Insights, engineering deep-dives, and product updates from the IBERANT team.
The AI provider is the part you rent and swap. The connections from your own systems, behind your own OAuth, are the part you own. Build those once with MCP and the same tools work in Claude, ChatGPT, Copilot, or a model you run yourself.
AI ate the 80% of the job that was execution. The 20% that's left, judgment, design, knowing what not to build, is now the whole job, and the same human brain has to hold more of it. That work didn't get cheaper. It got rarer and more expensive.
Anthropic's June 15 billing split exposed what was already true: $200 Claude and Codex subscriptions subsidize agent workloads at 15–30x. The metered reality is $500–$2,000 a month for heavy users. Here is what the math actually says, what other tools are doing about it, and why Churro Coder embeds the native CLI instead of calling the SDK.
Ralph loops, Playwright MCP, log and database access — what does an AI coder agent actually need to ship production work without a human in the loop? A grounded look at the stack, the limits, and a worked autonomous bugfixing workflow.
AGENTS.md, CLAUDE.md, skills, workflows, OpenSpec — the conventions stack keeps growing, but how do you actually know your repo is legible to coding agents? A look at the agent-readiness tooling, what the ETH Zurich research really says, and the minimum path to a sensible maturity level for a small team.
Vibe coding without a harness ships hallucinations. Heavy spec frameworks slow small teams to a crawl. Here is why Spec-Driven Development matters in the AI era, where the sweet spot lives, and why OpenSpec is the lightweight harness most teams actually need.
AI is a multiplier, not a generator. Multiply zero by anything and you still get zero. Here is how senior architects at IBERANT drive AI to ship code worth shipping, and why the people complaining about AI slop are usually the ones producing it.
How IBERANT approaches software development in the era of AI — where human expertise and machine intelligence work as one to deliver at an entirely different speed.
ChurroStack is our open-source answer to the complexity of deploying and managing AI infrastructure on-premises — as straightforward as making churros.