"What can you automate with AI?" and "what should you automate with AI?" sound like the same question. They're not, and the gap between them is where a lot of automation projects go wrong. Almost anything repetitive can technically be automated now. Far fewer things should be — at least not yet, and not without real thought about what breaks if the automation is wrong.
What can be automated — the honest, current answer
Drafting: emails, content, first-pass proposals, summaries of long documents. Research: gathering information, comparing options, pulling together a first look at a market or competitor. Routine communication: follow-ups, scheduling, standard responses to common questions. First-pass analysis: sorting through data, flagging what looks unusual, preparing a summary for a human to review. If a task is repetitive, has a knowable "good output" you'd recognize on sight, and doesn't require judgment specific to your business, it's very likely automatable today.
What should be automated — a different filter entirely
Ask three questions before automating anything. First: what happens if this goes wrong while no one's watching? A miscategorized expense is a minor annoyance. An automated message sent to the wrong customer at the wrong moment is a real problem. Weight your caution accordingly. Second: does this task carry your specific judgment, or your relationship with someone? If a client feels the difference between you and a generic response, that's a strong signal to keep it human, at least for the final pass. Third: how often does this actually happen? A task you do twice a year rarely justifies the setup time an automation requires — automate the things you do weekly, not the things you do occasionally and can just handle by hand.
The trap of automating too early
It's tempting to automate a process the moment you notice it's repetitive. Resist that instinct until you've done the task manually enough times to actually understand its edge cases. An automation built on your first three attempts at a task usually breaks on the fourth, because you didn't yet know what "normal" looked like. Do it by hand for a while, notice the pattern, then automate the pattern — not your first guess at what the pattern might be.
Where to actually start
Pick the task that's both frequent and low-stakes if it goes slightly wrong — that's your safest, highest-value first automation. Build it, run it alongside your manual process for a week, and compare the outputs honestly before you trust it fully. Only after that trial period should it run unsupervised.
How this connects to running the whole business
The businesses that automate well don't do it task by task with no plan — they treat it as an ongoing practice: notice what's repetitive, decide honestly whether it should be automated, build it carefully, and check it periodically as the business changes. That's exactly the kind of standing context Cameron is built to hold — not just executing individual automations, but tracking which parts of your business are automated, which aren't, and where the next opportunity actually is.
Not sure where to start automating? Bring the question to Cameron — it'll help you find the highest-leverage place to begin. Try Cameron for free.
Ready to put this into action? Waymaker helps you go from idea to your first paying customer, with AI doing the heavy lifting alongside you.
Start building free