How I Tell If a Task Should Be Automated, Agented, or Manual
I share the three‑question test I use to decide whether a task belongs in code, an AI agent, or stays human, so you can stop guessing and start building.
When I first walked into a client’s warehouse and saw a spreadsheet that had been manually copied into a ticketing system every night, I felt the same mix of frustration and curiosity that drives every operator. The spreadsheet existed because someone had decided that the task was too cheap to code, too noisy for an AI, and too routine to keep a person’s attention. I asked myself whether the effort was truly irreducible or whether we were just paying for the status quo. That moment sparked the framework I now use on every engagement: a three‑question test that tells me if the work belongs in a piece of software, in an autonomous agent, or stays in the hands of a human. The test is simple, but applying it consistently saves weeks of rework and eliminates the hidden cost of guessing.
The hidden cost of guessing
Most owner‑operators treat automation like a lottery ticket. They buy a SaaS tool, hope it will cover a pain point, and later discover that the tool only fits 70 percent of the workflow. The remaining 30 percent becomes a patchwork of manual steps, workarounds, and endless support tickets. That friction is not just a time sink; it erodes morale and inflates the cost of ownership because you keep paying for a product that never truly belongs to you. In the rebuild I led for a 200‑employee business, we replaced 21 SaaS tools with a single operating system that the owners could modify themselves. The difference was not the technology stack but the discipline of asking the right question before any line of code or model was written.
Three questions that decide the path
The test I rely on reduces every task to three binary answers. First, does the task have a well‑defined, repeatable rule set that can be expressed in code without ambiguity? Second, does the task require context, natural language understanding, or the ability to negotiate with external parties? Third, does the task involve judgment, empathy, or unpredictable variables that only a human can resolve? If the answer is yes to the first and no to the others, automation wins. If the answer is yes to the second and no to the others, an AI agent is the right fit. If the answer is yes to the third, keep it manual.
- Is there a deterministic rule set?
- Does the task need natural language or external negotiation?
- Does the task demand human judgment or empathy?
Automation: code that never sleeps
When a task can be reduced to a deterministic rule set, the payoff is immediate. Code runs 24/7, incurs no licensing fees, and can be versioned alongside the rest of the business’s operating system. In practice, this means pulling data from a sensor, applying a threshold, and triggering an alert—all without a human pressing a button. The danger is over‑engineering; you can spend months building a micro‑service for a rule that could be expressed in a spreadsheet. The key is to keep the implementation as thin as possible and to embed it directly in the core system you already own. That way the automation becomes part of the business’s DNA rather than an external add‑on that must be patched every quarter.