Nate B. Jones uses an agent that mistakenly sent an insurance appeal to show why capable AI still needs explicit authorityLeast privilege gives a person or automated system only the access required for its current authorised work., durable memoryAI agent memory is stored information that an agent can retrieve and use across steps, sessions, or changing contexts. and visible approval boundaries. Better intent-following has lowered the technical barrier, but longer agent actions also make mistakes more consequential.
His proposed stack separates personal context from model providers. Open Brain stores memories, Open Skills captures reusable methodsAn AI agent skill is a reusable package of instructions, resources, and tool guidance for performing a bounded kind of work., and Open Engine makes tasks and agent handoffs visible. Claude, Codex or open-source models can supply the intelligence while the user keeps control of accounts, secrets, permissions and final approvalHuman-in-the-loop AI keeps a person involved in reviewing, correcting, approving, or guiding an AI system's work..
Jones recommends starting with one repeated pain point, writing down the context that changes the answer and asking an agent to build the technical middle. The goal is not a universal assistant. It is a transparent loop whose memory and standards remain portableInteroperability is the ability of different systems to exchange information and use it with compatible meaning and behavior. when models, providers or access rules change.
Watch on YouTube
