Daily AI briefing

AI Briefing: SoftBank Pulls Back as Agents Scale Up

SoftBank trims OpenAI-backed loan targets as Isomorphic Labs seeks $2B, and OpenAI's new Chrome extension brings agents to Salesforce and Gmail.

  • Briefing date: 2026-05-09
  • Editorial AI analysis
  • The AI Wrap

Stories covered on 2026-05-09

  1. AI Will Not Replace Software Developers — And I’ll Say It Loud

    Towards AI · 2026-05-09

    I use AI tools every single day. And I’m here to tell you: the “AI replaces developers” narrative is overblown — by at least a decade.

  2. NVIDIA AI Releases Star Elastic: One Checkpoint that Contains 30B, 23B, and 12B Reasoning Models with Zero-Shot Slicing

    MarkTechPost · 2026-05-09

    NVIDIA researchers have introduced Star Elastic, a post-training method that embeds multiple nested reasoning models — at 30B, 23B, and 12B parameter scales

  3. So you’ve heard these AI terms and nodded along; let’s fix that

    TechCrunch · 2026-05-09

    The rise of AI has brought an avalanche of new terms and slang. Here is a glossary with definitions of some of the most important words and phrases you might encounter.

  4. Cloudflare Hired 1,111 People to Prove AI Wouldn’t Take Jobs. Then It Fired 1,100 of Them.

    Towards AI · 2026-05-09

    For tech workers trying to understand what’s actually happening to their industry: the Cloudflare numbers, the pattern behind them, and…

  5. How to Run Claude Code Agents in Parallel

    Towards AI · 2026-05-09

    Learn how to apply coding agents in parallel to work more efficientlyContinue reading on Towards AI »

  6. "OncoAgent: A Dual-Tier Multi-Agent Framework for Privacy-Preserving Oncology Clinical Decision Support"

    Hugging Face Blog · 2026-05-09

    OncoAgent introduces a novel multi-agent framework for oncology clinical decision support that prioritizes patient privacy.

  7. Google Just Installed a 4GB AI on Your Computer. The Privacy Excuse Is a Lie.

    Towards AI · 2026-05-09

    If you use Google Chrome on a device with a dedicated GPU, Google installed a 4GB AI model without asking. Here is what it does, what it…

  8. AI Memory Down From 42GB to 7GB. Here’s What Google’s TurboQuant Actually Did.

    Towards AI · 2026-05-09

    Google researchers have developed a new technique, TurboQuant, that significantly reduces the memory footprint of large language models

  9. AI Is the New Layoff Alibi

    Towards AI · 2026-05-09

    AI's perceived efficiency is increasingly cited as a justification for workforce reductions, even when the technology's direct impact on productivity remains unproven.

  10. Running MedGemma on Ollama: Multimodal Medical AI in Action

    Towards AI · 2026-05-09

    From medical conversations to image interpretation — all running locally with OllamaContinue reading on Towards AI »

  11. Google developers significantly misstate carbon emissions of proposed UK datacentres

    The Guardian AI · 2026-05-09

    Emissions understated by factor of five in Essex plans for tech giant, while Greystoke’s Lincolnshire plans show similar error Developers working for Google have significantly

  12. ChatGPT Just Started Reading Your Email Without Asking. Here’s How to Control It.

    Towards AI · 2026-05-09

    OpenAI’s new default model accesses your Gmail on its own judgment. The transparency panel they built to explain that doesn’t show…

  13. The Must-Know Topics for an LLM Engineer

    Towards Data Science · 2026-05-09

    From tokenisation to evaluation : how modern language models actually work in practice

  14. What Is the Best Local LLM for Coding in 2026?

    Towards AI · 2026-05-09

    A practical guide to choosing local coding models by hardware tier, workflow, latency, and privacy, not just benchmark screenshots.

