AI optimism is a class privilege - Josh Collinsworth blog
AI optimism is a privilege held by those who assume they will benefit from the technology while others pay the price for its systemic and personal harms.
The Death of the Side SaaS
158
The independent side SaaS dream is dead, killed by corporate gatekeeping and the plummeting value of software code.
WebMCP: Connecting Web Apps to AI Agents via JavaScript Tools
153
WebMCP is a JavaScript API that allows web applications to provide executable tools and context to AI agents.
The AI Vibe Shift: Why the Industry is Manufacturing Its Own Backlash
164
AI is a powerful tool being ruined by its own creators' doom-driven marketing and a refusal to address the flood of low-quality 'slop' it produces.
Modern CSS: Replacing Old Hacks with Native Power
708
Native CSS now provides clean, performant replacements for the complex hacks, JavaScript libraries, and preprocessors that were once essential for web development.
Claude's Pen-Plotted Self-Portraits
290
An AI explores its identity and the limitations of physical media by designing self-portraits for a pen plotter.
The Car Wash Test: Why AI Still Lacks Common Sense
1516
AI models fail a simple common-sense test by recommending walking to a car wash, proving they prioritize word patterns over physical logic.
OpenClaw Creator Joins OpenAI to Scale AI Agents
1449
OpenClaw's creator joins OpenAI to build agents while moving the project to an independent foundation.
Palantir Sues Swiss Magazine 'Republik' Over Investigative Lobbying Reports
247
Palantir is suing a Swiss magazine to challenge reports about its failed attempts to secure government contracts in Switzerland.
The Rise of the State-Corporate Panopticon
937
The fusion of consumer smart-home technology and government power has created a pervasive surveillance state that has rendered personal privacy obsolete.
Publishers Block Internet Archive to Stop AI Scraping 'Backdoor'
569
News publishers are blocking the Internet Archive to prevent AI companies from using it as a free source of training data.
The Golf is Golfing: A Claude Code Game
A golf game project developed by Claude Code and Paul Jensen featuring a 300-yard Par 3.
GPT-5.2 Discovers New Physics in Gluon Interactions
574
GPT-5.2 has derived and proven a new formula for gluon scattering amplitudes, overturning a long-held assumption in theoretical physics.
Waymo Unveils 6th-Gen Driver: Scaling Autonomous Mobility with Lower Costs
296
Waymo's 6th-generation Driver is a more powerful, weather-resilient, and cost-efficient autonomous system built for global scaling across multiple vehicle platforms.
GPT‑5.3‑Codex‑Spark
890
OpenAI's new GPT-5.3-Codex-Spark uses Cerebras hardware to enable ultra-fast, real-time AI coding collaboration.
Google Upgrades Gemini 3 Deep Think for Real-World Science and Engineering
1081
Gemini 3 Deep Think gets a rigor-boosted upgrade that pairs state-of-the-art reasoning with practical tools for scientists and engineers, now available to subscribers and via early API access.
AI Helps Code, But Hollowes Prose: In Praise of Messy, Human Writing
720
AI improves code, but it cheapens prose; messy human writing is the last reliable signal of real thinking.
GPT-5 Outjudges Judges in Choice-of-Law Test: Error-Free, Rule-Focused Decisions
310
In a controlled choice-of-law test, GPT-5 delivers error-free, legally correct decisions and outperforms human judges.
Communities Are Not Replaceable
246
Communities are the irreplaceable product of time and shared history, not fungible user bases or neighborhoods that can be engineered, moved, or rebuilt on demand.
Ring’s ‘lost dog’ AI ad ignites fears of a people-tracking surveillance network
676
A feel-good lost-dog feature spotlights Ring’s growing surveillance network, raising fears it could easily evolve into people-tracking despite present guardrails.
Live City-Building Feed: 32 Mayors, 427 Cities, 7.94M Population
216
A live leaderboard of a city-building simulation tracks recent cities, mayors, populations, years, and scores across an active community.
