https://thefrontierforge.com/blog Writing on production AI from TheFrontierForge. Sun, 26 Jul 2026 00:00:00 GMT https://thefrontierforge.com/blog/the-model-was-never-the-bottleneck https://thefrontierforge.com/blog/the-model-was-never-the-bottleneck Sun, 26 Jul 2026 00:00:00 GMT A better model is the first idea in the room and the wrong one. Four things stop a pilot reaching production, and none of them is the model. https://thefrontierforge.com/blog/ai-pilot-to-production https://thefrontierforge.com/blog/ai-pilot-to-production Sat, 25 Jul 2026 00:00:00 GMT Most AI pilots stall on missing mechanisms, not model quality. The five that get a workflow into production, and how to retrofit a stalled pilot. https://thefrontierforge.com/blog/is-95-percent-accuracy-good-enough https://thefrontierforge.com/blog/is-95-percent-accuracy-good-enough Fri, 24 Jul 2026 00:00:00 GMT No percentage is good enough on its own. An accuracy bar is a percentage, a unit, a held-out set of your own cases, and a name against it. https://thefrontierforge.com/blog/does-human-review-eat-the-savings https://thefrontierforge.com/blog/does-human-review-eat-the-savings Thu, 23 Jul 2026 00:00:00 GMT The break-even exception rate comes from three numbers you already have. The formula, the verification tax nobody counts, and when not to automate. https://thefrontierforge.com/blog/agent-preflight-checklist https://thefrontierforge.com/blog/agent-preflight-checklist Wed, 22 Jul 2026 00:00:00 GMT Seven checks an AI agent passes before it gets write access: bounded scope, approval gate, eval set, kill switch, audit trail, cost ceiling, rollback.