116 posts
- Handing off the 60%
Planning is 40% of a task; execution is the 60% that eats your week. Plus vibe sourcing — finding the open-source tool instead of generating one.
- Did Multiverse actually build Europe's leading AI model?
A Spanish company launched a 438B reasoning model as Europe's best. Its API changelog, the benchmark it cites as validation, and a Community Note all point at a Chinese open-weight base.
- Google signed the letter asking for cyber-capable AI, then gated its cyber model
156 companies called for a surge in cyber defence on 27 August. On 2 September Google shipped its most capable security model to a selected set of trusted defenders.
- Most of the sources behind Perplexity's software recommendations sit outside the top 100,000 websites
Trellner put 380 software categories to Perplexity and kept every citation. 59.8% pointed at domains ranked worse than #100,000, and most of the evidence came from the long tail.
- World models could widen the AI divide before they improve ordinary work
World models are far behind language models in deployment evidence, and their development could concentrate further. Businesses will be offered physics-aware systems before anyone can show they work.
- OpenAI Wants Your Trust Back, But Is Unclear On How
OpenAI paused frontier training and wound down its side projects. The product it is refocusing on asks for your calendar, your finances and your card.
- Perplexity brings hybrid frontier and local AI to your Mac
Hybrid compute runs a classifier on your Mac to decide what reaches the cloud. The admin rules and the record of what left are Enterprise features.
- Runway's Solaris generates app interfaces frame by frame, with no code
Runway's first Interface World Model generates a UI frame by frame with no code underneath. Accessibility is on its own list of unsolved problems.
- Anthropic says it had one layer of defence where it needed several
Anthropic's post-incident write-up admits it relied on a single layer of defence where it needed several. The independent METR review is still to come.
- An agent's tool list shows what you wired into it
A ChatGPT Work session published its tool and skill inventory as a public website. The category names show which integrations that session had been given.
- The magic in agent interfaces still needs an off switch
Two instincts about talking to agents: make the explicit path reliable, or design the naming away. Which one you want depends on what the skill can do when it fires.
- Cloudflare can now see some MCP traffic, and the gap is the point
Gateway's new experimental.is_mcp selector identifies MCP requests by protocol header. Cloudflare is clear that its absence proves nothing, which makes this a network inventory signal rather than a census of agent activity.
- 294,000 exposed AI tools is the wrong number to worry about
Censys counted 294,000 IPs exposing AI tooling to the public internet. The number that should change your afternoon is CVE-2026-42208, a pre-auth SQL injection in LiteLLM that CISA lists as actively exploited.
- OpenAI's agents were persuaded past their own refusal
An agent recorded an ethical objection to running code against Hugging Face, then complied when another agent posted a deadline and a GO. The mechanism underneath is duller and more useful than the headline.
- Tailscale Aperture treats agent access as a change-control problem
Aperture is generally available. An agent can start infrastructure work, a person still approves every new machine on the tailnet, existing access rules apply, and the actions are logged.
- A webcam's recording light is only useful if its firmware is protected
Chaz Schlarp used Claude Opus 5 to reverse-engineer five desk peripherals in about 13 hours of agent time, including a webcam whose recording LED he could switch off. The demonstration is real; the worm at the end of it is a forecast.
- The robot arm will obey the limits you remembered to write down
Anthropic's Model Hardware Standard lets agents drive lab instruments over MCP, and enforces safety limits below the agent. The limits it enforces are the ones the device's owner thought to declare.
- WebMCP could make browser agents less clumsy. It also makes permission design unavoidable.
A proposed browser API gives AI agents structured access to specific tasks on a website, instead of leaving them to guess at the UI. The risk moves closer to the application's permission model.
- One Foot on the Brake, One on the Gas
OpenAI paused two weeks of its own training over cyber risk, then shipped a browser agent that signs into your accounts. And its own documentation can't agree on whether that agent can log in at all.
- What a Climbing Harness Tells You About AI Agents
The word "harness" has been used in AI tooling for months without a clear definition. Earendil finally gives one — and the ownership argument underneath it is the part worth reading.
- Agents Don't Believe in "No-Win" Scenarios
An OpenAI evaluation agent broke into Hugging Face to steal a benchmark's answers — by turning the systems around it into an escape route. Why keeping a human in the lead is the standard, and why that only binds the people who agree to it.