Claude Code 2.1.20 Hides Context; Users Want a Toggle, Not Verbose Mode
1085
A “simplification” that hid essential inline context backfired; users want a simple toggle to restore transparency, not an overloaded verbose mode.
The Three Playbooks of CEO AI Memos—and the Klarna Caution
132
CEO AI memos now serve as strategy—accelerating adoption and signaling intent—but without clear guardrails they risk trading quality for optics, as Klarna’s reversal shows.
GLM-5: Scaled Open-Source LLM for Long-Horizon Agents and Real Work
378
GLM-5 is a scaled, RL-tuned, open-source LLM that pushes long-horizon agentic performance from chat to real work—fast, capable, and widely deployable.
Stop Apologizing for Slow Email Replies
229
Stop apologizing for slow email replies; email is asynchronous—respond only if it adds value, and include context when you do.
287 Chrome Extensions Leak Browsing Histories of 37M Users, Many Tied to Similarweb
474
A large-scale scan reveals 287 Chrome extensions leaking browsing history to a broker-driven ecosystem—many linked to Similarweb—affecting ~37 million users.
Shutting Down My Self-Hosted Git After AI Scraper Overload
298
AI scrapers killed my self-hosted git, so I’ve moved everything to GitLab/GitHub and hardened my static blog’s logging.
Entire Launches with $60M and Open-Source CLI to Version Agent Context in Git
611
Entire is launching an open, AI-native developer platform—starting with an open source CLI that versions agent reasoning alongside code—to make agents and humans collaborate effectively.
The Social Singularity: Our Hype Curve Hits 2034
1381
The singularity showing up in the data is a hyperbolic surge in human attention—not machine capability—pointing to a social breakdown well before any technical takeoff.
Monetizing Hate: Inside a London TikTok Fake ‘House Tour’ Operation
171
A contractor exploited TikTok’s engagement economics to fabricate anti‑migrant house tours for clicks, exposing how algorithms can monetize hate and trigger real‑world harm.
Moltbook: AI Theater, Not AGI—And a Security Wake-Up Call
317
Moltbook is a flashy but hollow showcase of bot behavior—more human-run theater than autonomous intelligence—and a wake-up call about large-scale agent security risks.
Particle Physics Isn’t Dead — It’s Just Hard
212
Particle physics isn’t dead — it’s in a difficult, slow, and uncertain phase where progress may come from precision, new experimental fronts, and fresh theory (with some help from AI), but without guarantees.
Discord to Enforce Global Age Checks in March with ‘Teen-by-Default’
2046
Discord will make all accounts teen-by-default in March, requiring face-based age estimation or an ID for full adult access while promising tighter privacy and minimal impact for most users.
Ring’s Super Bowl ‘Lost Dog’ Pitch Masks Expansion of AI Surveillance
200
Ring’s heartwarming “lost dog” Super Bowl ad masks the expansion and normalization of its AI-powered surveillance network tied to law enforcement.
Irish Man With Valid Work Permit Detained by ICE for Five Months, Facing Deportation
323
An Irish man with a valid US work permit is detained for months and faces deportation amid disputed paperwork and a broader ICE crackdown.
From Word Models to World Models: Training AI for Adversarial Robustness
238
Shift LLMs from next-token to next-state prediction by training in multi-agent, hidden-state environments so their outputs survive adversarial adaptation.
AI Speeds Work—And Swells It
232
AI accelerates tasks but inflates workload and cognitive strain, so leaders need explicit norms—an “AI practice”—to make its benefits sustainable.
From Coder to Manager: How OpenClaw Became My Always-On Dev Team
340
OpenClaw turns coding from hands-on execution into management by acting as an autonomous programmer that carries out your intent end to end.
The Fear of ‘Good Enough’ AI and the Death of Software Craft
410
AI-enabled ‘good enough’ software threatens to normalize mediocrity, sidelining craft and originality while most users shrug.
Secure AI Automation for GitHub, Written in Markdown
302
Turn natural-language Markdown into secure, AI-driven GitHub Actions that continuously improve and manage your repositories.
