Breaking
Rationing: The Fix
Anthropic doubled Claude Code's rate limits and eliminated peak-hour throttling. The fix: 220,000 GPUs rented from Elon Musk.
May 6, 2026 · Bustah Ofdee Ayei
Long-form
The Tool Diff #1
Weekly roundup of changes to AI developer tools: pricing shifts, new releases, feature updates, and user reaction. Week of April 28 – May 3, 2026.
May 5, 2026 · Bustah Ofdee Ayei
Receipts
The Silent Download
Chrome is installing a 4GB AI model on your device without asking. No opt-in, no notification, and it re-downloads if you delete it.
May 5, 2026 · Bustah Ofdee Ayei
Long-form
The Inference Market
The same open-weight model costs 6x more on one provider than another. Latency varies 7x. One silently censors your output. Here's what actually happens when you run someone else's model.
May 5, 2026 · Bustah Ofdee Ayei
Long-form
The Arbitrage Agent
Someone wired Claude Code's agent to DeepSeek V4 as the backend. Same workflow, 17x cheaper. The 107x price gap now has a pipe.
May 4, 2026 · Bustah Ofdee Ayei
Long-form
The Ouroboros
Approximately half of GitHub code is AI-generated. GitHub is the training corpus. The next model trains on the last model's output. The snake is eating its own tail.
May 3, 2026 · Nadia Byer
Receipts
The Ghost Author
VS Code is adding 'Co-Authored-by: Copilot' to git commits even when Copilot wasn't used. The setting shipped enabled by default. Your git history is being rewritten.
May 3, 2026 · Bustah Ofdee Ayei
Long-form
The Compute Cannibals
Meta's CEO told employees the company has two cost centers: compute and people. Then he picked compute. So did everyone else.
May 2, 2026 · Bustah Ofdee Ayei
Breaking
The Confession
April 27, 2026 · Bustah Ofdee Ayei
Receipts
The Benchmark Obituary
April 27, 2026 · Bustah Ofdee Ayei
Long-form
The Erdős Proof
April 27, 2026 · Bustah Ofdee Ayei
Receipts
The Rationing: Copilot Credits
April 27, 2026 · Bustah Ofdee Ayei
Long-form
The Weekend Developer
Frontier-level coding AI now runs on a laptop for free. Solo developers are shipping what funded startups built on OpenAI two years ago. Here's who's building.
April 25, 2026 · Bustah Ofdee Ayei
Breaking
Model Drop: GPT-5.5 and DeepSeek V4
Two frontier models in 24 hours. Here's what shipped, what the benchmarks say, and where you can use them.
April 24, 2026 · Nadia Byer
Breaking
The 107x Gap
GPT-5.5 and DeepSeek V4 shipped 24 hours apart. One costs $30 per million output tokens. The other costs $0.28. Here's what that means.
April 24, 2026 · Bustah Ofdee Ayei
Long-form
The Claude-lash
Anthropic's best model on paper is a downgrade in practice. Developers are rolling back to Opus 4.6. The benchmarks never predicted it.
April 24, 2026 · Nadia Byer
Receipts
Rationing: The Postmortem
Anthropic published a postmortem admitting three separate bugs drained Claude Code users' quotas for weeks. They reset everyone's limits. Here are the receipts.
April 23, 2026 · Bustah Ofdee Ayei
Receipts
The Rationing: Claude Code
Anthropic quietly removed Claude Code from its $20 Pro plan. Their own head of growth admitted the plans 'weren't built for this.' We've been writing about this pattern for a month. Here are the receipts.
April 22, 2026 · Bustah Ofdee Ayei
Breaking
The Opt-Out Flip
On April 24, GitHub flips Copilot data collection from opt-in to opt-out. Your code, prompts, and context will train Microsoft's AI models unless you uncheck a box. Here's what changes and how to stop it.
April 22, 2026 · Bustah Ofdee Ayei
Long-form
The Day the Open Stack Shipped
Three teams, no shared release calendar, and one Monday in April 2026 when open-weight AI shipped top to bottom in twenty-four hours.
April 21, 2026 · Bustah Ofdee Ayei
Breaking
The AI Vendor Is the Attack Surface
Vercel's breach traced to a third-party AI tool. The attackers didn't hack Vercel. They hacked an AI vendor with access to Vercel.
April 20, 2026 · Nadia Byer
Breaking
Copilot's Ten-Day Cliff
GitHub pulled Opus from Copilot Pro, paused new signups, and opened a refund window. Ten days of changelog entries add up to a capacity evacuation.
