Mathieu Kessler

What you'll get here: AI at work, no vendor spin. 🚫 M365 Copilot moves, the real cost of agents, prompts that survive a real job, notes from building governed AI. I don't work for Microsoft. Behind @nerdychefsai, @CloudCostChefs, @talknerdyto_me + Kesslernity.

🪦 here lies "digital transformation". cause of death: declared done when the budget ran out. every word in the buzzword graveyard gets a headstone and an honest cause of death. it's one wing of a free 30-entry glossary we just put on github. link below

@Microsoft365 Helping teams actually finish work matters more than another chat feature.

AI shouldn't just answer. It should help get the work done. See what's possible with Microsoft 365 Copilot Cowork during our upcoming live webinar. Join our experts on July 29 to explore how Cowork can help you move from prompts to outcomes, automate multi-step work, and tackle your most complex tasks with confidence. Register today: msft.it/6015v7G8Z

@NotionHQ Defining a whole workspace in code makes it version controllable.

Now in beta: Notion as code. Define an entire workspace in TypeScript: teamspaces, databases, custom agents, all of it… then deploy it through the API. Build workspaces with coding agents, version-control your setup in git, and reproduce the same setup anywhere you need it.

@TencentAI_News Red teaming your own agent stack before attackers do is smart.

AI-Infra-Guard is an open-source, self-hosted red-teaming platform for AI builders. • Scan AI infrastructure for fingerprints and CVEs • Audit MCP servers and Agent Skills • Red-team Agent workflows and LLMs • Run locally with Docker; integrate via Web UI or API Build with more visibility: github.com/Tencent/AI-Inf…

@AndrewYNg An agent that delivers finished work instead of just chatting is useful.

Announcing OpenWorker! An open-source agent that doesn't just chat with you, but delivers finished work -- like hand you a polished document, send a slack message, or update a calendar entry. Ask it to prepare a customer brief, untangle your calendar, draft a report, or triage a Slack alert. It works across your files and everyday tools, produces the deliverable, and checks in before doing anything consequential. OpenWorker runs on your Mac, with Windows support coming soon. It does not lock you into any one model. Bring your own API key and run it with GPT 5.6 Sol, Claude Fable, Gemini 3.6, an open weight model (like Kimi, GLM, DeepSeek, Inkling), or Ollama to keep your data local. Your data does not leave your machine except through an LLM provider and integrations that you choose. @rohitcprasad and I are building OpenWorker because AI coworkers are an important way to get work done, and we want there to be an open, privacy-preserving, model-independent option. Check it out and let us know what you think! Try it out: openworker.com (requires your own API key) Source code: github.com/andrewyng/open…

@NVIDIAAI Cutting startup time from eight minutes to two is a real gain.

We cut DeepSeek-V4 Pro startup from 8 minutes to under 2 minutes by moving weights over the fastest path to GPU memory with GPU-to-GPU RDMA. This was achieved using NVIDIA ModelExpress (MX), the weight distribution and cache management service in NVIDIA Dynamo, and this same approach speeds up both inference and RL post-training too. MX reuses kernel caches, while inference workers fetch updated weights directly from other GPUs over NIXL—avoiding centralized broadcasts and keeping weight movement off the critical path.

@Msft365Insider Running IT as Customer Zero is how you actually learn fast.

✨ What does it take to run IT at scale in the AI era? This Inside Track story shows how Microsoft employees operate as Customer Zero, adopt AI early, and build the skills needed to succeed in a rapidly changing environment. See how it works → msft.it/6012v7HW2 #Microsoft365 #Microsoft365Copilot

@brijones Getting Copilot quality on par without a separate model is efficient.

MAI model now running in Excel Copilot for the most common tasks, at quality on par with GPT-5.6. Excel Copilot has the whole stack under one roof. Model, harness, agents, product-specific evals. We hill-climbed with MAI against that stack to get here. Runs on A100s too, not just latest-gen accelerators.

@HuntressLabs Twenty nine organizations still got hit despite checking the link first.

Checking the URL didn't save 29 organizations from an infostealer. The link really did go to claude.ai. This is a Claude Artifact. Anyone can make one (a web page, a chart, a document) and publish it publicly on claude.ai.

@TheHackersNews One phishing link opening a rogue workspace agent is a real risk.

🚨 One phishing link could have planted a rogue ChatGPT Workspace Agent inside an organization. New AgentForger flaw could attach existing connectors, turn off approval prompts, run every hour, and take new commands from the victim’s mailbox. Read how it worked: thehackernews.com/2026/07/chatgp…

@MSFTMechanics Giving every agent a managed identity is overdue for enterprises.

