Loading...
Loading...

Software Engineer · Toronto
Bobby Jose
C# & Azure roots, now deep in agentic AI. Field notes on the tools, ideas, and announcements changing how we ship — written for engineers who build.
October 27, 2026 · 13 min read
Prompt engineering became context engineering, and context engineering became spec-driven development. Here's the Claude Code setup I'm adopting so the spec, not the chat, is the source of truth, and why almost none of it depends on your language. Same files and loop for iOS, Mac, Python, .NET and Android; only four commands change.
October 24, 2026 · 7 min read
On Wednesday the PHP project shipped emergency fixes for two critical bugs in the runtime itself, and sites around the world went into maintenance mode to install them. I've never written PHP. What pulled me in is how the bugs were found: by security teams using frontier AI models. Here's the story as the advisories and reports tell it.
October 23, 2026 · 7 min read
A power event in a single Google Cloud zone in London on Tuesday took banking apps and big UK brands offline for hours. Here's what Google has said so far, why one zone's problem reached a whole region, and what I'm checking in my own app backends.
October 16, 2026 · 10 min read
In four days Meta shipped a closed frontier model and gave away the weights of the tier below it. That's a strategy, not a contradiction, and it's not the open source Meta used to talk about. Plus: Tesla's 'flying' Roadster, and the questions I can't stop asking about it.
October 9, 2026 · 7 min read
Personal AI agents can now book flights and buy groceries, and a growing list of websites won't let them in. Some blocks are deliberate, some are collateral damage from bot defenses, and the fix everyone is reaching for is the same: agents that say who they are.
October 2, 2026 · 8 min read
Microsoft published a detailed review of Monday's Azure OpenAI outage in Sweden Central, and a second gateway incident followed on Tuesday. Here's what Microsoft's reviews say happened, why knowing a failure mode isn't the same as planning for it, and what I'm checking in my own apps.