Posts

Graphify: The Missing Map for Brownfield Codebases Back when I was digging into OpenSpec, I kept circling the same problem: on a brownfield codebase, nobodycan tell you with confidence what’s actually connected to what. Specs help, but a spec still describes behaviour. That’s the gap that pulled me into Graphify. Graphify turns a codebase — code, docs, SQL schemas, configs, even PDFs — into a queryable knowledge graph, and ships as a skill for Claude Code, Cursor, Codex, Gemini CLI, Copilot, Aider, and a dozen-plus other assistants. Its like a dependency map for your code base. Instead of an agent guessing at architecture from whatever files happen to be open, it gets a real map to query. ...

Beyond the Co-Pilot: My Q2 Plan for Multi-Agent Automation First off, if you noticed complete radio silence on here over the last quarter, I have a pretty solid excuse! Work has been an absolute whirlwind. Between non-stop flights between Australia, Singapore and India, setting up new offshore engineering teams, and establishing new engineering ways of working from scratch, my calendar completely ate my side projects for breakfast. On top of that, team resource constraints meant I had to jump back into active microservice development alongside my management duties—building out core service APIs, handling edge cases, and pushing production code. ...

Spec-Driven Development for the Rest of Us: OpenSpec Last month I found GitHub’s Spec Kit and how it flips the traditional dev workflow — spec as the durable source of truth, code as the disposable output. Spec Kit’s /constitution-first, plan-then-build ceremony assumes you’re starting mostly from a blank slate. Almost nothing I touch day-to-day looks like that. It’s always years old services, inherited conventions nobody remembers agreeing to, and test suites that are more archaeology than architecture. ...

Flipping the Script: A Look at GitHub’s Spec Kit I’ve spent the last few posts obsessing over agents that write, run, and heal tests. This time I want to zoom out one level, because I think I’ve been skipping over the artifact that actually matters most to an AI coding agent: the spec itself. That’s what pulled me into Spec Kit, GitHub’s open-source toolkit for what they’re calling “Spec-Driven Development” (SDD). The pitch is deceptively simple but genuinely inverts how most of us have worked for the last twenty years. In traditional development, the spec (if one even exists past the kickoff meeting) is a scaffold — you lean on it briefly, then throw it away the moment code starts shipping. Code becomes the source of truth, and the spec rots in Confluence somewhere, quietly lying to whoever reads it next. ...

The Hype: Jason Huggins Just Announced Vibium, and Browser Automation Will Never Be the Same If you’ve spent any time in the software testing space over the last two decades, you know the name Jason Huggins. He’s the guy who created Selenium back in 2004, basically founding modern web test automation, and later gave us Appium for mobile. So when Jason Huggins drops a new open-source project, the entire testing community stops what it’s doing and looks up. ...

Running huge regression suites every time I push a small change to a repo is super inefficient and slow, especially as projects start growing and getting complex. Lately, I’ve been diving deep into AI-Driven Risk-Based Selection to make my testing workflow fast, targeted, and lean. Smarter Builds with Risk-Based Selection Instead of blindly firing off every single test on every commit, I’ve been tinkering with a script that inspects the exact files changed in a Git diff. The script parses the modified code, maps out the underlying dependencies, and picks only the top 15% or so of tests that are actually affected by those changes. ...

Lately, it feels like we’ve crossed into a totally new era with AI tools. Over the past few months, my focus has shifted from simple code auto-completions to playing around with Autonomous Agentic Workflows. Instead of just having an assistant that finishes a line of code, an agent acts more like a goal-oriented script. You give it a high-level target, and it can plan out multi-step actions, check the results, and adjust its approach on the fly. ...

As I’ve been leaning heavier into AI-assisted coding and scripting for my own side projects over the past year, my API token usage shot up fast—it went from a cheap experiment to an invoice that made me do a double-take. Lately, my focus has been all about Token Optimization, Context Caching, and keeping my setup lean. Things I’ve Been Tweaking 1. Prompt Caching & Smart Context Truncation When OpenAI and Anthropic rolled out prompt caching, I jumped on it immediately for my personal scripts. By caching heavy system prompts, common schemas, and reusable code patterns, I managed to slash my monthly LLM API spend by over 45% while actually speeding up execution times. ...

With multimodal models like GPT-4 Vision and Claude 3 Opus getting so much sharper, test automation is finally moving beyond fighting with the HTML DOM and into actual Visual and Spatial Verification. Things I’ve Been Experimenting With 1. Screenshot-Based Visual Assertions In the past, visual testing meant wrestling with rigid pixel-by-pixel diff tools that would blow up your build over a tiny font rendering shift. Lately, I’ve been feeding raw UI screenshots directly to vision-capable LLMs along with plain English questions: “Is the checkout button properly aligned below the order summary? Is any text overlapping?” ...

Flaky tests caused by shifting DOM element IDs have been the bane of my existence for as long as I’ve been writing test automation. I got tired of constant pipeline failures over minor UI tweaks, so lately I’ve been tinkering with Runtime Self-Healing Mechanisms directly inside my Playwright test runners. What I’ve Been Building Recently 1. Dynamic Selector Healing I set up dynamic error interceptors inside my test suites. When a locator lookup fails (say, a TimeoutError: element '#submit-order-v2' not found), the test runner catches the exception, snags a snapshot of the surrounding DOM tree, and hands it off to an LLM to request an alternative CSS or XPath selector. ...