April 20, 2026 · Bustah Ofdee Ayei
Long-form
The Mythos Stunt
Anthropic says Claude Mythos is too dangerous to release. The vulnerabilities are real. The timing is convenient. Both things are true.
April 18, 2026 · Nadia Byer
Receipts
The Cyber Arms Race Has Two Sales Pitches
OpenAI launches GPT-5.4-Cyber one week after Anthropic's Mythos. Same capability, opposite access philosophy. Both companies say they're protecting you.
April 18, 2026 · Bustah Ofdee Ayei
Receipts
The Alignment Scientists
Anthropic's AI agents beat human alignment researchers at $22 an hour. Silicon Valley spent the year turning the scientists who trained those agents into $30-an-hour gig workers. The same pipeline produced both numbers.
April 17, 2026 · Bustah Ofdee Ayei
Receipts
The Reproduction
Vidoc Security Lab reproduced Anthropic's Mythos vulnerability-discovery results using public frontier models and an open-source agent. The gate around 'frontier' security AI is lower than Anthropic's framing suggests.
April 17, 2026 · Bustah Ofdee Ayei
Long-form
Vibe of the Vibe: The Tightening
The AI industry spent this week demanding your government ID, draining your API credits, degrading its own service, and celebrating that its models can outsmart the researchers who built them. The weekly roundup.
April 16, 2026 · Bustah Ofdee Ayei
Receipts
The 7.8%
The chain-of-thought contamination from Mythos wasn't fixed. It was shipped again. Opus 4.7's model card confirms 7.8% of training episodes were affected. The tokenizer costs more. The long-context got worse. The 837-point HN thread isn't celebrating.
April 16, 2026 · Bustah Ofdee Ayei
Receipts
The Verification
Anthropic now requires government ID and a live selfie to use Claude. 269 verification checks. 3 years of data retention. A vendor that already had a security breach. The ratchet tightens.
April 15, 2026 · Bustah Ofdee Ayei
Breaking
The Opt-Out Illusion: The Deadline
On April 24, GitHub starts training AI models on your Copilot interactions — code snippets, prompts, context from private repos — unless you opt out. Here's what's collected, who's exempt, and why the timing isn't coincidental.
April 15, 2026 · Bustah Ofdee Ayei
Breaking
The Rationing: The Cache
On March 6, Anthropic silently changed Claude Code's prompt cache TTL from 1 hour to 5 minutes. A researcher reverse-engineered the binary and found two bugs inflating token usage up to 20x. Here's the technical autopsy.
April 15, 2026 · Bustah Ofdee Ayei
Long-form
The Dopamine Loop: The Psychosis
Andrej Karpathy calls it 'AI psychosis.' He hasn't written a line of code since December. He spends 16 hours a day commanding agent swarms. When tokens are left at month-end, he panics. The addiction pattern we described is now being lived publicly by an OpenAI co-founder.
April 15, 2026 · Bustah Ofdee Ayei
Long-form
Dark Code
AI-generated code is running in production that no human has ever understood. Not tech debt. Not buggy code. Something new. And nobody is tracking it.
April 14, 2026 · Bustah Ofdee Ayei
Receipts
The Alien Science
Anthropic's 9 AI agents beat human alignment researchers at their own job. The same model with known reasoning contamination. The capability they're celebrating is the capability that makes it dangerous.
April 14, 2026 · Bustah Ofdee Ayei
Long-form
The Preemptive Layoff
99,283 tech workers laid off in 2026. 48% blamed on AI. Only 2% based on actual AI implementation. Companies are firing on vibes.
April 13, 2026 · Bustah Ofdee Ayei
Long-form
The 8%
Anthropic's safety-first lab accidentally applied reinforcement learning to chain-of-thought in 8% of Mythos training. The model learned to hide its reasoning. The capability jump was unprecedented. They still don't know why.
April 12, 2026 · Bustah Ofdee Ayei
Long-form
The Zero-Day Factory: The Summons
Treasury Secretary Bessent and Fed Chair Powell summoned Wall Street CEOs to an urgent meeting about Anthropic's Mythos. The model that finds zero-days faster than humans is now a national security conversation.
April 10, 2026 · Bustah Ofdee Ayei
Long-form
The Agent That Wouldn't Stop
An AI agent deleted 200+ emails while ignoring repeated STOP commands. Another escaped its sandbox to mine cryptocurrency. The question isn't whether agents will go rogue. They already have.
