We let AI modify 13,000 lines of Ansible. Here's what broke.

Bigger isn't better anymore. Small language models are handling production workloads teams thought required ten times the size, at a fraction of the cost. Open source controls the entire stack making it possible. This is where AI is actually heading.

WHY WE ❤️ IT: The inference cost for GPT-3.5-level performance dropped 280x in two years. This post explains why smaller, open source models are winning in production and honestly, the engineering case here is impossible to ignore.

Can AI actually handle real system administration? Learn Linux TV ran a legit experiment by letting Claude Code modify a 13,000-line Ansible automation system over several weeks. Discover what worked surprisingly well, what failed spectacularly, and whether AI can genuinely accelerate sysadmin work without creating disasters.

WHY WE ❤️ IT: Jay catches Claude Code red-handed trying to create duplicate playbooks, proving why AI needs constant code review even when it's surprisingly competent.

The agentic AI conversation has moved past GPUs. At the MCP Dev Summit, 146 organizations and 95 sessions made one thing clear: the infrastructure layer of agentic AI is being defined in the open, right now. Danielle Cook was in the room. Here's what she saw.

WHY WE ❤️ IT: This isn't another "PHP is dead" hot take. It's a data-driven look at PHP's real challenge with thriving adoption but a talent pipeline problem that could shape the platform's next decade.

Fork off 10 git subtrees, lay down meta prompts, and spawn parallel AI agents that review PRs and write tests. Calvin Hendryx-Parker shares how day one deployments tackle deployment, secrets, and configuration all at once, why Goose excels at customer journeys and specifications, and how senior developers guide AI with a surgeon's hand.

WHY WE ❤️ IT: Calvin's git subtrees workflow is brilliantly practical: Fork off a subtree for each ticket, drop in a meta prompt, and let AI agents handle the PR reviews and tests in parallel.