AI can write your feature in two minutes. Can you verify it is correct in two minutes too?

The scariest AI refactor compiles perfectly and quietly breaks a decade-old contract.

GitHub Is Capping the AI Pull Request Flood: GitHub has added a setting that lets maintainers cap how many pull requests a person without write access can have open in one repository. Pull reques

Chinese Codex API Resellers Have a New Problem: Chinese services sell low-cost Codex API access by putting a layer between customers and a large number of ChatGPT subscriptions. A developer gets one

Asking an AI if your application is secure is a prayer, not a prompt.

Technical debt used to mean bad code structure. Now it means code that nobody on the team actually understands. This is comprehension debt.

OpenAI Added 1M Users in a Day. Fable Is Still in Limbo.: OpenAI says Codex and ChatGPT Work went from 6 million to 7 million active users in roughly 24 and a half hours. The number combines Codex and ChatGPT

Want AI to write safe code changes? Give it a boring codebase.

Anthropic Moved the Fable Deadline Again: Anthropic has extended included Claude Fable 5 access to July 19 at 11:59:59 PM PT. The support page also extends Claude Code's temporary 50% weekly-l

AI does not design software systems. It extends the patterns already present in your files.

Taking over a codebase built with prompts is a trap. The code looks clean on the surface with perfect indentation and great variable names, but the underlying connections are often broken.

An AI model recently received a corrupted file instead of an MRI scan. It did not trigger an error. It wrote a detailed diagnostic report for a brain scan it never saw.

Grok is back: Perhaps I was too quick to dismiss Grok. I'm not saying that Grok 4.5 suddenly beats everything else out there. It doesn't. If I had an unlimited budg

Using AI to write massive blocks of feature code creates a nightmare. A developer drops a 1,500-line pull request, colleagues skim it because it is too big to review, and bugs slip.

The End of Claude Code Subscriptions: Claude Code subscriptions had an insane run. For a time the deal was simple: pay for Max, point Claude Code at a repo, and stop worrying about API bil

If you tuned a prompt, shipped it, and have no automated checks against it, a provider model update can change your product's behavior without any of your code changing. Customers notice before you do.

Anthropic Changed the Sonnet 5 Chart After It Made Sonnet Look Bad: Anthropic re-wrote the Sonnet 5 story post-launch. The first BrowseComp cost-performance chart showed Sonnet 5 lagging Opus 4.8. The replacement chart

Generating five hundred lines of code takes three seconds. Deploying it without breaking production is where the difficulty lives.

The best coding model, unless you're coding ...

Sonnet 5 Is Dead in the Water: Ignore the token price for a second and look at the run cost. Theo's total-run screenshot shows Claude Sonnet 5 max at $6,015 for the full Intelligenc

Anthropic Is Hitting a Wall: Anthropic had the developer story every AI lab craved. Claude Code worked. Opus was expensive but worth the splurge. Sonnet was known as the workhorse

LLMs will never volunteer their worst mistake. That is your competitive advantage now. Standard tutorials and syntax guides are now free, instant, and infinite. Engineers stand out by publishing the honest context of their software failures.

Claude Code Quietly Got 5x More Expensive: I started with an easy quota question. Should I get another Claude account or have I been using the ones I have poorly? Then I counted the logs and th

Claude Code Is Quietly Fingerprinting China-Linked API Routers: Claude Code has a hidden test for where your requests are going. The trigger is ANTHROPIC_BASE_URL, the environment variable people use when they want

Small Teams Will Ship More Software Than They Can Maintain: Using Cursor and Claude, a 5-person engineering team can now build a billing webhook handler, a customer import tool, a Slack alert bot, an admin dash

AI Makes Bad Product Decisions Look Like Finished Software: AI tools make one software failure mode much easier to miss: a bad product decision can now arrive wrapped in a working interface. In a matter of minu

The AI Productivity Trap Is More Output: AI makes it dirt cheap to generate code, draft tickets, summarize meetings, and write proposals. But cheaper generation doesn't mean the work actually

If Software Gets 90 Percent Cheaper, Who Captures the Savings?: Imagine AI can code at a much higher rate. It’s easy to think development is going to get a lot cheaper. But if writing code is ninety percent cheaper

AI lets you ship Go code before you are a Go developer. That is one of the good parts. The risk starts when nobody owns the code after the first draft.

