From the archive
Earlier articles retain their original dates. Software versions, prices and product availability may have changed since publication.
88 articles · Page 3 of 8
Real overlap math, sample standup schedules, and a side-by-side comparison of working with Philippine vs Indian dev teams from Sydney/Melbourne. Why a 2-3 hour gap improves shipping velocity.
Real AUD comparison: Sydney/Melbourne in-house vs Manila offshore for a 600-hour custom app. Loaded rates, total project cost, hidden line items, and the break-even math most Australian CFOs miss.
Why Australian builders and developers consider Philippine software teams: AEST/AEDT overlap, lower delivery cost, English fluency, and construction workflow experience.
ClawHub has 8,000+ exposed instances and skills with file-system access. Here's the manual audit checklist we run before any skill touches our agency machines.
How to configure the OpenCode Controller skill to route cheap tasks to Kimi K2 and complex work to Opus. Real token cost math from three months of agency production traffic.
You don't have to pick. Here's how we run OpenClaw and Cursor side-by-side — port config, model routing, and which tool to reach for per task.
Concrete sub-agent patterns from agency production work — task decomposition, parallel fan-out, review-and-merge, and when not to spawn another agent.
OpenCloud lets OpenClaw sub-agents run on cloud instances while your machine is off. Early architecture, who it's for, and where the friction points already are.
Step-by-step install and config for the opencode-controller skill from ClawHub. Switch providers, route models per task, and manage sessions without leaving OpenClaw.
How OpenClaw orchestrates OpenCode's free frontier models locally — including the opencode-controller skill from ClawHub for routing models per task. Production-grade agent workflows with zero ongoing costs.
GLM5 from ZAI is the first open-weight model that genuinely threatens frontier labs. Production-grade performance at 1/5 to 1/6 the cost with MIT licensing and 30% hallucination rate.
NVIDIA and researchers just dropped a bombshell. 25 FPS video object removal without training. Here's why this matters for builders.