Great results from Copilot & agents

Working with AI is a skill, and like any skill it has a few simple moves that separate “this is magic” from “this is useless.” The good news: the moves are easy, and you’ll have them within a day. This page is written assuming you may be using AI seriously for the first time — so nothing is taken for granted, and no question is too basic.

If you’d like to know what’s happening behind the scenes first, How AI reads your workspace sets the stage. Otherwise, dive in.

The right mental model

Picture the AI as a brilliant, fast, eager new teammate who started this morning. It’s read an enormous amount, it never gets tired, and it genuinely wants to help — but it’s literal, and it only knows what you tell it or point it at. It won’t roll its eyes at an obvious question, and it won’t read your mind either.

That single image explains almost everything: clear, specific asks get great results; vague asks get vague results. You’re not “programming” anything. You’re briefing a capable colleague.

There are no dumb questions

“Explain this database to me like I’ve never seen it.” “What did we decide in this thread?” “Is this email too harsh?” All perfectly good asks. The AI would much rather you ask plainly than struggle in silence.

Be specific — about the goal and the output

The most common reason an answer disappoints is that the request was broader than the asker realized. Say what you want, and say what you want it to look like.

Compare:

Weak:   write something about the launch
Strong: write a 3-sentence announcement for our #general channel about
        the v2.4 launch, friendly tone, mention the new dashboard

The strong version names the format (3 sentences), the audience (#general), the tone (friendly), and the key point (the new dashboard). You’ll get something usable on the first try — and if you don’t, you’re now in a great position to nudge it.

A handy checklist when an ask feels too vague: What exactly? For whom? How long? What tone? What must it include?

Give it the right context

The AI is much smarter when it can see the thing you’re talking about. Because it works by retrieval, pointing it at the right material is half the battle.

• Ask where the work is. Mention the Bridge copilot inside the relevant chat, task, or page, so “summarize this” has an obvious this.
• Be explicit about the source. “Draft release notes from this page.” “Pull action items from this thread.” “Compare these two database views.”
• Bring the facts to it. If something it needs isn’t written down anywhere in your workspace, paste it in. It can’t use what doesn’t exist.

Iterate — don’t start over

Here’s the move that feels unnatural at first but changes everything: when the answer is close but not right, don’t rewrite your whole request — just nudge it. The AI remembers the conversation, so you can steer in small steps:

You: draft a project update from this task list
AI:  (writes a solid first draft)
You: shorter, and lead with the risks
AI:  (revises)
You: perfect — now turn the action items into tasks

Each follow-up builds on the last. This back-and-forth is normal and expected — it’s a conversation, not a slot machine. Most great results come from the second or third turn, not the first.

Stay the editor

The AI is confident — even when it’s wrong. That’s not a flaw to fear, it’s just a fact to plan around: you remain the editor. Skim what it produces, check anything that matters (numbers, names, claims, anything you’ll send to a customer), and keep what’s good.

Trust, then verify

For low-stakes drafting, let it run. For anything that leaves your team — an external email, a published doc, a financial figure — give it a human read before it goes out. The AI is a wonderful first draft and a poor final approver.

This is also why agents ask before doing anything sensitive — the system is built to keep you in the loop on exactly these moments.

Keep a conversation to one topic

Each conversation builds up context. That’s helpful while you’re on one topic — and unhelpful when you switch. If you’ve been refining release notes and now want to plan next quarter, start a fresh conversation. It keeps the AI focused and its answers sharp.

(Behind the scenes, very long conversations get summarized automatically so things stay fast — but a clean start for a genuinely new topic still beats one endless thread.)

Copilot or agent — which do I use?

Both are AI, but they’re for different jobs:

Use the Bridge copilot when…	Build an agent when…
You want quick, ad-hoc help right now	You have a repeatable role to fill
The task is general (summarize, draft, explain)	The job needs specific knowledge, tools, or rules
You don’t want to set anything up	You want it to work tasks or sit in a channel ongoing

Rule of thumb: reach for the copilot for a moment’s help; build an agent when you find yourself asking for the same kind of help over and over.

Good first things to try

If you’re not sure where to begin, ask the copilot to:

• Summarize a long thread or meeting into a few bullet points.
• Extract action items from a discussion and turn them into tasks.
• Draft a first version of a doc, message, or announcement.
• Explain a database, a page, or “what we decided” in plain language.
• Reword something — shorter, friendlier, more formal.

Each one takes ten seconds to ask and shows you, immediately, what good context and a clear request can do.

Related

• How AI reads your workspace — why context and clarity matter so much.
• Knowledge that makes agents smart — set agents up to perform like experts.
• Copilot and Agents — the two ways AI shows up.
• How an agent works a task — what happens when you hand AI a real job.