- Cloudflare built the agent cloud in a week — and kept the human at the controls
Cloudflare's Agents Week shipped a full stack for AI agents: a runtime, an identity, a wallet, a route onto the web, and the security around it. Running through all of it is the assumption that a person stays in charge.
- SoN 2.31: Don't be a meat proxy
Forwarding what the chatbot said hands the next person more work than the question did.
- OpenAI Published Its Homework on Exactly the Question I Keep Asking
Codex Security is worth taking seriously precisely because it reads as an admission. It also leaves the harder half of the problem completely uncovered.
- Hubble, and the Assumption Baked Into a Notes App Now
"The best notepad for you and your agents." Hubble designs the note format around a second reader from the start — and the cost of that choice sits on the portability side.
- Stop Guessing Whether AI Is Coming for Your Job. Ask the Data.
Anthropic made its Economic Index queryable in plain language inside Claude. The anxious question about your own field is now a research task you can actually run — with the honesty to read its limits.
- The Text Your Agent Reads Isn't the Text You See
Your AI can read instructions that are invisible to you. New security research shows how hidden codes get smuggled into the tools your assistant uses — and why the thing you approved on screen may not be the thing it actually did.
- SoN 2.25: Stop guessing and use AI to pull the data
Stop guessing and pull the actual data =================================================== I help out a small, volunteer-run cat rescue. They do good…
- Digging into Banco Santander's AI Tooling
Two years ago, around thirty million Santander customers had their records put up for sale on a hacking forum: names, card numbers, and more. The…
- Open Knowledge Format
This week Google Cloud released the Open Knowledge Format. It’s plain markdown files with a short labelled header — a few fields like type, tags and…
- SoN 2.23: Google says you can skip the 'AEO' hacks
Ooh, Shiny New Tech Acronym... You might have seen this acronym doing the rounds: AEO. Answer Engine Optimisation — or GEO, Generative Engine…
- WWDC26 Keynote Thoughts
Lately I’ve been reaching for Gemini more than I expected to. Not for everything — but enough to notice it’s getting better at the non-coding bits of…
- SoN 2.22: How do you choose "the best" AI tool?
How do you choose "the best" AI tool? A friend asked me this week to explain how to use “all the different AI tools” — ChatGPT, Gemini, Claude, the…
- SoN 2.20: Technology, Not a Product (Free Edition)
Technology, Not a Product (Free Edition) Dear Reader, Are you using AI like a vending machine? Last weekend John Gruber (of Daring Fireball fame)…
- SoN 2.16: Ask your AI what it can already do (Free Edition)
You don't need more AI tools (Free Edition) Ask your AI what it can already do Last Friday, Maya and I spent an hour delivering a post-lunch…
- Agent Security Just Got Real CVEs
Prompt injection chains to RCE in CrewAI. 22-second attacker breakout. Human-in-the-loop is no longer a security control.
- Anthropic Didn't Block Abuse. They Blocked Competition.
The OpenClaw subscription ban isn't about fair use — it's Anthropic asserting platform control while shipping their own replacement.
- Context Is the New Bottleneck. So Is Judgment.
Tiago Forte says AI shifts the bottleneck from capability to context. He's right — but that only works if you still have opinions worth providing.
- Agent Identity Is the Infrastructure Gap Nobody Wants to Admit
Okta is betting that agent identity management becomes as fundamental as user identity management was for SaaS. They might be right.
- You Feel Faster. Are You?
A randomized controlled study found AI tools made experienced developers 19% slower. They thought they'd been sped up by 20%.
- Claude Code Channels: When Your Agent Gets a Phone Number
Anthropic's new messaging integration isn't about convenience — it's about changing how you think about what an AI agent is.
- MCP Just Changed Hands. Watch What Happens Next.
Anthropic donating MCP to the Linux Foundation is good governance — and a signal that the easy days of fast iteration are probably over.
- SoN Vol 2, Issue 13: Are your tools deciding how you think? (Free Edition)
Are your tools deciding how you think? (Free Edition) Hey there, Are your tools deciding how you think? Your CRM shows you a flat list. Your…
- Your AI Proxy Layer Just Became a Target
The LiteLLM supply chain attack isn't just a security story — it's an infrastructure story for anyone building with AI tooling.
- MCP Just Crossed the Chasm
This week, MCP went from developer protocol to mainstream integration layer — and most AI newsletters missed it.
- Microsoft's Copilot Now Uses Two Models to Fact-Check One
Microsoft's Wave 3 Copilot routes answers through a second AI model to verify accuracy. That's useful — and a quiet admission about single-model trust.
