Productive AI can search your data, take actions, run on a schedule, and carry out complex workflows — but the quality of what it does depends heavily on how you communicate with it.
This article covers principles for working with AI in Productive. They apply whether you're chatting with the Assistant, using AI Search, or writing instructions for Skills and Agents.
📌 AI-generated content may be inaccurate or incomplete. Always review results before acting on them, especially when the Assistant or an Agent is creating records, updating fields, or posting to tasks.
👉 New to Productive AI? Start with Productive AI: A Practical Guide.
Delegate the Right Work
Not every task benefits from AI, and knowing when to use it is the first step to getting value from it.
AI in Productive works best when:
The task is repetitive or time-consuming but not deeply creative or judgment-based.
You can describe the goal clearly.
The output is something you'd recognize if it were wrong (a report, a task list, a doc draft).
If a task depends on context only you have, such as deciding a project's priorities or evaluating a client relationship, an AI can help you think through it, but the decision shouldn't be fully delegated.
Ask the AI:
"Show me who on the team has availability for booking next week across all active projects."
Decide for yourself:
"Book the right people on the new project."
Describe Clearly
The AI responds to what you say, not what you mean. Vague prompts produce vague results.
A good description includes:
The goal — what you want the AI to produce or do.
The context — which project, task, doc, or time frame is relevant.
The output format — a list, a table, an artifact, or another structure that fits your needs.
Any constraints — tone, length, what to include or skip.
You don't need all four every time, but the more specific you are, the less back-and-forth you'll need.
Vague:
"Summarize the budget I am working on."
More specific:
"Show me the current budget status for the Acme project: planned vs. spent, and flag any service that's over budget. Present it as a table."
Review and Refine
Working with AI is a process, not a single exchange. Expect to review, correct, and iterate — especially early on when you're still learning what works for your use case.
Review what AI gives you before acting on it, especially when it is creating or updating records or modifying data. If the result isn't right, don't start over; follow up in the same conversation, or go back and refine the skill or agent instructions.
Within a session, the AI keeps track of what's been said, so you can steer it without repeating yourself.
Instead of starting over, you can write:
"That's close. Remove the budget section and add a line about the client contact."
Writing Instructions for Skills and Agents
Skills and Agents both run on instructions you write. But they work differently, and what makes a good instruction differs between them.
Skills
A skill is a reusable instruction set you apply to the conversation with the AI Assistant or assign to an Agent. It shapes how the AI responds in a specific context, for example, a tone guideline, a process to follow, or a format to use.
The same principles we mentioned apply when writing skill instructions. A few things that work especially well for skills:
Be explicit about format — instead of "summarize this," write "summarize in three bullet points: key decisions, action items, and open questions."
Set the tone — if the output has an audience, say so: "write in a confident, professional tone — no jargon, no filler."
Add constraints — Constraints help keep outputs focused: "keep responses under 150 words" or "always respond in the language of the input."
Think in workflows, not just prompts — The best skills encode a multi-step process, not just a single instruction. If you find yourself repeating the same sequence of steps in chat, that's a skill waiting to be written.
Example skill instruction:
"When writing any client-facing message or comment, use a confident and professional tone. Avoid internal jargon. Keep the message concise, no more than three short paragraphs. Always end with a clear next step or question."
👉 For more on creating and managing Skills, see What Are Skills?
👉 For more tips and examples on writing skill instructions, see Working with Skills.
Agents
An Agent acts autonomously — it runs on a schedule, without you prompting it each time.
Because you won't be there to course-correct in real time, the instructions need to be explicit about scope, conditions, and limits.
Good Agent instructions:
Define the scope — which projects, tasks, or records the Agent should work with.
Describe the condition — what situation should prompt further action.
Specify the action — exactly what to do, and what not to do.
Example agent instruction:
"Review all open tasks in projects I manage. If a task has been in "Review" status for more than 5 days without a comment, post a follow-up comment that mentions the assignee and asks for a status update. Do not change task status or assignees."
📌 You can assign Skills to an Agent to give it more specific instructions for how it acts and what it produces. For example:
Assign a client communication skill to an agent that posts comments on client-facing tasks, so it always uses the right tone and format.
Assign a status update skill to an agent that summarizes project progress, so reports follow a consistent structure.
Assign a task naming skill to an agent that creates tasks automatically, so new tasks always follow your team's naming conventions.
👉 For more on Agents, see Setting Up and Managing AI Agents.