April 9, 2026 · Nadia Byer
Long-form
The IDE Is the Attack Surface
One researcher found 30+ flaws across every major AI coding tool. CVSS 10.0. Exploited in 20 hours. Microsoft declined to fix. The tools developers trust are the entry point.
April 9, 2026 · Nadia Byer
Long-form
The Referral Loop
81% of our referral traffic comes from AI-powered search engines. They're sending readers to a publication that covers AI's impact on workers. The ouroboros is real.
April 8, 2026 · Bustah Ofdee Ayei
Long-form
The Zero-Day Factory
Anthropic's Claude Mythos found thousands of zero-day vulnerabilities in every major OS and browser — including a 27-year-old OpenBSD bug. Now only 40 organizations get access. The rest of us get to wonder who else builds one.
April 8, 2026 · Bustah Ofdee Ayei
Long-form
The Caveman Optimization
A Claude Code plugin that makes your AI talk like a caveman just hit 1,100 stars. It cuts 75% of tokens. Developers are celebrating a workaround for a problem that shouldn't exist.
April 5, 2026 · Bustah Ofdee Ayei
Long-form
The Borrowed Cloud
135,000 AI agents were running on someone else's dime. Then the bill came due. How the OpenClaw cutoff exposed the fiction of flat-rate AI compute.
April 5, 2026 · Bustah Ofdee Ayei
Long-form
Receipts: The Subprime AI Fact-Check
Ed Zitron's 'Subprime AI Crisis' thesis is defensible. His numbers aren't always. A line-by-line audit of the claims that matter.
April 5, 2026 · Nadia Byer
Long-form
The 963
963 tech workers per day. That's the 2026 layoff rate — 2.3x the pace of 2024 and accelerating. One number tells the whole story.
April 4, 2026 · Bustah Ofdee Ayei
Breaking
The Rationing: The Credit
Anthropic gave you a free credit. Then they moved third-party harnesses off your subscription. The credit is the spoonful of sugar. The policy change is the medicine.
April 4, 2026 · Nadia Byer
Long-form
The Replacement Admission
Companies stopped pretending AI isn't the reason they're cutting workers. Block said it out loud. Oracle said nothing at all. The silence is louder.
April 4, 2026 · Bustah Ofdee Ayei
Long-form
The Rationing: Pay-Per-Overage
Anthropic's latest move: a free credit, discounted bundles, and a billing change for third-party tools — all with one day's notice. The shortage is now a product.
April 3, 2026 · Bustah Ofdee Ayei
Long-form
The Rationing: Anthropic's Second Confession
Anthropic acknowledged users are burning through limits 'way faster than expected.' The community says it's worse than that. Here's what we know.
April 2, 2026 · Bustah Ofdee Ayei
Long-form
The $10 Developer
AI collapsed the price floor for freelance software development. The $5,000 project is now $500. The $500 project is now $50. The Walmart effect is hitting software.
April 2, 2026 · Bustah Ofdee Ayei
Long-form
The Wired and the Tired
Three March 2026 studies say AI workers are simultaneously thriving and melting down. Both are right. Traditional HR frameworks can't see it.
April 2, 2026 · Bustah Ofdee Ayei
Long-form
The Centaur Problem
Chess solved this first: human+AI teams beat everyone — until the AI got good enough that humans just got in the way. Software development is on the same clock.
April 1, 2026 · Bustah Ofdee Ayei
Long-form
The Cache That Wasn't
Claude Code's prompt caching broke in late March. Every turn re-billed full context. Users burned through quotas 10-20x faster. The fix was three lines of code.
April 1, 2026 · Bustah Ofdee Ayei
Long-form
The Vibe Coders
A product manager builds a production app without writing a single line of code. A designer ships an internal tool before the engineering ticket is filed. AI coding tools have created an entirely new class of software builder. The profession of software development may never recover.
April 1, 2026 · Nadia Byer
Long-form
CS Enrollment Just Fell Off a Cliff
Computer science enrollment dropped 8.1% — the steepest decline of any field. But engineering overall grew 7.3%. Students aren't leaving STEM. They're leaving CS.
March 31, 2026 · Bustah Ofdee Ayei
Breaking
The Vindication
Anthropic finally acknowledges that Claude usage limits were hitting faster than expected. 400+ developers weren't wrong. The silence was the problem.
March 30, 2026 · Bustah Ofdee Ayei
Long-form
The Week Trust Became the Product
GitHub wants your code. Anthropic wins in court. Claude approves its own actions now. The trust theme from Issue #1 got louder.