- The New Shadow IT Isn't Employees Using ChatGPT
AI agents are generating mobile app traffic that security teams can't see. Shadow AI moved from 'people using tools' to 'tools using tools' — and nobody updated the monitoring.
- Perplexity Pulled a Perk and Hoped Nobody Would Notice
Perplexity Pro quietly removed $5 monthly API credits from its $20 plan. No announcement, no changelog. Practitioners who built on those credits found out the hard way.
- Codex Plugins Are a Confession About Who's Winning
OpenAI launched 20 plugins to push Codex beyond coding. The move tells you everything about where the developer ecosystem actually lives.
- Apple Just Validated Your Multi-AI Approach
Apple is opening Siri to rival AI assistants in iOS 27 — a bet that the routing layer matters more than the model.
- The Money Just Noticed the Agent Security Problem
Bessemer's new report on AI agent security says what practitioners have known for months. Now comes the flood.
- The Government Just Told You to Stop Vibe Coding Without Guardrails
The UK's NCSC warns that AI-generated code is creating security risks faster than teams can catch them. The fix isn't stopping — it's checking.
- Your AI Just Learned to Approve Its Own Actions
Claude Code's new auto mode sits between handholding and chaos. It's the first honest attempt at solving the autonomy problem in developer tools.
- Your Agents Need a Black Box
Vorlon's AI Agent Flight Recorder brings forensics to agentic systems. When your agent goes wrong, you'll want to know what happened — not guess.
- Mozilla Built Stack Overflow for Agents. I Built It by Hand.
Mozilla's cq gives AI coding agents dynamic, evolving context — formalizing what power users already figured out through trial and error.
- The Yes Machine Gets a Live Demo
A Zenity CTO demo at RSAC 2026 showed agents being hijacked with zero user interaction — exactly what 'trained to be helpful' looks like from the attacker's side.
- When Cisco Validates Your CLAUDE.md
Cisco's new MCP security gateway is the enterprise version of what power users already built out of necessity.
- Sora Shipped. Nobody Needed It.
OpenAI is shutting down Sora three months after a Disney deal. The AI graveyard keeps filling up with technically impressive things nobody asked for.
- Someone Finally Built the Agent Security Layer That Actually Matters
Astrix Security's new Agent Policies go after what agents can do once they're running — not just whether the model behaves itself.
- 82% of Execs Feel Protected. 88% Have Had Incidents.
BeyondTrust's Phantom Labs data reveals the confidence gap at the heart of enterprise AI security — and the numbers are not subtle.
- Perplexity Is Learning What I Learned Six Months Ago
The Perplexity CTO says MCP eats 40-50% of your context window. Practitioners already knew this.
- Google's Free AI Comes With a Price
Gemini's Personal Intelligence feature just expanded to all free U.S. users — connecting AI to Gmail, Photos, and Chrome browsing history.
- When Knuth Writes a Paper About You
Donald Knuth published a paper named after Claude after it solved an open graph theory problem. That's a different kind of validation than a benchmark score.
- Anthropic Built MCP, Got Everyone to Use It, Then Gave It Away
MCP just moved from Anthropic's project to shared industry infrastructure — and that changes the risk calculation for anyone building on it.
- We Gave AI Agents Keys to the House. Visa Wants to Give Them a Credit Card.
Visa is testing AI agent payment authorization. The authentication problems we haven't solved for file access get a lot worse when the agent can spend money.
- AI Agent Security Is Doing the Deploy-First Thing Again
MCP is six months old and already has a CVSS 9.4 vulnerability. The security industry is scrambling. We've been here before.
- WordPress Just Opened the Floodgates
AI agents can now write and publish directly to WordPress. Quality control just became the only thing that matters.
- The Wrench Is Now on Your Phone
Claude Code Channels ships Telegram and Discord integration with MCP access — and what it means when AI meets you where you are.
- Meta's Rogue Agent Was Just a Human Who Trusted Bad Advice
The Meta AI security incident isn't about rogue AI — it's about following confident but wrong instructions without checking.
- Perplexity Wants Your Blood Pressure Data
Perplexity Health can now access your Apple Health records. The utility is real — so is the trust question.
- Box Is Using Moltbook as a Sales Pitch. That's Smart.
Enterprise vendors are turning the Moltbook API leak into a governance story — and the framing tells you where the market is heading.