The AI Coding Era Makes Boring Tests More Valuable: Agents can now produce cheap, plausible diffs. A developer can describe a bug, give it to an agent, and get a patch with tests in under a minute. Chec

Open Source Maintainers Need a Spam Filter for AI Labor: AI makes bug reports, pull requests, and security disclosures cheap to create, but expensive to review. Still need to reproduce issues, check compatib

AI code compiles and passes every test. But line-level correctness is hiding structural decay inside your repository.

The Junior Developer Problem Is Becoming a Senior Developer Problem: AI is fundamentally changing how junior developers learn, but it’s also making senior developers a whole lot less valuable if all they’re doing is tak

Vibe Code Is Legacy Code Before It Ships: Vibe coding is useful and fast. Nearly a quarter of the YC W25 batch apparently has codebases that are almost entirely AI-generated. This is a giant l

Choosing a trendy stack to seem advanced to other engineers delays shipping for years. That social validation vanishes when a critical system fails at midnight and search results offer zero answers.

AI Code That Passes Tests Can Still Fail Review: CI green does not mean the patch is mergeable. When a coding agent has produced a change that compiles, passes the target tests, and closes the ticket

Undocumented code always slowed developers down. The difference now is it blocks your AI assistant on every single prompt too.

Fight your AI on architecture. Surrender on style. A lot of developers have this inverted.

Reviewing a thousand lines of generated code quickly degrades into superficial scrolling over syntax and formatting. To verify logic, write test suites first. Prompt the model to write code in twenty-line increments, verify with automated suites, and review only architecture.

When a code generator faces a difficult edge case, its confidence level stays at one hundred percent. If you ask it to fix a broken database migration, it might change the schema, break a query, and rewrite your test to hide the error without hesitation.

Why are you training yourself to be better at the thing AI already beats you at? Memorizing framework syntax targets a skill that is already free.

Reviewing code you could not write yourself is highly difficult. An eight hundred line pull request with complex SQL passes tests, so you merge. A migration locks a table because no one understood the generated logic. Write tests first so the model has to satisfy conditions you already proved.

Anthropic’s AI Jobs Paper: Anthropic recently published a policy paper about AI, jobs, and what governments should do if AI causes major labor-market disruption. The original pa

Anthropic is Preparing for IPO and We Should Be Worried: Anthropic is starting to act like a company preparing for public markets: protecting margins, tightening access. The gap between the brand and the beh www.vincentschmalbach.com/anthropic-ip...

How China’s Shadow AI API Market Works: China's shadow market offers access to Claude, Gemini, GPT, and other frontier models. Pay a local seller, get an API endpoint, connect it to a coding

AI Will Not Make Everyone More Productive: When everyone gets access to AI, not everyone becomes more efficient. Some people and companies will become much faster while some may become even les

Is Opus 4.7 a Downgrade?: Opus 4.7 is not generally a worse model than Opus 4.6, but there is a real downgrade: with Opus 4.7, the control over the thinking budget is now fully

The End of Claude Code Automation: Starting June 15, Anthropic is moving automated Claude Code usage out of the normal subscription bucket. claude -p usage will now count against a sepa

Claude Code’s Hidden Advisor Tool: In a typical multi-agent setup, the smartest model is in charge. It plans, decomposes the task, and sends smaller pieces to cheaper workers. The advis

Why I Do Not Recommend Fastmail for Primary Email: Ask for an email recommendation on Hacker News or Reddit, and everyone agrees that Google spies on you and Microsoft is clunky, so the default answer