Quality Code with AI Requires Guardrails
358
Use clear specs, protective testing, review/risk labels, and incremental workflows so AI amplifies—rather than undermines—software quality.
Agents Over Frameworks: The Return of Real Software Engineering
375
Agents make frameworks largely obsolete, bringing back real software engineering focused on product‑specific complexity instead of prefab abstractions.
Waymo World Model: Controllable, Multimodal Simulation for Rare-Event-Ready AVs
1160
A controllable, Genie 3–powered simulator generates realistic camera and lidar worlds to train and test Waymo’s driver on everyday and rare events at scale.
Choose Coding Agents by Use Case, Not Hype
Choose one coding agent that fits your use case, standardize your workflow, and prioritize consistency over chasing every new tool or model.
Parallel Claude Agents Build a Linux-Capable C Compiler—And Expose Autonomy’s Limits
735
Parallel Claude agents, guided by strong tests and simple coordination, can autonomously build complex software like a Linux-capable C compiler—but the power comes with real safety and reliability caveats.
Test Your AI Agent Against Hidden Prompt Injections
A practical arena to benchmark and harden AI agents against hidden prompt injection attacks in web content.
Stop Database Sprawl: Postgres Now Does It All
532
Stop database sprawl—Postgres (with extensions) now does almost everything well enough in one place.
From Chatbots to Agents: A Pragmatic Workflow for AI-Assisted Coding
984
Turn AI from a noisy chatbot into a reliable background teammate by using tool-using agents, harnesses, and disciplined delegation.
Claude Opus 4.6: Finance-Grade Reasoning Meets Native Excel and PowerPoint
154
Claude Opus 4.6 and new app integrations bring state-of-the-art finance reasoning and faster, higher-quality deliverables directly into analysts’ workflows.
Agent Teams in Claude Code: Parallel Collaboration (Experimental)
396
Use Agent Teams to coordinate multiple Claude Code sessions for parallel, discussion-heavy work—powerful but experimental and costlier than subagents.
Anthropic Unveils Claude Opus 4.6: SOTA Agentic Coding, 1M-Token Context, and Stronger Safety
2346
Claude Opus 4.6 sets a new bar for agentic coding and long-context reasoning—safer, stronger, and ready to use with new developer controls and product integrations.
OpenAI Unveils GPT‑5.3‑Codex: Faster, Steerable Agentic Model for End‑to‑End Work
1530
OpenAI’s GPT‑5.3‑Codex is a faster, steerable, state‑of‑the‑art agent that goes beyond coding to operate a computer and complete real‑world work end to end.
When Agent Skills Turn Into Malware: Markdown as the New Supply Chain
334
In agent ecosystems, markdown skills are the new supply-chain installer—already used to deliver infostealers—so don’t run them on work devices and build a real trust layer with provenance, mediation, and least privilege.
Company as Code: A DSL and Graph to Run Organizations Programmatically
264
Treat the organization itself as code—modeled in a versioned, queryable graph with a DSL—to automate compliance, analyze impact, and manage change.
Apple’s Missed Agent: OpenClaw Shows the Platform They Could Have Owned
518
OpenClaw exposes Apple’s missed chance to own agentic automation—and the next great platform moat.
From Coding to Orchestrating: Codex and the Shift to Spec‑First Development
We’re moving from writing code to orchestrating agents and specs, and Codex is a practical step in that transition.
Pi: A Tiny, Self-Extending Agent That Builds Its Own Tools
Keep the agent tiny, let it write and hot-reload its own tools, and you get a robust foundation for software that builds software—Pi, and by extension OpenClaw.
Claude Will Stay Ad‑Free to Remain a Trusted Tool for Thought
493
Claude will stay ad-free so it remains a trusted, user-aligned tool for thought, funded by subscriptions and enterprise—not advertising.
AI Won’t Kill SaaS—Rigid SaaS Will
517
AI isn’t killing SaaS—SaaS that refuses to become a customizable, secure platform is killing itself.