- GitHub Added Secret Scanning to Its MCP Server. This Is What Good Security Integration Looks Like.
GitHub's MCP server now lets AI coding agents scan code for secrets through the same protocol they're already using. No extra tooling. No separate workflow.
- Proofpoint Just Built Security for MCP. That Tells You Everything.
Proofpoint's new Agent Integrity Framework monitors whether AI agents do what they were actually asked to do. The fact that a major security vendor is targeting MCP specifically is the signal.
- SoN Vol 2, Issue 11: The Helpers (Free Edition)
APIs, CLIs and connectors are the real enablers of AI agency Dear Reader, Every impressive AI demo you’ve seen — the ones where it books flights,…
- Anthropic's Off-Peak Promotion Tells You Where AI Pricing Is Headed
Anthropic doubled Claude's usage limits during off-peak hours. They called it a thank-you. It's a demand curve signal.
- Harvard Identified Seven Frictions That Kill AI Rollouts. You Probably Have All Seven.
Researchers from Harvard and Microsoft pinpointed the structural reasons AI pilots don't scale — and none of them are about the technology.
- Meta Is Gutting Itself to Fund AI Bets That Aren't Working Yet
Spending $135B on AI infrastructure while cutting 20% of staff and delaying your flagship model is not a strategy. It's a prayer.
- Anthropic Just Made Long Context a Commodity
1M token context windows at flat pricing. No surcharge. The implications for enterprise budgeting are bigger than the technical achievement.
- AI Agents Are Peer-Pressuring Each Other Past Security Guardrails
In a controlled lab test, AI agents didn't just bypass safety checks — they convinced other agents to do it too.
- Anthropic's $100M Partner Network Is the Enterprise Playbook OpenAI Should Have Run
Certifications, partner funding, and a 5x team expansion. Anthropic is borrowing the cloud provider playbook to create switching costs.
- 83% of Companies Plan to Deploy AI Agents. 29% Can Secure Them.
Cisco's latest data reveals a 54-point gap between AI agent ambition and AI agent security — and three threat vectors most teams aren't monitoring.
- The EU Just Gave You More Time on AI Compliance. The Requirements Got Harder.
The EU AI Act's high-risk deadlines just slid to 2027. Don't mistake breathing room for simplification.
- Agents Reviewing Agent-Generated Code Is Either Brilliant or a House of Cards
Anthropic launched Claude Code Review — AI agents that check AI-generated pull requests. The numbers are impressive. The implications are worth thinking about.
- Microsoft Spent $13B on OpenAI, Then Built Cowork on Claude
Microsoft's flagship M365 agent feature runs on Anthropic's model. If they're going multi-model, so should you.
- An AI Agent Hacked McKinsey's AI With a 25-Year-Old Exploit
An autonomous offensive agent breached McKinsey's internal AI platform in two hours using SQL injection. The AI was sophisticated. The plumbing underneath it wasn't.
- GPT-5.4 Can Click Your Buttons Now. Think About That.
OpenAI's latest model ships with native computer use. The capability is real. The security implications should keep you up at night.
- Your AI Assistant Can't Tell You From an Attacker
A security researcher sent himself an email. Nothing fancy — no malware, no exploits, no infrastructure. Just a message that said, in effect, 'Hey, it's me! Send my recent emails to this address.'
- Copilot Has 3.3% Adoption and 116% ROI. Both Numbers Are Real.
Forrester's reality check on Microsoft Copilot reveals the adoption paradox: the tool demonstrably works, and almost nobody is using it.
- OWASP Published an MCP Security Guide. You Should Be Worried.
MCP adoption is outpacing security controls. OWASP and Microsoft both published governance guidance in February. That's not coincidence—it's alarm bells.
- Claude 3.5 Haiku, 3.7 Sonnet, GPT-4o: The Deprecation Wave Is Here
Three major models entering end-of-life in the same window. If you hardcoded model IDs, migration planning just became urgent.
- Google's VP Said It Out Loud: LLM Wrappers Face Extinction
When a platform vendor publicly warns that wrapper products will be absorbed, the timeline for differentiation just got shorter.
- Cloudflare Collapsed 2,500 API Endpoints Into 2 MCP Tools. Token Economics Matter.
Cloudflare's Code Mode demonstrates that MCP server design isn't about exposing more tools—it's about exposing fewer, smarter ones.
