Tips & best practices

What this page is for

The agents themselves are powerful. But the teams that get the most value out of them follow a few patterns that take a week to learn and pay back for years.

This page collects those patterns. Skim the headers — pick the ones that match where your team is today.

Patterns for individuals

1. Lead with the outcome, never with the topic

The single biggest unlock is starting your brief with what you want to ship, not what you want to think about.

Weak: “Let’s talk about the customer call yesterday.”

Strong: “Write a five-bullet follow-up email recapping the customer call yesterday and naming what we owe them by Friday.”

2. Name the audience explicitly

Every audience changes tone, length, and assumed knowledge. The agent does its best work when you tell it who the reader is.

“For our exec team” — short, confident, decision-oriented.

“For a new customer’s onboarding manager” — friendlier, more explanatory, less jargon.

“For our engineering leads in standup” — terse, action-oriented, no preamble.

3. State exclusions, not just inclusions

What the agent shouldn’t do matters as much as what it should.

• “Don’t mention launch dates that aren’t confirmed.”
• “Don’t include the competitor’s name in the public-facing version.”
• “Skip everything before the third agenda item — that’s housekeeping.”

Without exclusions, you’ll spend a turn or two removing things the agent assumed you wanted.

4. Refine in place — never restart

A first draft you don’t love is rarely an agent problem. It’s almost always a brief problem. Refine: tell the agent what to change, one change at a time. Three small revisions usually land you somewhere great.

5. Use real sources, not vibes

Attach the file. Paste the transcript. Connect the app. Agents are dramatically better when they’re reading something specific than when they’re guessing at “what a release note usually looks like.”

Patterns for teams

1. Build a shared library of winning briefs

The first time a teammate writes a great brief for a Monday update, save it. Mark it for the team. From that day on, every Monday update starts from a known-good prompt instead of a blank page.

A useful saved brief includes:

• The outcome it produces — what shape of output comes out.
• The agent it pairs with — different agents respond differently.
• The sources it expects — what to attach or connect for it to work.
• A worked example — a real run that landed well.

2. Standardize the recurring stuff first

The fastest wins come from the most boring tasks. Look across your team for any of these:

• Weekly or monthly status updates
• Release notes
• Sprint retrospectives
• Customer call follow-ups
• Stakeholder one-pagers
• Backlog refinements
• Competitor briefs
• Onboarding messages

Pick one. Build a great brief for it. Save it. Move to the next. In a quarter, you’ll have replaced a dozen recurring chores with reliable agent runs.

3. Designate a “prompt champion” per team

One person on each team becomes the brief curator: they write the first version, refine it through real use, and update it when the shape of the work changes. The role rotates quarterly. Quality compounds.

4. Pair agents with sources, not just prompts

A great brief gets better when paired with great sources. Keep your connected apps fresh:

• Refresh Confluence and Google Drive connections when permissions change.
• Add the team’s working folder when new projects start.
• Disconnect sources that have gone stale — they hurt more than they help.

See VDF AI Data for the deeper story on sources.

5. Measure the time you save

You’re not building habit if you can’t see the win. Once a quarter, list the recurring tasks your team automated with agents. Estimate the time saved per run × the number of runs. Share the number.

Teams that share their savings expand usage faster than teams that don’t.

Patterns for organizations

1. Run a “brief audit” twice a year

Once every six months, look at your saved-brief library. Some briefs will have drifted — the audience changed, the format changed, the source apps changed. Refresh them. Retire the dead ones.

2. Map agents to job families, not job titles

A “release note drafter” is a job family pattern — engineering managers, product managers, and tech leads all use it. Map agents to the work, not the title. You’ll see broader adoption with fewer agents in the catalog.

3. Let agents draft, let people decide

Agents are great at drafting. They are not great at deciding for your company. Use them to remove the first-90% friction; keep humans on the last-10% judgment.

This is also how you build trust: every output gets read by a person before it ships. People who trust the output expand usage. People who don’t, won’t.

4. Combine agents into networks when work has clear stages

If a task always goes through a sequence — research → draft → critique → final — that’s not three agent runs. That’s a workflow. Build it once in VDF AI Networks and run it forever.

5. Choose between picking from the library and building your own

When a recurring task starts feeling slightly off — close, but not quite right for your team — that’s the moment to decide.

• Customize a library agent when the existing one is close. Most adjustments to tone, audience, and constraints work as customization on top of a library template. See Agent library.
• Build your own when the job is genuinely yours — a deliverable specific to your team that no library agent targets. See Creating your own agent.
• Build a network when the job has multiple stages that always happen in the same order — not a single agent at all. See VDF AI Networks.

A useful test: read the agent library quickly. If the first three minutes don’t surface something close, your job is probably specific enough to deserve its own agent.

6. Treat the brief library as a real internal product

Once your team has 20+ saved briefs, treat the library like a product:

• It needs an owner.
• It needs a backlog.
• It needs feedback loops with the people who use it.
• It needs to be discoverable.

The teams that take this seriously turn agents into a durable advantage.

Anti-patterns to avoid

• Asking for “more detail” without saying which detail. Specify what’s missing.
• Combining three asks in one brief. One outcome per brief.
• Skipping the audience. “Just summarize this” produces tone-flat results.
• Trusting outputs without reading them. Agents are confident; that doesn’t make them right. Read before you ship.
• Saving every prompt as a template. Only save the ones that produced great results — otherwise the library becomes noise.
• Treating agents as a way to skip the thinking. They remove drafting friction, not strategic judgment.

A weekly habit that compounds

Once a week, ten minutes:

1. Look back at your agent runs.
2. Pick the one that went best.
3. Save the brief.
4. Pick the one that went worst.
5. Rewrite the brief so it would have gone better — and save it too.

Over a year, you’ll have ~50 great briefs and 50 lessons. Your individual output will be measurably faster, and the patterns will quietly spread to the people you work with.

Where to go next

• Use cases — worked scenarios you can copy.
• Working with assistants — the brief-context-output loop in detail.
• Agent library — the pre-built starting points.
• Creating your own agent — when no library agent fits.
• VDF AI Networks — when your work needs several specialists in sequence.