That question will save more tokens than any prompt trick. FAQ What is a multi-agent workflow runtime? A multi-agent workflow runtime is the orchestration layer that manages agent tasks, routing, shared state, tool permissions, validation, tracing, retries, and approval gates. It turns a set of agents into a controlled workflow instead of a loose conversation. When should developers use multiple AI agents? Use multiple AI agents when the task has distinct roles with different context, tools, permissions, outputs, or evaluation criteria. If all agents would see the same context and make the same decision, a single well-designed agent is usually better. What is the biggest risk in multi-agent systems? The biggest operational risk is uncontrolled coordination: agents repeating work, passing vague messages, escalating costs, and making decisions without clear ownership. Security risk also increases when multiple agents have broad tool access. How do you debug a multi-agent workflow? Trace every task envelope, model call, tool call, state write, handoff, routing decision, validation result, and approval. Then replay failed runs against fixed prompts, schemas, and tool outputs so you can isolate whether the issue came from routing, context, tools, or a specific agent. Is LangGraph, ADK, or the OpenAI Agents SDK better for multi-agent workflows? The best choice depends on your stack and workflow shape. LangGraph is strong for graph-based stateful workflows, ADK is designed for agent development patterns and deployment paths, and the OpenAI Agents SDK focuses on agents, handoffs, guardrails, and tracing. The architecture matters more than the logo: choose the tool that best supports your runtime contract. How do you keep multi-agent workflows from getting expensive? Set budgets on turns, tokens, tool calls, and runtime duration. Use smaller models for routing or extraction, cache stable tool results, pass only relevant state slices, stop loops early, and measure cost per successful task rather than cost per model call. Sources and Further Reading Google Agent Development Kit documentation LangGraph documentation OpenAI Agents SDK documentation Anthropic Engineering: How we contain Claude GitHub Copilot documentation Multi-Agent Workflow Runtime: How to Build Agent Teams That Don’t Turn Into AI Meetings was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.
- Classrooms lean into analog learning in the AI era
Parents, educators and lawmakers are pushing schools to curb classroom screen use amid concerns about student attention, behavior and learning. Why it matters: School cellphone bans are expanding into broader efforts to limit screen use and establish guardrails around AI in education. The big picture: At least 16 states — both red and blue — have introduced bills to limit classroom technology. Case in point: Schools Beyond Screens formed with fewer than a dozen parents in Los Angeles Unified School District last year, but the nonprofit has grown to include thousands of parents and educators nationwide, SBS policy director Kate Brody tells Axios. SBS worked with the school board in the nation's second-largest district to pass a resolution limiting classroom screen time and eliminating school-issued devices for students in first grade and younger. Similar resolutions have emerged in New York City and Washington, D.C . Zoom in: McPherson Middle School principal Inge Esping told Axios that the suspension rate at her Kansas school fell 70% after cellphones were banned in 2022. Students also started speaking more with one another and with teachers. Suspensions continued to decline as she implemented more safeguards, but Esping says "we weren't quite making the progress that we wished, behaviorally and culture-wise." The school's discipline data showed behavioral issues involving Chromebooks, including gaming in class, so the school secured a grant allowing students to return the devices after it shifted to storing them in carts, Esping says. The school doesn't cap Chromebook use. Instead, teachers are encouraged to use devices only when they enhance learning, Esping says. The intrigue: Epsing says students tell her binders and pencils feel lighter than Chromebooks and help them focus better. Zoom out: Though challenges remain, Body says she's optimistic that analog learning is coming to more classrooms. It helps that some teachers appear on board with the shift. Context: The American Federation of Teachers, the second-largest teachers' union in the U.S., released a 10-point plan to introduce AI and screen-time guardrails in classrooms. The plan would limit AI use and ban screens for students in prekindergarten through second grade "unless there is a compelling reason," such as supporting students with special needs. The teachers' union wants to get "the balance right to harness the benefits of technology while mitigating the harms," Randi Weingarten, AFT President, said during Wednesday's announcement. Screenshot: American Federation of Teachers The other side: "When used with intention and balance," ed tech "can support differentiation, accessibility, enrichment, and workforce preparation in the K-12 classroom," says a 2025 report from the Consortium for School Networking (CoSN), a nonprofit for ed tech leaders that has corporate partners including Amazon, Google and Microsoft. Beth Holland of K-12 education nonprofit FullScale said during a policy briefing for state leaders posted to CoSN's YouTube channel last month that ed tech is designed to address "that fundamental relationship between teachers and students and content." Esping agrees with Holland's assessment that "not all screen time is the same" and that there are differences between ed tech and Big Tech. But she says educators need to be mindful of when tech can be bad and how it enhances learning. The bottom line: "We also need to recognize that tech alone is not an enhancement of learning," Esping says. Teachers remain "the most important" factor in what students "will or won't learn," she adds.