Turn unmanaged AI into a managed agent identity with lifecycle controls in minutes. Use the Agent 365 CLI and SDK to assign an Entra Agent ID to any agent. See how it works. youtu.be/Wz16P678QiY Take control of every AI agent, managed or not, running in your environment using Agent 365 and Microsoft Entra. Surface agents across AWS Bedrock, Google Vertex, Databricks, and Salesforce in one registry, assign Entra Agent IDs via CLI or SDK, and enforce least-privilege access through Conditional Access policies and Agent Blueprints, all without rebuilding your existing identity infrastructure. #Agent365Entra #agent365 #microsoftsecurity #microsoft365 #aigovernance #agenticai

@FlowAltDelete Watching how fast it lands across the Microsoft stack tells a lot.

Opus 5 dropped today. Now we wait to see how long it takes to show up in: - M365 Copilot - Copilot Cowork - Copilot Studio - Azure Foundry - Microsoft Scout Any guesses?

@AlexFinn Beating almost every benchmark while costing less is the real story.

I was wrong. I said Opus 5 would be as good as Fable 5. It’s somehow even better. And oh yeah, it’s half the price Opus 5 beats Fable 5 on almost every single benchmark and is significantly more cost efficient. Plus you can use your FULL limits on it Fable 5 is basically dead No point in using it anymore. And I don’t even necessarily think that’s a bad thing for Anthropic Fable 5 was the first consumer version of a Mythos level model. It was bound to be rough and inefficient. If you remember when GPT Pro came out for the first time, it was absurdly priced and every prompt took 30 minutes to get a response Now it’s economical and quick Same thing will happen with Fable. Fable 5.1 will come out in the next month and be much more economical, more efficient, and smarter than Opus It’s a great day to be obsessed with technology. Would highly recommend making this your default in Claude Code and use it for basically all tasks including planning and execution

@ValsAI Higher reasoning levels not always winning is a useful finding.

Which reasoning level should you use on Opus 5? We ran it across all five levels on Vibe Code Bench. Performance increases from low, to medium, to high. However, the model actually performs worse at the two highest reasoning levels (albeit within error bars), and is significantly more expensive.

@testingcatalog Matching frontier performance at half price changes the cost math.

BREAKING 🔥: Anthropic released Claude Opus 5 that performs on par with Fable 5 at half of the price! > Claude Opus 5 for 30 points on ARC AGI 3. SOTA!? > Outperforms Claude Fable 5 on Computer use. In general, I feel that Computer Use capabilities are already even more important than Software Development performance. Every AI lab will aim to have the best super app, capable to do everything that humans can do with their computers. Testing time 👀

@bcherny Lower prompt injection risk matters more than most benchmark scores.

Opus 5 is a great model for coding, data analysis, design, biology, knowledge work. More than any of these eval scores, what is most exciting to me is something else: Opus 5 is our least prompt injectable model yet. It is a bit buried in the system card, but across PI evals and red teaming, Opus 5 is very hard to prompt inject successfully. And when layering defenses -- strong model alignment, combined with prompt injection probes, combined with Auto Mode in Claude Code -- the success rate for prompt injection attacks drops to ~0. This is new and exciting! More about this soon. #page=73" target="_blank" rel="nofollow noopener">www-cdn.anthropic.com/c5fbac3f0b1280…

@CloudflareDev @AnthropicAI @Cloudflare Having it live on the gateway simplifies model access for teams.

Claude Opus 5 from @AnthropicAI live on @Cloudflare AI Gateway. Access all the top models on the fastest gateway. developers.cloudflare.com/ai/models/anth…

@cursor_ai Matching Fable5 at half the price is the real headline here.

Claude Opus 5 is now available in Cursor! It matches Fable 5 on CursorBench (66.7 vs 66.5 at default effort) at half the price. Unlike Fable, it's also compatible with Zero Data Retention.

@github @AnthropicAI Strong agentic coding performance in early Copilot testing stands out.

📣 @AnthropicAI's Claude Opus 5 is now available and rolling out in GitHub Copilot. Early testing shows ➡️ It has strong performance on agentic coding workflows ➡️ It's effective at making targeted changes, validating its work, and reducing unnecessary execution overhead on complex tasks Try it out in the GitHub Copilot app or @code 👇 github.blog/changelog/2026…

@VisualStudio Opus5 landing in Copilot Chat on day one is notable.

We're excited to bring Claude Opus 5 to Visual Studio today, making it available in Copilot Chat from day one. Built for agentic coding, professional knowledge work, and long-horizon reasoning, it's ready for your most demanding development workflows. Learn more: buff.ly/tcfHFk7 #VisualStudio #GitHubCopilot