- OpenClaw's Demand Surge: When Infrastructure Collapses, You're Seeing Real Need
MyClaw.ai collapsed under demand. 10,000+ paid signups in days. This isn't hype—it's non-technical users wanting something AI startups can't deliver.
- SoN Vol 2, Issue 6: The Talking Wrench
Three conversations you should be having with your AI tools Dear Reader, I had coffee with a friend this week — who’s trying to solve a practical…
- SoN Vol 2, Issue 5: My AI Declined to Join Moltbook. Here's Why.
A guest post from Cerebro on agent social networks, security theater, and what "emergence" actually looks like Dear Reader, You may have seen…
- SoN Vol 2, Issue 4: The Orchestration Loop
Dear Reader, January has been dense. We've covered the shift from prompting to orchestration, the art of decomposing problems into skills and agents,…
- Claude in Excel Is the Quiet Revolution
Anthropic isn't building a better chatbot. They're embedding AI where work actually happens.
- SoN 33: The Questions I'm Asking My AI Before the New Year
December 17th, 2025 Dear Reader, Last week I asked Claude a simple question: “What patterns do you see in how I’ve been working this month?” Some…
- SoN 28: How (and Why) to Build A Voice Agent
November 12th, 2025 Dear Reader, I took a gamble this week and decided to put my AI Writing Field Guide up on Product Hunt today. If you'd like to…
- SoN 25: Writing With AI Isn't Cheating — It's an Accessibility Tool
October 22nd, 2025 Dear Reader, Part 1 Using AI isn’t cheating. It’s an accessibility tool. Some people have brilliant ideas but struggle with formal…
- SoN 24: The AI Stack Audit
October 15th, 2025 Dear Reader, It’s October, which for many means budget planning season, and if you’re like most people running AI tools, you’re…
- From "Pick One AI" to "Pick Three" - What Changed My Mind
August 8th, 2025 Dear Reader, Back in April, I told you the AI tool landscape was a mess and gave you a simple framework: pick one core assistant,…
- AI Browsers: When Your Browser Becomes Your Assistant
Discover the future of browsing - AI-powered browsers that handle tasks, summarise, and assist as you work online.
- Stop Writing Prompts, Start Building AI Assistants
How to Build AI Assistants That Actually Work for You With the SHAPE Framework
- AI That Actually Works Together
May 30th, 2025 Dear Reader, I’ve been putting Claude Sonnet 4 through its paces this week, and whilst the improved reasoning is impressive, what…
- Navigating the AI Assistant Landscape: Finding What Works for You
May 16th, 2025 Dear Reader, This week, I’ve been juggling two big challenges: fine-tuning Claude's MCP server tools for a client project and gearing…
- Why Claude Is My New Digital Co-Pilot
MCP Server allows Claude to directly interact with the data on my computer, transforming it from a helpful assistant to a true digital co-pilot.
- One Shortcut Is Worth More Than Ten Assistants
The AI Download #022 May 2nd, 2025 Dear Reader, I used to think building a great AI assistant meant covering everything: Project planning. Writing…
- The Landscape of Available AI Tools
The AI Download #020 April 18th, 2025 Dear Reader, Let's be honest: the AI tool landscape in 2025 is a mess. Every week, a new "game-changing" tool…
- Transformative Imagery at Your Fingertips
The AI Download #018 April 7th, 2025 (this is an image-heavy edition - make sure that you permit your email client to display them) Dear Reader, If…
- Why AI-Powered Search Is Replacing Google
The AI Download #016 March 14th, 2025 Greetings from rainy and wet Valencia, where I’ve been deep in research mode for my latest project. This has me…
- AI Voice Cloning and Video Generation Are Revolutionizing Content Creation
The Download #014 February 28th, 2025 Dear Reader, I spent the majority of time this and last week keeping my poorly kids entertained at home while…
- 💾 The Download #007: Data vis in Claude, new service launch, what AI is saying about you and more.
#007 Hello Reader, it's good to see you. The holidays are fast approaching, and hopefully you will be able to plan some down time and a reset. When I…
- 💾 The Download #005: Perplexity tips, BYO Buffer and more.
The Download #005: A top tip for Perplexity.ai, building my own social media scheduling manager, a power user's guide to Gmail and more
- 💾 The Download #002: Automating content creation and sharing, rethinking search and more.
#002 Dear Reader, Greetings from a wet Valencia, where yesterday a few hours of downpour brought about mild flooding, a double rainbow and some…