AI Automation: The Biggest Challenge Is Not How, but What, Where, When, and How Much
Before automating a workflow, decide what to automate, where it matters, when the process is ready, and how much control AI should have.
AI is changing how engineering teams write code, test, and review changes. But just because we can automate something, does it mean we should?
The hard part is deciding what to automate, where it creates value, when the process is ready, and how much control AI should have. Then we need to check whether the result is better.
What to automate: Focus on problems, not just tasks#
Consider generating similar pieces of code and investigating why a system stopped working. AI can help with both, but the risks differ. The first is predictable and easy to check. The second requires context and judgment.
Good candidates for automation tend to happen often, follow clear steps, and produce results people can verify. But first ask: do we need to do this task at all? Making unnecessary work faster does not make it valuable.
Where to automate: Fix what slows everyone down#
Imagine a team that writes code twice as fast with AI, while changes still wait two days for review. Customers receive improvements no sooner.
Look at the whole journey: planning → designing → building → testing → reviewing → releasing → improving. Where does work get stuck? Where are people waiting, repeating steps, or correcting mistakes?
Automate where it removes a real obstacle, even if that is not the easiest place to introduce AI.
When to automate: Make sure the process is ready#
Suppose a company wants AI to prioritize technical problems, but teams disagree on what is urgent. Documentation and ownership are unclear. Automation may make that confusion worse.
Before introducing it, ask:
- What does a successful result look like?
- Is the information reliable?
- How will we recognize mistakes?
- Who is responsible when something goes wrong?
Sometimes the smartest first step is to organize information, clarify responsibilities, or simplify the process. A confusing process does not become a good one just because AI is involved.
How much to automate: Choose the right level of control#
Automation does not have to mean giving AI complete control. One useful way to think about it is:
- AI suggests. People make the decisions.
- AI prepares. A person reviews and approves the work.
- AI acts within limits. It completes defined tasks under agreed rules.
- AI operates independently. It makes and carries out decisions with limited human involvement.
For an online shop, AI might notice that the site is slowing down and recommend a fix. It might prepare a change for approval, or make the change itself. The right choice depends on the potential impact of a mistake. If thousands of customers could be affected, a person approving the action may be the better option.
The goal is to give AI the right amount of control, not the maximum amount.
How much value: Are we actually improving anything?#
Hours saved are not the whole story. If AI saves a team five hours a week, but people spend six hours checking and fixing its output, the work has not become easier.
Real value = benefits gained − setup and maintenance costs − the cost of mistakes and review.The result should simplify work, improve quality, or help the team deliver better outcomes beyond the first demonstration.
Final thought#
Successful AI adoption means choosing where automation helps, keeping human judgment in place, and giving people more time for problems that need their experience.
What is the biggest mistake you see teams make: automating the wrong work, introducing AI too early, or giving it too much control?