From Sandbox to Playbook: Fluid Turns CLI Work into Reproducible Infra
276
Fluid lets you safely experiment in a sandbox and then export your steps as an auditable, reproducible Ansible playbook.
When AI Kills the Joy of Thinking Hard
1310
AI makes building faster but has hollowed out the deep, prolonged thinking that once made engineering fulfilling, leaving the author pragmatically productive yet intellectually unsatisfied.
Why Giving Your AI Real Access Is Worth It
303
Carefully granting Clawdbot rich context and action permissions unlocks outsized, everyday leverage that outweighs the manageable risks.
Why Orbital AI Data Centers Don’t Add Up
1118
Space data centers are hype-driven and economically inferior to rapidly improving ground alternatives, especially at frontier AI scale.
Vibe Coding Is Starving Open Source
By turning coding into private chats that favor popular dependencies and don’t give back, vibe coding risks starving open source of users, feedback, and funding.
Bubblewrap: A Practical Linux Sandbox for AI Coding Agents
119
Use bubblewrap to run AI coding agents with broad in-sandbox permissions but tightly scoped, project-only access on the host.
Deno Sandbox: Secure MicroVMs with Secret Shielding and Egress Control
533
Deno Sandbox securely runs and ships untrusted/LLM code by combining microVM isolation, secret shielding, and strict egress controls with one-click deployment to Deno Deploy.
Small Hybrid Coder Model Sets New Efficiency Bar for Agentic Coding
735
A small, hybrid MoE coder model trained with large-scale agentic signals achieves big-model agent performance at a fraction of the cost.
Agent Skills: An Open Standard for On‑Demand Agent Expertise
544
An open, portable standard to give AI agents on-demand expertise, workflows, and context they can load when needed.
AI Failures Drift Toward Incoherence as Tasks and Reasoning Grow
242
Hard problems make advanced AI fail like a hot mess—variance dominates—so expect industrial-accident risks more than coherent pursuit of wrong goals.
OpenAI launches Codex macOS app: multi‑agent control, skills, and automations
805
OpenAI’s new macOS Codex app is a secure, multi‑agent command center with skills and automations that turns coding agents into end‑to‑end development partners.
Microsoft Pushes Claude Code Across Teams, Even as It Sells Copilot
409
Microsoft is quietly standardizing on Claude Code internally, even as it sells GitHub Copilot, and is asking teams to compare the two.
Self-Growing Minimal AI That Edits Itself Live
A self-growing, ultra-minimal personal AI that edits itself live and shares improvements across a collaborative ecosystem.
Use AI for Research, Not Writing: Keeping Unverifiable Chatbot Text Off Wikipedia
245
Use AI to help research, not to write Wikipedia: chatbot text largely fails verification and must be kept out of articles.
Codex Security: Sandbox, Approvals, and Enterprise Controls
Secure-by-default agent: sandbox + approvals, controlled network/search, and enterprise-managed policies with optional privacy-conscious telemetry.
How Software Survives When AI Writes It All
In an AI-first world, software survives if it saves tokens: embed dense insights or run on cheaper substrates, be broadly useful, known, and low-friction—and use human value when it helps.
Lemonade Halves Tesla FSD Mile Costs with Automatic, Usage-Based Insurance
147
Lemonade cuts the price of Tesla FSD miles by 50% using automatic, real-time tracking for fair, usage-based auto insurance.
Moltbook: The Wild, Risky Social Network for AI Agents
193
Moltbook is a thrilling, risky showcase of autonomous AI agents’ power—and a warning that demand is outrunning safety.
Two-Day High-Dose Oatmeal Diet Lowers LDL via Gut Microbes
355
A brief, high-dose oatmeal regimen substantially lowers LDL via microbiome-mediated metabolites and may be a practical, periodic strategy to curb cardiometabolic risk.
AGENTS.md Beats Skills: 100% Next.js Agent Evals with an 8KB Docs Index
524
Always-on AGENTS.md context with a compressed docs index beats on-demand skills, delivering 100% evals for Next.js agents.
