You don’t need AI to automate your workflows. A bash script and a cron job will handle 80% of what most people use AI automation for. But for the remaining 20% — the tasks that require understanding, interpretation, and judgment — AI transforms automation from “follow these exact steps” to “figure out what needs to happen and do it.”
Here’s the honest breakdown of what AI adds to workflow automation, what it doesn’t, and where the hype exceeds reality.
What AI Automation Actually Does Better
Handles ambiguous input. Traditional automation: “If email subject contains ‘invoice’, route to accounting.” This fails when the subject says “billing question” or “payment issue” or “can you check our account?” AI automation: “If the email is about financial matters, route to accounting.” The AI understands that “payment issue” and “billing question” are financial matters even though they don’t contain the word “invoice.”
Adapts to variations. Traditional automation breaks when input format changes. Someone sends a date as “March 15” instead of “2024-03-15” and the parser fails. AI handles format variations naturally because it understands the content, not just the pattern.
Generates rather than transforms. Traditional automation moves data from A to B, maybe reformatting along the way. AI automation can generate new content: summarize a document, draft a response, create a report from raw data. This is a fundamentally different capability.
Makes judgment calls. “Is this customer email urgent or routine?” Traditional automation needs explicit rules (keywords, sender lists). AI can assess urgency from context: a regular customer asking about a delayed order is more urgent than a prospective customer asking for general information.
What AI Automation Does Worse
Deterministic operations. “Move file from folder A to folder B.” AI adds nothing here. A simple script is faster, cheaper, and more reliable.
Math and data processing. “Calculate the sum of column B in this spreadsheet.” The AI might get it right, or it might hallucinate a number. A formula is always right. Use code for math.
High-frequency operations. If you need to process 10,000 items per hour, AI is too slow and too expensive. Traditional automation handles volume; AI handles complexity.
Safety-critical workflows. Financial transactions, medical alerts, security operations. AI’s probabilistic nature makes it unsuitable as the sole decision-maker in high-stakes scenarios. Use AI for recommendations, not for final decisions.
The Practical Architecture
The most effective automation architectures combine traditional and AI approaches:
Traditional layer handles the plumbing. Triggers, scheduling, data movement, API calls, file operations. These are deterministic tasks that should be handled deterministically.
AI layer handles the thinking. Classification, generation, summarization, interpretation. These are cognitive tasks that benefit from AI’s natural language understanding.
Human layer handles the judgment. Approvals, edge cases, sensitive decisions. Some things should have a human in the loop regardless of how good the AI is.
My morning briefing workflow illustrates this:
– Cron job triggers at 8 AM (traditional)
– Scripts fetch email count, calendar events, server metrics (traditional)
– AI summarizes the data into a natural-language briefing (AI)
– AI identifies anything that needs attention and explains why (AI)
– Briefing is posted to Slack (traditional)
– I review and take action on flagged items (human)
Each layer does what it’s best at.
Getting Started
If you’re new to AI workflow automation, start with one workflow that currently requires manual judgment:
1. Pick a task you do daily that involves reading, interpreting, and acting on information
2. Build the traditional automation part first (triggers, data fetching)
3. Add AI for the interpretation and generation parts
4. Keep a human review step for the first month
5. After a month, evaluate: is the AI’s judgment reliable enough to remove the human step?
Common first automations: email triage, daily status summaries, content categorization, customer inquiry routing. All of these combine traditional triggers with AI interpretation.
Don’t start with complex multi-step workflows. Start with one step that benefits from AI understanding, get it working reliably, then expand.
The Bottom Line
AI doesn’t replace traditional automation — it extends it. Traditional tools handle the reliable, deterministic operations. AI handles the ambiguous, context-dependent operations. Together, they cover ground that neither could handle alone.
The hype says “AI will automate everything.” The reality is more modest but still transformative: AI automates the 20% of tasks that traditional tools couldn’t handle, which happen to be the most time-consuming 20% because they required human judgment.
That’s not everything. But it’s a lot.
🕒 Last updated: · Originally published: December 23, 2025