- This creepy blob robot will keep going even if you break its legs
While Argus looks like a sea urchin, its designers took cues from physics, not biology. The post This creepy blob robot will keep going even if you break its legs appeared first on Popular Science .
- Tired of AI Overviews? I found 9 Google Search alternatives that showed me links again
Fed up with trying to avoid AI when I search, I went looking for a Google replacement. I found plenty, but not many made the cut.
- AI Tool of the Week: This OpenAI tool tells you if an image is AI-generated
A free OpenAI tool now lets users verify whether an image was generated by AI, using hidden watermarks and metadata that survive screenshots, edits, and format changes.
- Zuckerberg’s yacht, Meta’s layoffs, a robot pizza flameout, and a reality check on AI expenses
On this week's GeekWire Podcast: Mark Zuckerberg's superyacht arrives in Seattle the same day Meta discloses nearly 1,400 local layoffs, robot pizza startup Picnic shuts down and sells its assets to a mystery buyer, and companies grapple with the rising cost of corporate AI. Read More
- ‘What a joke’: Github Copilot’s new token-based billing spurs consternation among devs
The golden age of Microsoft's Github Copilot appears to be at an end.
- Hollywood’s A.I. Flirtation & A Fox-ESPN Proxy War
Hollywood’s A.I. Flirtation & A Fox-ESPN Proxy War Puck
Score: 30🌐 MovesMay 30, 2026https://puck.news/newsletter_content/hollywoods-ai-flirtation-a-fox-espn-proxy-war/ - Second phase of AI Passport project in pipeline
The Ministry of Digital Economy and Society (DES) plans to move ahead with development of the second phase of the TH-AI Passport project at a cost of 900 million baht, despite ongoing criticism of the first phase.
Score: 30🌐 MovesMay 30, 2026https://www.bangkokpost.com/business/general/3262915/second-phase-of-ai-passport-project-in-pipeline - Docling + VectorLess + Gemma 3.5 Flash To Get Higher Accuracy
If you don’t have a Medium subscription, use this link to read the full article: Link Continue reading on Towards AI »
- Sacred Mysteries: Babel or Jerusalem – our urgent choice on AI
Sacred Mysteries: Babel or Jerusalem – our urgent choice on AI The Telegraph
Score: 30🌐 MovesMay 30, 2026https://www.telegraph.co.uk/news/2026/05/30/sacred-mysteries-babel-jerusalem-urgent-choice-ai-pope/ - Robots vandalized while evaluating Hollister pedestrian infrastructure
Robots vandalized while evaluating Hollister pedestrian infrastructure East Bay Times
- Accenture Workshop: 'Gemini is my (new) business partner!'
Accenture Workshop: 'Gemini is my (new) business partner!' University of Oxford
Score: 29🌐 MovesMay 30, 2026https://enspire.ox.ac.uk/event/accenture-workshop-gemini-is-my-new-business-partner - Ministry's AI project faces heat
The Democrat Party has called on the Ministry of Digital Economy and Society (DES) to review its proposed TH-AI Passport project, citing concerns over transparency, procurement practices and value for money.
Score: 28🌐 MovesMay 30, 2026https://www.bangkokpost.com/thailand/politics/3263438/ministrys-ai-project-faces-heat - Drone strike takes Putin’s shadow war to dangerous new level
Drone strike takes Putin’s shadow war to dangerous new level The Telegraph
Score: 28🌐 MovesMay 30, 2026https://www.telegraph.co.uk/world-news/2026/05/30/drone-strike-putin-shadow-war-europe-dangerous-new-level/ - Meta-Cognitive Regulation Might Be the Most Important AI Skill Nobody Is Talking About
As AI gets smarter, the real differentiator may be how well humans regulate their own thinking. The post Meta-Cognitive Regulation Might Be the Most Important AI Skill Nobody Is Talking About appeared first on Towards Data Science .
