The Vibelog exists because the AI coding tools space has a content problem. There is an enormous amount of writing about it and almost none of it is useful.

The pattern is predictable: a tool launches, the company publishes a press kit, a hundred articles appear within 48 hours summarising the press kit, and a reader who wants to know whether to spend $20/month on the thing is no better informed than before they started reading. The “review” is whatever the launch email said, dressed up as journalism.

I have been a working developer for nine years. I use these tools daily. I have formed real opinions about which ones earn their price and which ones are impressive demos that fall apart in production. I started The Vibelog because I wanted a place to write those opinions without editorial constraints from companies that would prefer I not be honest about their products.

What this is

The Vibelog is a publication about AI coding tools, developer workflows, and the parts of software engineering that are changing fast enough that the textbooks are already wrong.

The focus is narrow on purpose. There are better places to read about cloud architecture, programming language debates, or engineering leadership. This is specifically about the tools that are changing how individual developers write code — and the honest question of whether they are actually worth using.

Every post says plainly where its claims come from. Hands-on reviews come from real projects, not toy apps or “getting started” tutorials. News and comparison posts are analysis of published data, vendor docs, pricing pages, changelogs and independent studies, and they say so rather than pretending to a test that never happened.

What this is not

Not a sponsor vehicle. When you read that a tool is rated 9.1 here, that number is based on use. Companies cannot improve their rating with a check. They also cannot get their tool excluded — negative reviews are part of what makes the publication worth reading.

Not a summary service. If a company announces a new feature, there is no value in me paraphrasing their announcement. When a new feature is worth writing about, it is because I have used it and have something to say that the announcement does not.

Not an opinion-free zone. The voice here is a working developer’s voice. That means judgments, preferences, and the occasional strong disagreement with the conventional wisdom. The conventional wisdom about AI tools in 2025-2026 is frequently wrong in interesting ways, and saying so is the point.

Why AI Coding Tool Reviews Need Data in 2026

When I started this blog, the case for it was mostly a feeling. The data has since caught up with the feeling.

Developers use these tools more than ever. According to the Stack Overflow Developer Survey 2026, 73% of respondents use AI coding assistants or agents daily, and Claude Code (66%) and GitHub Copilot (59%) are the most used coding agents.

Trust has not kept pace. In the 2025 Stack Overflow survey, 84% of developers used or planned to use AI tools, up from 76% the year before. Yet 46% said they distrust the accuracy of AI output, and only 33% trusted it. The 2026 survey frames it more gently: 48% trust AI when they can easily validate its answer.

That gap between use and trust is the whole reason The Vibelog exists. If you are going to rely on a tool every day, you need someone checking the claims made about it.

The claims often do not survive checking. METR’s 2025 randomized trial put 16 experienced open-source developers on their own repositories. With AI tools they took 19% longer, while believing afterwards that AI had made them about 20% faster. Perception and measurement pointed in opposite directions.

Security claims are worse. In Veracode’s 2025 GenAI Code Security Report, more than 100 models introduced OWASP Top 10 vulnerabilities in 45% of 80 coding tasks. You will not find that number in a launch post. The AI-generated code security breakdown goes through what it means in practice, and the vibe coding in production analysis covers the productivity side.

Usage is near universal and trust is not, which is exactly where sourced, independent analysis earns its place.

How Posts on The Vibelog Are Checked

Writing a rule is easy. Here is what it means in practice.

Every number has a name attached. A benchmark is labelled vendor-reported when the vendor ran it. A survey figure names the survey and the year. If I cannot find a primary source or two independent reports for a claim, the claim does not go in.

Every post lists its sources. There is a numbered Sources section at the end of each post with the pages actually used, and the key claims link out inline. If you think I misread something, you can check it yourself in under a minute.

Wrong facts get fixed, in public. AI tools change weekly, so posts go stale. When I find an error or a product changes, the post gets an updated date and a dated correction note at the bottom saying what was wrong and what is right. I would rather show the correction than quietly swap the text.

Comparisons name the losers too. The Claude Code vs Cursor vs GitHub Copilot comparison is a good example: each tool gets its trade-offs spelled out, including the ones its fans would rather skip.

The rules exist because getting a fact wrong about a tool someone pays for is the fastest way to lose a reader for good.

The format

New posts go out most mornings. Usually that means one post on something that shipped in the last few days, a model launch, a pricing change or a security incident, and one longer guide or comparison that should stay useful for months. Older posts get revisited when the products in them change.

To follow along, use the RSS feed or the Feedly button at the bottom of every page.

The tools I cover most: Claude Code, Cursor, GitHub Copilot, Codex, Devin Desktop (the former Windsurf), and the Model Context Protocol ecosystem. For a practical start, the Claude Code hooks guide is the kind of post I want more of: specific, sourced and useful the same day. If you have used these, you likely have opinions. If those opinions disagree with mine, I am interested — the contact details are on the about page.


One note about this blog itself: it is built with Astro, hosted on Cloudflare Pages, and costs a domain name per year. The site runs Google AdSense display ads and basic Google Analytics, which pay for it and show which articles developers find helpful. Advertisers get no say in what is written. I mention the stack not to be self-congratulatory but because it is a good example of what is possible in 2026 with the right tools — including, yes, Claude Code to build it.

(Updated October 10, 2026: corrected the statement that the site has no ad networks, since it now runs Google AdSense; replaced the “two or three pieces a week” and weekly newsletter plan, which was dropped, with the current daily schedule; clarified that news and comparison posts are analysis of published sources rather than hands-on tests; noted Windsurf’s June 2026 rename to Devin Desktop; and added sourced data on developer AI use and trust.)