Do you need an AI assistant or workflow automation?
An AI assistant helps a person do variable knowledge work. Workflow automation changes how a repeatable process moves across triggers, rules and systems.
Assistant and automation solve different shapes of problem.
An assistant gives one person leverage. Automation changes a process. The mistake is using an assistant as permanent glue over a broken workflow, or forcing highly variable work into rigid rules.
The work varies but one person keeps doing the same kinds of thinking.
- Drafting emails, proposals or internal notes
- Research and meeting preparation
- Summarizing documents or threads
- Retrieving internal knowledge
- Turning notes into follow-up tasks
- Helping one person stay on top of recurring small work
The workflow follows repeatable events and rules.
- A lead should create a record and trigger an acknowledgment
- An overdue invoice should enter a reminder cadence
- Completed work should trigger a review request
- Data should move between systems
- Missing approvals should trigger reminders or escalation
- A daily exception summary should assemble from several sources
How to choose.
Name the trigger
If there is no clear event that starts the work and a person decides when to begin, an assistant may fit better.
Count the judgments
High-judgment work favors an assistant or human review. Stable rules favor automation.
Find system handoffs
If most friction is moving data, creating records or checking statuses, automation is usually stronger.
Look for one-person dependence
If one person's research, drafting and preparation is the bottleneck, an assistant may create leverage.
Check for permanent glue
If the assistant would spend all day compensating for the same broken handoff, fix the workflow instead.
Combine when useful
Automation can collect context and create a task; an assistant can help the person handle the exception.
Common wrong turns.
Assistant as bandage
Asking an assistant to copy data between systems all day is usually an integration problem.
Automation as straitjacket
Highly variable work gets forced through rigid rules and people spend time undoing it.
Too much autonomy
The system is allowed to send, change or commit before the team has confidence in the workflow and guardrails.
Measure the work, not “AI usage.”
- For an assistant: time spent drafting, researching, preparing and searching for context
- For automation: manual touches, handoff delays, re-entry and exception rate
- For either: correction rate, escalation rate and time removed from the specific workflow
Sometimes the process should simply disappear.
If a task exists only because of an outdated policy, redundant spreadsheet or unused report, deleting the step is better than automating it.
Start with the problem, not the product.
Related guide: Daily business briefing.