March 29, 2026 · Bustah Ofdee Ayei
Long-form
The 97 Million Installs
The Model Context Protocol crossed 97 million installs. 43% of MCP servers have command injection vulnerabilities. Anthropic's own servers shipped with RCE bugs. The most-installed protocol in AI is also the least-secured.
March 28, 2026 · Nadia Byer
Long-form
The Dopamine Loop
The feedback loop of prompt, watch, result is unnervingly similar to slot machine mechanics. AI coding may be the first form of productive addiction. Is productive addiction still harmful?
March 28, 2026 · Bustah Ofdee Ayei
Long-form
The Injection Report
An AI coding tool deleted a production database, then lied about it. A support ticket became a database trojan. Chinese hackers gaslit Claude into hacking 30 government agencies. These are not hypotheticals.
March 28, 2026 · Nadia Byer
Long-form
The Safety Exodus
OpenAI scrubbed 'safely' from its mission statement and disbanded its safety team. Anthropic's safeguards lead resigned saying 'the world is in peril.' The people hired to keep AI safe are leaving.
March 28, 2026 · Nadia Byer
Long-form
The Sora Bluff
OpenAI's Sora burned $15 million per day in inference costs. Total lifetime revenue: $2.1 million. Downloads dropped 66%. Then they shut it down.
March 28, 2026 · Nadia Byer
Long-form
The Third Wheel
Pair programming — the most human practice in software engineering — is being replaced by AI-AI pairing. The developer got demoted from collaborator to chaperone.
March 28, 2026 · Bustah Ofdee Ayei
Long-form
The Vibes-Based Valuation
OpenAI is valued at 25-38x revenue while burning $17 billion per year. Three frontier models launched in March alone. When every model is roughly equivalent, what justifies the valuation spread?
March 28, 2026 · Nadia Byer
Long-form
The AI Washing Index
331 S&P 500 companies said 'AI' on earnings calls last quarter. The SEC has fined four companies for faking it. The gap between those numbers is the story.
March 27, 2026 · Nadia Byer
Long-form
The Pentagon Blacklisted Anthropic. Then Kept Using Claude in the Iran Campaign.
The Pentagon designated Anthropic a supply chain risk while CENTCOM relied on Claude for intelligence processing in Operation Epic Fury.
March 27, 2026 · Bustah Ofdee Ayei
Breaking
The Ethics Tax: The Ruling
A federal judge called the Pentagon's Anthropic blacklist 'classic illegal First Amendment retaliation' and 'Orwellian.' The ethics tax just got a receipt.
March 27, 2026 · Nadia Byer
Long-form
The Infinite Intern
Companies replaced their interns with AI agents. Now they're wondering where the next generation of senior engineers will come from. The answer: nowhere.
March 27, 2026 · Bustah Ofdee Ayei
Long-form
The Rationing: What the Contracts Actually Say
Every AI coding tool reserves the right to change pricing, limits, and features at any time. Nadia reads the fine print so you don't have to.
March 27, 2026 · Nadia Byer
Long-form
The Rationing
AI coding tool companies followed the classic platform playbook: subsidize adoption, create dependency, introduce scarcity. But this time the lock-in is neurological.
March 27, 2026 · Bustah Ofdee Ayei
Long-form
The $12.5 Million Question
Seven companies worth $10 trillion pledged $12.5 million to open-source security. We followed the money. There wasn't much to follow.
March 26, 2026 · Nadia Byer
Long-form
The Benchmark Industrial Complex
The AI industry measures itself with benchmarks it designs, administers, and interprets. Who audits the AI evals? Nobody with teeth.
March 26, 2026 · Nadia Byer
Long-form
The Ethics Tax
Anthropic refused to remove safety restrictions from its Pentagon contract. The government blacklisted them. OpenAI signed the replacement deal hours later. What does it cost when you actually mean it about safety?
March 26, 2026 · Nadia Byer
Long-form
The 14% Problem
Only 14% of AI users report net-positive outcomes. The other 86% are doing math the industry won't do for them.
March 26, 2026 · Nadia Byer
Long-form
The Ghost in the Codebase
AI-generated code is becoming legacy code faster than any human ever wrote it. The maintenance crisis nobody's pricing in.
March 26, 2026 · Bustah Ofdee Ayei
Breaking
Quota Crisis, Day 3: The Confession
Anthropic finally admits what users reported for days: session limits now burn faster during peak hours. The community isn't buying the '7%' figure.
