Ai tag
View All Tags
Google turned Search into a task-execution engine, OpenAI admitted reasoning traces resist deterministic control, and Cloudflare replaced static access rules with continuous behavior scoring. Meanwhile, 2,622 leaked TLS certificates are still valid in the wild. Here is what each of those means for your deployment pipeline.
Anthropic built a whole landing page — claude.com/import-memory — for the groundbreaking technology of copying text from one browser tab and pasting it into another. The "feature" is marketed as a way to "transfer your preferences and context from other AI providers to Claude." There is no transfer. There is no protocol. The entire mechanism is a prompt you paste into ChatGPT, copy the output, and paste it into Claude's memory settings. This could have been a tweet in 2023.
Every few months someone announces a "revolutionary AI productivity tool" that turns out to be three API calls in a trench coat. Google's Workspace CLI is not that. It ships a dynamic command surface built from Discovery Service metadata, a skills pack north of 100 entries, and structured JSON output that agents can actually parse without hallucinating a schema. This is just another Gmail script bundle.
Donald Knuth publicly credited Claude Opus 4.6 with solving an open math problem he'd been working on for weeks, CISA added two actively exploited CVEs to the KEV catalog, and half a dozen ICS/OT advisories dropped with CVSS 9.4 scores. Meanwhile, Google and OpenAI shipped cheaper models and Next.js 16 quietly became the default scaffold.
Google made per-request inference meaningfully cheaper, three webapp exploit classes from the mid-2010s showed up again in production software, and the PHP ecosystem started admitting out loud that it cannot sustain its own patch cadence. A useful week if you pay attention to the right parts.
Pantheon Content Publisher shipped to general availability after months of pre-release messaging since October 2025. The AI-assistive capabilities target content acceleration and quality — two words that marketing departments love combining and operations teams learn to distrust. For CMS teams evaluating this, the question worth asking is whether the controls keep up with the speed, or whether you're just publishing mistakes faster.
Nothing screams "move fast and break users" quite like encrypting someone's data with a key they can never recover. February delivered that gem alongside agent tooling that finally works, Drupal plumbing nobody will celebrate, and platform defaults that only protect you if you bother wiring them up.
February 2026 felt like the month where everyone agreed AI coding is "real now," then immediately discovered the bill: secrets sprawl, shaky governance, fragile tests, and communities trying to stay relevant without becoming AI-themed marketing departments.
Cloudflare viNext is the fastest path today to run modern Next.js apps on Workers with less adapter glue. But the safe adoption pattern is not "AI rewrites everything." The practical pattern is AI for scoped transforms, deterministic checks for every change set, and a canary rollback plan.
That combination gives speed without losing production stability.
SearXNG in Drupal AI assistants matters because it gives you controllable, inspectable retrieval without funneling your org's prompts and context into yet another black-box "trust us" API.
I needed to get data out of Jira. Not just the title, but the full description, comments, and all attachments, packaged neatly for use in other scripts. The official way involves wrestling with an API that feels like it was designed by a committee that never spoke to each other. The unofficial way involves paying $20/month for a SaaS tool that is just a glorified curl command wrapped in a pretty dashboard. I chose the third way: build it myself.
WordPress 7.0 Beta 1 is here, and it is not an incremental update. The two big moves are real-time collaboration and native AI integration via the Abilities API. Whether that makes WordPress "an intelligent workspace" or just a more complicated CMS depends entirely on the execution.
I dug into the developer-facing features. Some are genuinely good. Some need caveats.
Agentic AI moves fast. A few lines of code, a powerful LLM, and suddenly an agent is doing something that looks impressive. The rapid iteration is addictive, but it leads to a development style I call "vibe coding" -- tweak a prompt, rerun, and if the output feels right, ship it.
This works for a demo. It is a recipe for disaster in production.
You're paying $20-200/month for an LLM subscription. You have access to Claude, GPT, or Gemini. You have Playwright. You have Python. So why are you also paying for Crawl4AI, Moltbot, Firecrawl, or whatever the "AI-powered web scraping" flavor of the month is?
Because marketing works. That's why.
Dries Buytaert has introduced "Drupal Digests," a new initiative that uses AI to summarize development activity across Drupal Core, Drupal CMS, Canvas, and the AI Initiative. This is one of the most practical uses of AI in open-source project management I have seen.
Instead of building AI into Drupal, this is using AI to make the Drupal project more navigable. Smart move.
A recent report from Security Boulevard, based on a Gravitee study, puts a number on the AI agent security gap: approximately 1.5 million AI agents operating in large US and UK organizations with zero active oversight. That is not a theoretical risk. That is a ticking time bomb with a billing address.
I read the full report. The findings are worse than the headline.
I shipped an OpenAI o3-mini planner for my Drupal CMS 2 AI agent and kept a local rule-based fallback, because Drupal's 2026 AI direction is clear: intelligent automation is coming fast, but production teams still need predictable behavior when external AI calls fail.
If you want your WordPress content to appear in AI-generated answers, prioritize this sequence: answer-first writing, schema accuracy, clean crawl controls, maintained plugins, and weekly validation in Search Console/Bing Webmaster Tools. As of February 17, 2026, this is practical on WordPress 6.9.1 and should be treated as an operational SEO workflow, not a one-time checklist.
I built Project Context Connector, a Drupal module that makes site configuration programmatically accessible to AI agents, automation scripts, and monitoring tools through multiple interfaces including native Model Context Protocol support.
Every AI coding tool has its own config format for MCP servers. Claude uses JSON, Codex uses TOML, Gemini uses a different JSON schema. Setting up the same 18 servers across all three means editing three files, remembering three formats, and hoping you didn't typo a credential. I built mcp-web-setup to do it once.