[
  {
    "id": "a-model-answers-an-agent-acts",
    "url": "https://oneninenordic.com/perspectives/a-model-answers-an-agent-acts",
    "markdown": "https://oneninenordic.com/perspectives/a-model-answers-an-agent-acts.md",
    "title": "A model answers. An agent can act.",
    "summary": "The one distinction leadership teams need before they decide anything about AI.",
    "date": "2026-10-21T00:00:00.000Z",
    "author": "Jesper Andersen",
    "body": "An AI model: you provide the input, the model generates a response. It does nothing on its own. Everything stops when you stop.\n\nAn AI agent: you provide the goal. The agent uses tools and rules to carry the work forward. Model + tools + rules = an agent that can work within defined boundaries.\n\nStart with the task, not the technology. Agree what AI may do and who checks the results."
  },
  {
    "id": "cloud-or-on-premises",
    "url": "https://oneninenordic.com/perspectives/cloud-or-on-premises",
    "markdown": "https://oneninenordic.com/perspectives/cloud-or-on-premises.md",
    "title": "Cloud or on-premises?",
    "summary": "Choosing a model and choosing where to run it are separate decisions.",
    "date": "2026-10-21T00:00:00.000Z",
    "author": "Jesper Andersen",
    "body": "Cloud API: the provider runs the model. Fast setup, less operational work. Review data terms, retention, region and cost.\n\nPrivate cloud: your environment, more control over networks and data, option to host weights. You still own capacity, security and operations.\n\nOn-premises: the model runs on your hardware. Local control, can work offline. Requires hardware, skills and model maintenance.\n\nHybrid works. Open weights can run in the cloud or locally, and cloud does not automatically mean training on your data. GDPR requires appropriate agreements, processing and safeguards, not a particular location."
  },
  {
    "id": "fluent-output-is-not-evidence",
    "url": "https://oneninenordic.com/perspectives/fluent-output-is-not-evidence",
    "markdown": "https://oneninenordic.com/perspectives/fluent-output-is-not-evidence.md",
    "title": "Fluent output is not evidence",
    "summary": "Why adoption depends on judgement, not on how good the answer sounds.",
    "date": "2026-10-21T00:00:00.000Z",
    "author": "Jesper Andersen",
    "body": "Sustained use of AI depends on ownership, judgement and a useful daily routine.\n\nOwn the workflow: a process owner sets the outcome, the review standard and the escalation route, and makes AI part of an existing task with time to practise.\n\nBuild judgement: train on real tasks and failure cases. Check sources, assumptions and omissions. Fluent output is not evidence; review against the task and its context.\n\nLearn from use: track repeat use, accepted outputs, rework and time saved after checking. Use the feedback to improve guidance, retrain, or stop what does not work."
  }
]