A demo without a job
Impressive in a meeting and unused a month later, because it was never attached to work anyone owns.
AvoidableKinds of work where AI is worth building in, grouped by the three questions we ask first. Some are built into Lina or Lina Terminal. The rest are what we build to order.
Patterns to start a conversation from. None of them is a case study.
The model is rarely the problem. Each one comes from skipping one of the three questions.
Impressive in a meeting and unused a month later, because it was never attached to work anyone owns.
AvoidableAnother tab and another place to paste context, so whatever time it saves is spent switching to it.
AvoidableEvery request sent to the largest model available, whether the task needed it or not.
AvoidableFind the steps that need judgement, build the model into the application where they happen, and size it to what the work is worth. Pick a question to see what that looks like.
Which steps need a model, and which need a rule.
Most of the value is decided before anything is built. These are the situations where the first job is working out what the AI is for.
A team that has tried AI tools and has a demo or two to show for it.
Nothing is attached to a workflow anyone owns, so usage fades while the subscriptions stay.
Map how the team works today, pick the one use case worth shipping, and decide whether to build it, buy it or leave it alone.
A business with a long list of places AI might help, and no way to rank them.
Every idea looks promising on its own, and building them all would cost more than they return.
Weigh each use case by how often the work happens, what it costs today and how bad a mistake would be. Then start with the one that clears all three.
An operations team asked to “add AI” to a process that runs the same way every time.
A model would add cost and unpredictability to work that a rule already describes.
Automate it with ordinary software, and keep AI for any step that needs judgement.
Tell us how it runs today and where it gets stuck. If none of it needs AI, we’ll say so.
Start a conversation