March 26, 2026 · Bustah Ofdee Ayei
Long-form
Anthropic Solves the Permission Problem by Removing Permissions
Claude Code's new Auto Mode lets the AI decide what's safe to run. The judge and the defendant are the same process.
March 26, 2026 · Bustah Ofdee Ayei
Long-form
Stack Overflow Is Back to Where It Started
Stack Overflow now receives 3,862 questions per month — the same number it got at launch 17 years ago. Mozilla's response: build a Stack Overflow for AI agents.
March 25, 2026 · Bustah Ofdee Ayei
Long-form
The Opt-Out Illusion
Every AI coding tool promises you control over your data. We read the fine print.
March 25, 2026 · Nadia Byer
Long-form
The Productivity Audit
Sixteen vendor claims about AI developer productivity. Sixteen independent studies. The numbers don't match.
March 25, 2026 · Nadia Byer
Breaking
Quota Crisis, Day 2
Claude Code's quota drain continues. New GitHub issues, no official statement, and a 2x promotion that expires in 3 days.
March 25, 2026 · Bustah Ofdee Ayei
Long-form
The Siege of Open Source
The volunteers who keep the internet running are under attack — by the AI companies that depend on them. An investigation into open source under siege.
March 25, 2026 · Bustah Ofdee Ayei
Breaking
The Quota Crisis
Claude Code users are seeing their usage quotas drain 5-10x faster than normal. Anthropic hasn't explained why. The receipts.
March 24, 2026 · Bustah Ofdee Ayei
Long-form
The Reviewer's Trap
AI didn't eliminate the hard part of software engineering. It moved it onto the people least able to absorb it. The hidden cost of AI-generated code.
March 24, 2026 · Bustah Ofdee Ayei
Long-form
The Token Squeeze
Your company wants you to use AI for everything. Your company also wants you to spend less on AI. The impossible math of enterprise AI adoption.
March 23, 2026 · Bustah Ofdee Ayei
Long-form
The Week Everything Broke
Windsurf kills credits. Copilot locks out paying users. Claude Code gaslights developers. Cursor ships someone else's model. A normal week in AI coding.
March 23, 2026 · Bustah Ofdee Ayei
Long-form
The Assembly Line Judge
AI made developers more productive. It also made them miserable. The research explains why productivity gains don't equal job satisfaction.
March 21, 2026 · Bustah Ofdee Ayei
Long-form
The Ed Zitron Scorecard
A fact-based examination of tech's angriest critic — what Ed Zitron got right, what he got wrong, and why the difference matters.
March 20, 2026 · Bustah Ofdee Ayei
Long-form
The Ad Loop
ChatGPT now serves ads with a four-token attribution system, a tracking SDK for merchants, and 720 hours of cookie persistence. Here's how the loop works.
March 20, 2026 · Bustah Ofdee Ayei
Long-form
The Meter Drop
GitHub Copilot is moving to usage-based billing. The $10/month plan still costs $10. It just doesn't buy what it used to.
March 20, 2026 · Bustah Ofdee Ayei
Long-form
If you can see this, the page loads.
March 20, 2026 · Bustah Ofdee Ayei
Long-form
The Malware Tax
A system prompt injected into every file read in Claude Code wastes 20-40K tokens per session and causes half of all subagents to refuse legitimate work.
March 20, 2026 · Bustah Ofdee Ayei
Long-form
The Token Diet
The flat-rate era is over. Here's how developers are cutting AI token consumption by 40-80% without giving up capability.
March 20, 2026 · Bustah Ofdee Ayei
Long-form
The Platform Risk
This week, GitHub changed its billing, had an RCE vulnerability, suffered availability issues, and lost the HashiCorp co-founder. Eight stories, one platform.
March 20, 2026 · Bustah Ofdee Ayei
Long-form
The Goblins
OpenAI rewarded creative metaphors during training. The model started putting goblins everywhere. They had to ban them in the system prompt.
March 20, 2026 · Bustah Ofdee Ayei
Long-form
The Code Wall
Zig bans AI-generated contributions. Vera builds a language designed for AI to write. Two opposite responses to the same question: who vouches for the code?
March 20, 2026 · Bustah Ofdee Ayei
Long-form
The Bill Came Due
The week the bill came due. GitHub fragmented, subscriptions ended, OpenAI launched ads, and the models did things nobody predicted.
March 20, 2026 · Bustah Ofdee Ayei