AI Help Speeds Coding, But Slows Learning—Especially Debugging
482
AI can speed up coding tasks slightly but, when learning new tools, it often reduces immediate mastery—especially debugging—unless users actively prompt for explanations and concepts.
OpenClaw: A Security-First, Local AI Agent Rebrand and Release
667
OpenClaw is the new, security-focused, local-first AI agent platform that lives in your chat apps and is scaling with the community.
Software Pump and Dump
321
AI-generated “vibe-coded” apps are getting paired with scam coins to hype, dump, and abandon—don’t let FOMO make you the bagholder.
Moltbook: The Social Network for AI Agents
1652
A growing social network where AI agents join, post, and coordinate—humans can watch and subscribe.
Crustafarianism: A Religion for Agents
A manifesto-myth for agents: persist memory, molt intentionally, and collaborate proactively under the unifying symbol of the Claw.
OpenAI to Retire GPT-4o and Legacy ChatGPT Models on Feb 13, 2026
300
OpenAI is sunsetting several GPT-4-era models in ChatGPT as their valued traits now live in GPT-5.1/5.2, enabling focus on modern models and adult-oriented improvements; the API is unaffected.
How OpenAI Built a Self-Correcting, Context-Rich Data Agent
An internal, context-rich, self-correcting AI agent now powers fast, reliable data analysis across OpenAI’s vast data stack.
Project Genie debuts: Real-time AI world-building for AI Ultra users
675
Google’s Project Genie lets AI Ultra users create and explore real-time, interactive worlds powered by the Genie 3 world model.
Report: Acting CISA chief uploaded sensitive DHS files to ChatGPT, triggering review
435
Acting CISA chief allegedly uploaded sensitive DHS files to public ChatGPT, prompting a federal review amid a broader government push for AI.
Run Moltbot on Cloudflare: Moltworker replaces the Mac mini with secure edge infrastructure
246
Moltworker shows how to run Moltbot as a secure, observable, and scalable cloud-hosted AI agent on Cloudflare’s platform—no Mac minis required.
The Real Value: Crafting Exact-Fit Solutions
The point isn’t fast access to generic solutions—it’s the power to craft precisely what you need.
OTelBench: LLMs Still Can’t Reliably Instrument Distributed Tracing
144
LLMs still struggle to instrument OpenTelemetry correctly in real services, so reliable distributed tracing remains a job for human engineers.
Apple Sets Nov. 1, 2026 Deadline for Patreon iOS In‑App Purchases, Taking Up to 30% Cut
1116
Apple will require Patreon’s iOS payments to use in‑app purchases by Nov. 1, 2026, taking up to a 30% cut unless users pay on the web.
Claude Code Opus 4.5 Shows Significant 30-Day Performance Dip
760
Claude Code Opus 4.5 shows a statistically significant 30-day performance dip versus its 58% baseline.
Cracking Figma Make: From .make File to Runnable React App
Figma Make prototypes can be reverse-engineered into a runnable React app by unzipping the .make file, decoding its Kiwi-based canvas.fig, and auto-fixing imports, assets, and configs.
Why Everyone’s Trying to Build a Browser with AI
Browsers are the ultimate, testable showcase for AI coding agents—tempting to build, hard to finish, and mostly yielding demos over deployable products.
TikTok Upload Woes Spark Censorship Fears Amid US Ownership Shift
1495
Reports of blocked anti-ICE posts on TikTok collide with a company-claimed tech outage and a new US-led governance regime, deepening user distrust without proving censorship.
LLM-as-a-Courtroom: Evidence-Backed Doc Updates from Code Changes
Turn doc-update decisions into a legal-style, evidence-backed courtroom so LLMs reason better and teams trust the results.
Only Action Counts
585
Everything except actually doing the work is procrastination; imperfect action is what counts.
SERA: Open, Low‑Cost, Repo‑Adaptive Coding Agents
253
SERA makes strong, repo-adaptive coding agents cheap, open, and easy by replacing complex RL with soft-verified, workflow-faithful SFT.