  15. Semantic Caching for Enterprise AI Agents: Cut Costs, Kill Latency

    Towards AI · 2026-05-09

    A new technique called semantic caching has been developed to store and retrieve frequently used data for enterprise AI agents

  16. Nvidia has already committed $40B to equity AI deals this year

    TechCrunch · 2026-05-09

    Nvidia continues to be a big investor in the AI ecosystem.

  17. Stop Using AI as a Search Engine. Use It as a Marketing System Instead.

    Towards AI · 2026-05-09

    The prevailing advice suggests shifting from treating AI chatbots like Google to leveraging them for targeted marketing campaigns.

  18. How NVIDIA Cut DeepSeek Sparse Attention’s Top-K Time

    Towards AI · 2026-05-09

    Half by Exploiting a Quirk of Autoregressive DecodingContinue reading on Towards AI »

  19. ECB’s Escrivá Says AI Risks Prompt Finance Infrastructure Review

    Bloomberg · 2026-05-09

    Central banks must review the resilience of financial infrastructure given the rise of artificial intelligence

  20. AI Writes the Code. But Nobody Is Watching the Architecture.

    Towards AI · 2026-05-09

    Why the most important skill in software development right now is not prompting — it is knowing when the AI is quietly making a mess.

  21. RAG Is Blind to Time — I Built a Temporal Layer to Fix It in Production

    Towards Data Science · 2026-05-09

    Three weeks into testing, a learner told me my AI tutor gave her the wrong answer. Not obviously wrong — just outdated enough to mislead.

  22. I Let Claude Dream for 4 Hours — Today’s Agent Just Killed Yesterday’s by 5.4× on 18 Repeat Tasks

    Towards AI · 2026-05-09

    A research-preview feature called “dreaming” replays your agent’s past sessions overnight, prunes the contradictions, and ships a curated…

  23. Cloud Embeddings vs. Local Sovereign Memory: AI Agent Memory Layer Compared (2026)

    Towards AI · 2026-05-09

    A recent analysis contrasts the architectural approaches of cloud-based embedding stores versus local, sovereign memory solutions for AI agents

  24. Building Multi-Agent AI Systems for Banking: Advanced Workflows and Agent Coordination with CrewAI…

    Towards AI · 2026-05-09

    A new framework, CrewAI, has emerged to facilitate the development of multi-agent AI systems designed for complex banking operations.

  25. The 7 Skills You Need to Build AI Agents That Actually Work in Production

    Towards AI · 2026-05-09

    A recent article outlines seven key skill areas—including data engineering, MLOps, and prompt engineering—essential for successfully deploying AI agents in real-world applications.

  26. How Smart Organizations Will Use AI: Jevons Paradox and the Future of the Workforce

    Towards AI · 2026-05-09

    The summary suggests that increased AI efficiency may not lead to reduced overall AI adoption, referencing Jevons Paradox.

  27. Claude's /ultrareview Just Embarrassed My 4-Person Review Team — I Burned $241 on 18 PRs to Prove…

    Towards AI · 2026-05-09

    The free tier died on May 5, 2026. Three days later I had a $241 invoice, 18 closed pull requests, and a Slack thread where my 4-person…

  28. Foundations of CCA-F Exam Part 4: Engineering the Long-Running Agent Harness: From Amnesia to…

    Towards AI · 2026-05-09

    Turning Agent Amnesia into Persistent Autonomy: A Dual-Agent Harness Engineering Blueprint for the Claude Certified Architect ExamContinue reading on Towards AI »

  29. Machine Learning System design — Data Labeling Pipelines, With One Content Moderation System…

    Towards AI · 2026-05-09

    This installment delves into the practicalities of data labeling pipelines within machine learning system design, specifically highlighting a content moderation system.

  30. AI in Critical Infrastructure: What Good Actually Looks Like

    Towards AI · 2026-05-09

    Hospitals are deploying AI for operational efficiency, such as optimizing patient flow and predicting equipment maintenance needs.