AI and agents

Getting great results from AI

How to ask so orchit builds and answers exactly what you mean.


orchit's AI builds workspaces, modules, forms and plans from plain words, and it answers from your real data. A few habits get you what you mean the first time.

Describe the work, not the software

Say who it's for, what moves through it, and where it ends:

  • Vague: "Make a CRM."
  • Clear: "A workspace for our three-person agency to track leads from first call to signed contract, run client projects, and send monthly invoices."

The second gives Orchit AI enough to choose the right modules, lists, card types and process.

Name what must be tracked

If a detail matters, say it: "Each lead needs a budget, a source and a decision date." Those become fields. Anything you leave out is filled in from what similar teams use, and orchit tells you what it assumed.

Answer its questions, or let it guess

For open-ended requests, Orchit AI asks a few questions first. Answer them, or let it go ahead with its best guess. Either way, you see the plan before anything is built.

Read the plan before you build

Whether it's a workspace, a module, a list or a single card, orchit shows you a draft first. Look closely at what it says it's not covering and what it assumed, and correct it in plain words: "Drop invoicing, add a content calendar."

Ask the right agent

You want to... Ask
Think, write or research for yourself Mind
Know about or change one module That module's Main tab
Know about the whole workspace Peak
Ask without leaving the page Ask AI (✨)

An agent in a module already knows that module's cards, people, forms and process, so you don't need to explain them.

Be specific about people, dates and amounts

"Assign the three overdue bugs to Priya and move them into next sprint" works better than "sort out the bugs". When something's missing, the agent asks before it acts.

Choose the output

For a document you'll send, ask for a Word file or an artifact. For a quick answer, leave the output on Auto. In Mind, switch to Think for hard problems, or Plan for a step-by-step plan.

Give it your knowledge

Agents answer best from your own documents. Add your handbook, product docs or policies to a knowledge base, link it to the agent, and ask.

Check what matters

AI can make mistakes. Check names, numbers and dates before you send or rely on something important. Every change an agent makes is listed in its Activity, with undo.