- How should bosses talk about AI?
Employees are being asked to embrace a technology that causes fear
Score: 28🌐 MovesMay 30, 2026https://www.livemint.com/ai/how-should-bosses-talk-about-ai-11780135877278.html - REXTRIX LAUNCHES THE WORLD’S FIRST AI-NATIVE INTERACTIVE ENTERTAINMENT PLATFORM
REXTRIX LAUNCHES THE WORLD’S FIRST AI-NATIVE INTERACTIVE ENTERTAINMENT PLATFORM azcentral.com and The Arizona Republic
- Multi-Agent Fan-Out: When Parallelism Bites Back
Scatter-gather is one of the most seductive patterns in distributed systems: split a hard problem into N pieces, run them in parallel… Continue reading on Towards AI »
- NotebookLM is quickly becoming the podcast app I didn’t know I needed
AI podcasts may sound like a grim idea on paper, but Google's app is enabling a version I'm actually excited about.
Score: 26🌐 MovesMay 30, 2026https://www.androidauthority.com/notebooklm-podcast-app-didnt-know-i-needed-3669559/ - An AI CEO explains how much he spent on Codex last month — and why he's still 'very nice' when prompting the tech
An AI CEO explains how much he spent on Codex last month — and why he's still 'very nice' when prompting the tech Business Insider
- AI Summit 2026
A conference for AI research and applications
- VidspotAI Launches Long-Form AI Video Production Service for Creators and Businesses
VidspotAI Launches Long-Form AI Video Production Service for Creators and Businesses azcentral.com and The Arizona Republic
- Fobi AI Provides Corporate Update Regarding FFCTO Revocation Process and Q3 Interim Filings
Fobi AI Provides Corporate Update Regarding FFCTO Revocation Process and Q3 Interim Filings Toronto Star
- Cancel your Claude sub and get lifetime access to ChatGPT, Gemini, Claude, and more for $60
ChatPlayground gives you lifetime access to 20+ AI tools for $70.
Score: 23🌐 MovesMay 30, 2026https://mashable.com/tech/may-30-chatplayground-ai-unlimited-plan-lifetime-subscriptions - SEI Webcast: Rethinking and Maturing AI Adoption
SEI Webcast: Rethinking and Maturing AI Adoption Carnegie Mellon University
Score: 22🌐 MovesMay 30, 2026https://events.cmu.edu/event/31615-sei-webcast-rethinking-and-maturing-ai-adoption - Flights temporarily suspended at Munich Airport after reported drone sighting
Munich Airport closed twice within 24 hours in October following suspected drone sightings.
Score: 22🌐 MovesMay 30, 2026http://www.euronews.com/my-europe/2026/05/30/flights-suspended-at-munich-airport-after-reported-drone-sighting - These 3 AI Shortcuts Turn Ordinary Founders Into 10x Operators
These 3 AI Shortcuts Turn Ordinary Founders Into 10x Operators entrepreneur.com
Score: 22🌐 MovesMay 30, 2026https://www.entrepreneur.com/science-technology/these-3-ai-shortcuts-turn-ordinary-founders-into-10x/503484 - AI Filmmaker Compares His Tech to Something That Gets Worse the More You Think About It
Something about sex and babies. The post AI Filmmaker Compares His Tech to Something That Gets Worse the More You Think About It appeared first on Futurism .
Score: 20🌐 MovesMay 30, 2026https://futurism.com/artificial-intelligence/ai-filmmaker-normal-analogy-of-all-time - I asked ChatGPT if a ₹1 lakh salary can buy a ₹1 crore home. The AI verdict was a reality check
With a monthly take-home of ₹1 lakh and taking expenses and investments into account, buying a ₹1 crore flat may be financially imprudent. Here's what ChatGPT advised me to do instead.
- 🧠 Community Wisdom: Catching people using AI during an interview, org design when everything lives in one person’s head, when to rename your product, from nurse to health-tech PM, and more
Community Wisdom 187