Automation & Agents

Automation has been quietly running businesses for years — the rule that files an email into a folder, the workflow that routes an expense for approval. What’s new is that AI can now handle messier, less predictable steps, and that agents can chain several steps together to pursue a goal. That’s powerful, but it raises an old question in a sharper form: which work should we hand to a machine, and which still needs a person? Getting that line right is where most of the value, and most of the risk, lives.

Two kinds of automation

It helps to separate two ideas. Rule-based automation follows fixed instructions you set in advance: if an invoice is under a set amount and matches a purchase order, then approve it. It’s predictable and easy to audit because it does exactly what the rules say. An AI agent is different: you give it a goal and some tools, and it decides the steps itself — reading a request, looking something up, drafting a reply, maybe calling another system. That flexibility is useful for fuzzy tasks, but it also means the agent can take a path you didn’t foresee. The trade-off is real: rules give you predictability and easy auditing, while agents give you adaptability at the price of needing closer oversight, especially early on.

What’s safe to automate

A good rule of thumb: the better suited a task is to automation, the more it is high-volume, repetitive, rule-clear, and low-stakes when wrong. Sorting incoming tickets by topic, extracting figures from standard forms, drafting routine confirmations — these are strong candidates, because the patterns are stable and a mistake is cheap to catch and fix. The payoff is freeing people from drudgery so they can spend time on the parts of the job that need a human brain. A good early sign is a task people already do the same way every time and quietly resent — that consistency is exactly what makes it automatable.

Automate the typing, not the deciding. The more a task involves judgement, ethics, money, or someone’s wellbeing, the more a human should own the final call.

Where human judgment stays

Some tasks resist automation not because the technology can’t attempt them, but because the cost of being wrong is too high to leave unchecked. Approving a large refund, making a hiring decision, interpreting an unusual contract clause, handling an upset customer in a delicate situation — these involve context, fairness, and accountability that a model doesn’t carry. AI can assist here by gathering information or drafting options, but a person should make the decision and answer for it. The aim isn’t to remove humans; it’s to put their attention where it’s worth the most.

Designing the handoff

The practical pattern is human-in-the-loop: let the AI or automation do the legwork, then route the result to a person at the moment that matters. An agent might draft a customer refund and prepare the paperwork, but pause for a manager to approve anything above a threshold. This gives you the speed of automation with a safety gate on the decisions that carry real consequences. Set the threshold deliberately — too low and people are swamped with approvals, too high and risky actions slip through unseen.

Spot it: automate, assist, or keep human?

Read each situation and decide for yourself, then tap a card to flip it and check your answer.

Sort the steps

Drag each workflow step into the bucket it belongs to — or tap an item, then tap a bucket. Hit Check placement when you’re done.

How to use it

Look at one workflow you own and split it into steps. Mark each step as safe to automate (repetitive, rule-clear, low-stakes) or needs a person (judgement, money, fairness, sensitive). Then design the handoff so AI handles the first kind and routes the second to a human. Ask questions like: “What’s the worst outcome if this step runs wrong unattended?” and “Where should a person approve before anything irreversible happens?” Done well, you get faster routine work and humans focused exactly where their judgment counts.