Automate, augment, or strategize: how to sort every task in your business for AI

Most task lists don't fail because you're lazy. They fail because you treat every line the same way, and AI makes that worse. You point Claude at your whole business, get a wall of half-right output, and end up as the quality check on all of it. The fix is to sort each task before you hand it over. Every job in your business falls into one of three buckets: automate, augment, or strategize. Get the sort right and the agent does the boring work while you keep the calls that actually need you.

The three buckets, defined by who's in the loop

The buckets aren't about how "advanced" a task is. They're about how much judgment the task needs and who has to be in the loop when it runs.

Automate is work with a clear input, a clear output, and rules that don't really change. Support tickets that follow a pattern. CRM data entry. Intake and claims processing. Inbox triage. Calendar booking. Missed-call callbacks. These are the tasks an agent should own end to end, running without you watching. This is where agents win. Not someday demos, the boring queue work that keeps a business running.

Augment is work where you still make the call, but a draft saves you an hour. A first pass at a proposal. A research report you'll fact-check. A tricky customer reply you want to sign off on. The agent produces, you decide. You stay in the loop on purpose because the cost of a wrong answer is high or the judgment is yours to make.

Strategize is the thinking work. What to build next. Which market to enter. How to price. Here the agent is a sparring partner, not an operator. You use it to pressure-test a plan, list the risks you're not seeing, or model a decision. It never ships anything on its own.

Here's the quick test. Ask two questions of any task: How often do the rules change? and What breaks if the answer is wrong? Rules steady and low blast radius, automate it. Rules steady but high stakes, augment it. Rules shift with judgment every time, strategize it.

How to sort your actual task list in 20 minutes

Don't theorize. Pull up the work you did last week and sort it.

The task that shows up 30 times a week and takes four minutes each is worth more automated than the flashy project you do once a month. Frequency times minutes is your real ranking.

Manual sorting vs sorting with Claude

You can do this on paper, and you should the first time so you understand the buckets. But once you've got 40 or 50 tasks, Claude Code will sort faster and catch overlaps you'll miss.

Paste your list into Claude Code and ask it to sort each task A/U/S, explain the call in one line, and rank the A's by weekly frequency. Then ask it to write the first agent for your top automate task. That's the move newer Claude models made cheap enough for almost any business to run. With the right setup, Claude builds the workflow, executes it, and tests it before handing it back, so you stop being the quality check.

What "automate" looks like when it's real

The buckets aren't abstract. Take "Emily", a voice agent we built to run a law firm's phone line. Answering the phone is a textbook automate task: clear input, clear output, rules that hold steady. Emily booked 453 new clients in 8 months, handled 571 calls hello-to-booked, and hit a 96.5% self-serve rate with zero human help. All 176 after-hours calls got caught instead of going to voicemail. That's worth $10K to $25K a month back to the firm.

Nobody sat and watched Emily work. That's the whole point of the automate bucket. The tasks you sort into it should run without you in the room.

Common pitfalls when sorting

FAQ

What if a task is part automate, part augment?

Split it. The research pull is automate, the decision on what to do with the research is augment. Most "mixed" tasks are really two tasks wearing one name, and sorting them separately is the point of the audit.

How many tasks should I automate first?

One. Pick your highest-frequency automate task, ship one agent, and prove it works before you add the next. Trying to automate ten things at once is how people end up back in AI hell, overwhelmed and convinced AI isn't ready.

Do I need to write code to automate these?

No. The automate tasks most businesses start with, inbox triage, CRM updates, intake, callbacks, run on copy-paste Claude Code setups. You describe the workflow in plain language and the agent runs it.

How do I know the agent is actually working?

Pick the number the task exists to move before you launch. For Emily it was bookings and self-serve rate. For an inbox agent it might be response time. If the KPI moves and you're not touching the work, it's a real automate win.

Sort your last week into the three buckets, find your top automate task, and build one agent for it. If you want the copy-paste templates and the masterclass that walks through setting up Claude Code the right way, the free Claude Community is where builders and founders trade the exact setups that work.