What you should be using AI for

Most people have watched AI do something impressive in a chat window and then stalled on the obvious next question: how do I achieve the massive productivity unlock everyone is talking about? Watching Claude draft a decent email in ten seconds is easy. Knowing which parts of your week to outsource to AI is harder, but it's where the real efficiencies are.

The fastest way I know to get unstuck is to stop thinking of AI as software and start thinking of it as a hire. Specifically, a sharp intern. Once you have that frame, the delegation question gets much easier to answer. This guide helps you understand the type of work AI is good for, what questions to ask yourself, real-world examples of the work AI handles well, and what NOT to use AI for.

Treat AI like a talented intern

One of the most useful ways to start offloading more work to AI is to think of it as an intern or executive assistant with exceptional writing, organization, analytical (and coding) skills. You wouldn't hand an intern unfettered access to your personal finances or a decision that could sink the company. You also wouldn't hire one to do basic one-off tasks with no real long-term value add. You'd give them real, well-defined work and check it before it ships. Start by thinking about AI the same way.

That frame also clarifies the payoff. A good intern helps you in one of two ways: they 1) save you time on work you're already doing, or they 2) take on work you couldn't otherwise afford to staff. AI does both. It either gives you back hours in your week, or it lets you raise the quality and scope of your business without hiring for it.

The type of work AI is good for

AI churns through work that's repeatable and data-heavy. But it can also flex to work as a thought partner or coach, as long as you clearly explain what you're looking for. Below are the best types of work for AI:

Repetitive, templatized workflows with consistent rules. Basically, if you could write the rules or format down, AI can run them. Use AI for anything that follows the same steps every time: filling in forms, weekly summaries, scheduling and inbox management, answering common inquiries (FAQs), generating standard documents (content briefs, marketing materials, contracts) from a template.

Working with data. You shouldn't be manually copying and pasting data anymore, ever. You probably shouldn't even be building first drafts of financial models anymore either. Use AI for extracting, sorting, and organizing information, especially from PDFs; building spreadsheets and financial models; running calculations.

Compiling and summarizing information. This is, in my experience, one of the areas I've gotten the most out of AI, especially when the source material is scattered across multiple places — email threads, chat messages, documents, meeting notes. Pulling all this information into one coherent view is exactly the kind of tedious synthesis it does well, as is creating a project or database of information and querying it.

Organization. Turning a mess into a system — file structures, task lists, notes, trackers. If your problem is that everything lives somewhere different and nothing is sorted, AI is a good first pass.

Content creation and review. First drafts of documents, presentations, marketing copy, images, even simple apps. The key word here is draft — you're getting to a strong starting point fast, though the finished product almost always requires your review.

Refining your thinking. Brainstorming, pressure-testing an argument, finding the holes in a plan before a client does. AI is a useful thinking partner precisely because it's fast, available, and won't get tired of your fifth revision. You can also ask it to take on specific roles (your boss, a 3rd party) to help refine drafts before you finalize them.

Building and coding. You don't need to know how to code to build with AI anymore. Describe what you want in plain language and AI can write the script, build the simple app or tool, or automate the repetitive task — a spreadsheet macro, a one-off data cleanup, a small internal tool your team can reuse. You won't be shipping production software, but you can solve problems that used to call for a developer.

Still not sure? Ask yourself

1. Does it meet any of the above criteria? If it does, try AI.
2. Are you still learning by doing this yourself? Early on, doing the work is how you learn it. Don't outsource the reps that build the skill you sell. Once a task is rote for you — once you've stopped learning from it — it's safe to delegate.
3. How much do you hate doing this every day? Everyone has repeated tasks that they loathe. For me, it's managing my inbox and drafting summary emails. It's 100% worth the time to set up AI to handle this for me.
4. Does this involve copy and paste or toggling back between apps and screens? If so, AI can do it faster.

The one big test you should always ask yourself: Is your ownership of this task critical to its success? Things like your relationship with your biggest client, high-consequence decisions, and deep strategy work are your real value-add areas. This type of work really shouldn't be outsourced to AI, as tempting as it may be.

