AI Workflow Templates: Research, Meetings, and Writing

An AI workflow template defines what goes in, what the AI does with it, what you review, and when you stop and decide. Templates save time because you stop designing the same process each session and start from something that already works.

A template without a review step is a risk, not an asset. The value of a workflow is in knowing what a good output looks like and what should trigger a stop before you use the result. Tools like fenbs store your reusable prompts and the decisions behind each workflow so every session reads the current version.

What a useful AI workflow template includes

A minimal workflow template has four parts: the inputs the AI needs to start, the instruction for what to produce, the criteria you use to review the output, and the stopping rule — the condition under which a person must decide before the workflow continues.

• Inputs: the facts, documents, or context the AI needs before it starts.
• Instruction: what the AI should produce and in what format.
• Review criteria: what a good output looks like and what would make it inadequate.
• Stopping rule: when the AI should stop and surface the decision to a person rather than continuing.

The stopping rule is the most important and the most commonly skipped part. Without one, an AI will make a plausible-sounding choice at every fork rather than surfacing the decision to the person who should make it. See AI agent approval workflow for more on how to build approval steps into the design.

Research workflow template

Use this template when you need to understand a topic, a market, or a competitor before making a decision.

RESEARCH WORKFLOW

Inputs:
  The question to answer (one sentence)
  The decision this research will inform
  Any sources you already have and trust
  What would change the decision

Instruction to AI:
  "Research [question]. Focus on [most relevant aspect].
  List your sources. Flag anything you cannot verify.
  Do not state a conclusion as a fact without a source."

Review criteria:
  Does every factual claim have a source you can check?
  Are any claims from a roundup rather than a primary source?
  Does the output answer the actual question asked?

Stopping rule:
  If the research returns conflicting answers from credible
  sources, stop. Bring both to the decision-maker before
  continuing.

Save the sources alongside the output. A research note without its sources cannot be checked or reproduced. This principle also applies to the AI risk assessment template, where traceability matters as much as the conclusion.

Meeting workflow template

Use this template before and after a meeting to reduce preparation time and ensure follow-up actions are captured.

MEETING WORKFLOW

Before the meeting — inputs:
  Meeting goal (what decision or outcome do we need?)
  Attendees and their roles
  Relevant project context and recent decisions

Before the meeting — instruction:
  "Given this goal and context, list the three most important
  things to confirm in this meeting, and flag any open
  questions from the project record."

After the meeting — inputs:
  Notes or transcript from the meeting
  Decisions made and by whom
  What was not resolved

After the meeting — instruction:
  "From these notes, list: (1) decisions made and who made
  them, (2) tasks with owner and due date if given,
  (3) open questions still waiting on a person."

Review criteria:
  Does every decision have a named decider?
  Is every task specific enough to be actionable?
  Are open questions flagged, not resolved by inference?

Stopping rule:
  If a key decision from the notes is ambiguous, stop.
  Confirm with the attendees before recording it.

The after-meeting output can be pasted directly into a project tool. For how fenbs handles Decisions and rules and Where we left off from meetings, see what an AI brain is.

Writing workflow template

Use this template for any writing task where you want a draft to react to rather than a blank page.

WRITING WORKFLOW

Inputs:
  What you are writing (email, report section, proposal)
  Who it is for and what they need to feel or decide
  Constraints: length, tone, what to include or exclude
  Examples of writing in the right tone, if available

Instruction:
  "Draft [piece] for [audience], aiming to [goal].
  Use [tone]. Stay under [word limit].
  Include [required elements].
  Do not include [excluded content]."

Review criteria:
  Does it answer the reader's question in the first paragraph?
  Does every claim belong in this piece?
  Would the reader understand every sentence without
  specialist knowledge?

Stopping rule:
  If the piece requires a commitment, a promise, or a
  factual claim you cannot verify, stop and confirm
  before the draft continues.

Storing and reusing your templates

A workflow template stored only in your head gets forgotten or modified informally over time. A template in a document gets out of date when the process changes. A template stored where your AI assistant reads it at the start of each session is the version that actually runs.

When a template changes — because a review criterion was too loose, a stopping rule fired and you learned something, or the process itself changed — update the stored version and note why. "We added the source-check step after an unchecked claim in the research workflow" is a decision, not a comment. Record it as one so future sessions understand the rule.

Common mistakes in AI workflows

The most common mistake is skipping the review criteria. Without them, you have no basis for judging whether the AI's output is good. You end up either accepting everything or rejecting things based on vague dissatisfaction. Write the criteria before you run the workflow, not after you see the output.

The second common mistake is giving the AI too much context at once. A workflow that starts with "here is everything about the company" produces outputs that are general rather than targeted. Give it only what it needs for this specific task. If the AI asks for more context mid-task, that is often a sign the instruction was too broad.

The third is treating a template as permanent. A workflow that worked well six months ago may need updating when the process changed, the tool changed, or a near-miss revealed a gap. Templates need an owner: someone who updates them when they drift from how the work actually runs.

Testing a new template on a low-stakes task first helps before you rely on it for important work. A template that has never fired its stopping rule may simply have never been given a case where it should have stopped. Running it against a known-tricky example reveals whether the rule works.

How to do this with fenbs

1. Create a free account and open a project for your AI workflows.
2. Add each template to Skills and prompts — give it a name, the full text, and a one-sentence description of when to use it.
3. Record the decision behind any stopping rule or review criterion in Decisions and rules.
4. Connect your AI assistant from Integrations so it can list and read the current templates at the start of each session.
5. When a template is updated after a near-miss or a lesson, update the stored prompt and add a Lesson entry so the change is documented in one place.