./blog
Notes from building and breaking
Write-ups on AI agents, Claude, and the tooling around them. No newsletter, no pop-up. The RSS feed is at the bottom.
The state of AI agents: what works, what doesn't, what's next
A grounded look at where AI agents stand: the categories that deliver value today, the persistent problems with reliability, cost, and security, and how to place sensible bets without buying into either the hype or the dismissal.
AI search and SEO: what developers should do now
Search is changing as AI answers appear above the links. Here's what's still true about SEO, what's shifting, and a practical checklist for technical sites: crawlability, structured data, original content, and measuring what actually brings visitors.
Add Claude to a Laravel app with the HTTP client
You don't need a package to call Claude from Laravel. This tutorial builds a small service class on the HTTP client, adds config, error handling, queued jobs, a tool-use example, and tests with Http::fake, so the integration stays simple and easy to change.
Computer use and browser agents: what they can and can't do yet
Agents that look at a screen and click, type, and scroll can operate software that has no API. Here's how they work, where they help, why they're slower and less reliable than API-based agents, and how to run them safely.
Choosing the right Claude model for the job
Bigger models cost more and run slower, and most tasks don't need the biggest. This guide shows how to think about model tiers, test the choice on your own data, route between models, and avoid paying for capability you never use.
LLM tokens and costs, explained with real arithmetic
Tokens are the unit everything in an LLM bill is measured in. This guide explains what they are, why output costs more than input, how agents multiply spend, and the practical levers that bring a bill down without hurting quality.
Getting reliable JSON out of an LLM
Asking a model for JSON and hoping is fragile. This guide compares the approaches that work, from forced tool calls to schema-constrained output, and shows how to validate, retry, and design schemas that models can fill correctly.
Why LLMs hallucinate and what actually reduces it
Hallucinations aren't random glitches. They follow from how language models are trained and evaluated. Understanding the causes points to the fixes that work: grounding, tools, constrained answers, verification, and honest evaluation.
Slopsquatting, leaked secrets, and other AI coding risks
AI coding assistants bring risks that go beyond buggy code: invented packages, secrets in prompts, poisoned instruction files, and agents with too much autonomy. Here's what the risks are and a practical program for managing them.
Prefer a feed reader? New posts show up in the RSS feed as soon as they're published.
RSS feed