The combination matters more than any single answer. A task you don't need to own, where getting better wouldn't change much, and where you've stopped learning, is an easy hand-off — give it to AI. A task that's central to the business, where mastery compounds, and where you're still building the skill, stays with you. Most tasks fall in between, and that's the useful middle: hand AI the first draft or the grunt work, and keep the judgment, the final call, and the relationship for yourself.

Real-world examples: when to use AI

The categories above become concrete once you map them to real jobs. Here are use cases worth stealing — at work first, then at home. Not all of them run from a plain chat window. The more automatic ones — scheduled summaries, a competitor brief that writes itself, an Instacart order placed for you — need AI connected to your tools or able to control your browser, through features like connectors, scheduled tasks, or computer use. That's the line between AI as a chat and AI as a system.

Work use. Some of the ways you can use AI at work:

• A queryable second brain. Point AI at your notes, project docs, and past decisions so it becomes a source of truth you can ask — "what did we decide about pricing in March?" — and a brainstorming partner instead of a folder you dig through.
• Morning email and Slack summaries. Schedule a digest of overnight email and Slack so a short brief is waiting before you sit down: what's urgent, what changed, what needs you.
• A reviewer in someone else's voice. Have AI review a draft as a specific persona — your skeptical boss, a detail-obsessed client — and flag what they'd push back on before they see it.
• A weekly competitor brief. Set up a scheduled search that checks competitors' websites for changes — a price cut, a new feature, a repositioned homepage — and writes you a weekly summary.
• First-draft financial models. Hand AI your assumptions and raw numbers to build a starting forecast you then refine. Faster than a blank spreadsheet, and you still own the final version.
• Calendar management. Let AI handle the back-and-forth of scheduling and rescheduling, and hold your focus blocks, so booking a meeting stops eating your morning.
• A reusable tool for the team. Build something once — a branding checker, a formatting tool — that other teams run on their own work to stay consistent without routing through you.
• Self-updating dashboards and reports. Connect AI to your data so recurring dashboards and reports refresh themselves with new numbers instead of you rebuilding them every week.
• New-hire onboarding. Turn your scattered docs into an assistant a new hire can ask — where to find things, who owns what — so they ramp without booking time with half the company.

Personal use. Some of the ways I've used AI in my own life:

• A prioritization coach and to-do list manager. I tell it everything on my plate and it helps me figure out what actually needs to happen today, order my day, and stay focused when I drift.
• Managing big, scattered projects. My favorite use so far has been running a nanny search — candidates, notes, and decisions in one place instead of across five inboxes and tabs. The same setup works for a home renovation or design project, or preschool applications.
• Grocery ordering. I dictate what we need, Claude checks my saved preferences and instructions, opens Instacart, and adds everything to the cart — so ordering groceries stops being a weekly chore.
• An annual household budget. I paste in our financial statements — without any account numbers — and have it build a budget from our real ongoing spend, then check us against it.
• Inbox cleanup. I have it clear spam, unsubscribe me from listservs I never open, and write Gmail filters that auto-sort incoming mail — so the sorting keeps running even if I turn off AI access to my email later.
• File cleanup. I point it at the junk drawer of old files on my computer and Google Drive and have it organize them into something I can actually search.

Where AI falls short

AI is strong, but it shouldn't be used for everything, and, contrary to what you might see on YouTube, it does fall short. Here are the things I'd never fully hand to AI:

High-stakes decisions. You can use AI to gather information, lay out options, and pressure-test your reasoning, but you can't let it make the decision for you. And you should carefully review everything it puts together first.

Finalized numbers or outputs. AI drafts; you pressure test and verify. AI is confidently wrong often enough that "it looked right" is not a control.

Sensitive or regulated data and PII. Be careful what you paste in. On consumer plans (Free, Pro, Max), your conversations can be used to train future models unless you've turned that setting off — which makes them a poor place for client PII, health information, or anything regulated. Business plans (Team and Enterprise) don't train on your data by default, but even then, check your own obligations before putting regulated data anywhere. When in doubt, leave it out.

The sunshine test. This is the simplest gut-check I know: if you'd be ashamed to admit you didn't write or check something yourself, don't ship it without writing or checking it yourself. The proposal a client is paying for, the analysis your reputation rides on, the email to your most important relationship — if it wouldn't survive being shown in full daylight as "AI did this and I didn't look," it's not a task to hand over unsupervised.