Anthropic announced that future Claude models will generate watermarked text and attach signed credentials to created files. This change complies with the European Union AI Act and applies globally.
What does this mean for someone using Claude, and how does the underlying technology work? This post breaks down the announcement in simple terms.
You can watermark a sentence by influencing word choices. But an image has pixels, an audio clip has sound measurements, and a video has thousands of frames. Can those very different files carry a hidden signal that survives being shared, cropped, trimmed, or compressed?
This final post explains how watermarking works across images, audio, and video. It also separates an embedded watermark from file metadata, and shows why a positive or negative check is evidence with limits rather than a final answer about a file.
Ask ChatGPT or Claude to write a function, a SQL query, or a JSON document, and much of the output has to follow exact rules. If a text watermark works by gently favouring one valid next word over another, what happens when there are very few valid choices?
This second post answers that question. It shows where code still has room for a watermark, what rewriting and translation can do to the signal, and why a detector score needs to be read in context.
Public benchmarks can tell you a great deal about AI model capability. They still may not tell you which model will perform best on the work you actually need done.
This final post in a 3-part series shows how to investigate that gap with a small personal evaluation: a representative task, controlled user-visible conditions, and a rubric written before testing. It also examines what happened when I applied that method to three model systems.
Can you tell whether a paragraph was written by ChatGPT, Claude, Gemini, or a person? Looking at its style is not a reliable way to know. One possible answer is a text watermark: a hidden signal built into an AI's normal word choices.
This first post explains how that signal can work and what a matching detector can actually establish. Part 2 looks at code, strict formats, and editing; Part 3 moves to images, audio, and video.
I built HTML Deck Studio because I wanted slide changes to appear as ordinary text diffs. PowerPoint, Keynote, and Google Slides support useful visual editing, but their native workflows do not treat each slide as source code that an AI agent or script can edit directly.
HTML Deck Studio stores presentations as HTML, CSS, and JavaScript. A browser renders the deck, Git records text changes, and a script can export it to PDF. This post explains that file-centric design and the extra engineering responsibility it creates.
In this post, I will explain how HTML Deck Studio works, why I designed it as a file-centric viewer, and how it simplifies presentation authoring for developers and AI workflows.
While building a personal project that visualizes live operational data on a map, I ran into a classic frontend challenge: I needed to render thousands of moving coordinates at 60 FPS while keeping the interface keyboard-navigable, clickable, and accessible to screen readers.
I quickly discovered that treating the UI as an "all-or-nothing" choice between the DOM and a <canvas> element was the wrong approach. HTML, SVG, Canvas, and WebGL each handle different trade-offs between rendering throughput and browser ergonomics.
I’m sharing my notes on how to combine these surfaces into a clean hybrid architecture—and a simple rule of thumb for deciding which tool to reach for.
When multiple AI coding agents or developers work on the same repository simultaneously, code conflicts are only half the battle. One agent might require Python 3.10 to evaluate legacy scripts, another might need Python 3.12 for modern async features, and a third might test breaking dependency upgrades. If all instances share a single local environment, runtime collisions become inevitable.
Combining Git worktrees with VS Code Dev Containers addresses both kinds of isolation. Worktrees give each task a separate directory and branch checkout. Dev Containers give each directory its own runtime dependencies. This post explains the extra Git metadata mount that makes the combination work.
Knowing that benchmark scores depend on their test setup is only the first step. The next challenge is recognizing what different benchmark families actually measure—and where their evidence ends.
This field guide maps six useful categories that appear in model announcements and provides a repeatable way to read them together. The categories are practical rather than exhaustive, and some benchmarks span more than one.
VS Code Insiders updates frequently, so a new build can introduce a crash, extension failure, or slowdown before a fix is available. Reverting means installing a specific earlier Insiders build and pausing automatic updates long enough to keep it in place.
This guide shows how to record your current version, find a known working commit, build the correct download URL for your platform, install it, and later restore updates. It also explains which settings and extensions the reinstall should preserve.