I spent a long time using AI as a fancy search engine. Ask a question, get an answer, close the tab. Then do the same task again tomorrow from scratch.
The thing that changed it for me wasn’t a better tool — it was building actual workflows around the tools I already had. Same input format, repeatable process, output I could actually use. Once I had that, I stopped spending time doing things over and over that didn’t need to be redone.
Here are 7 workflows I actually use every week. None of them require code, paid tools beyond a basic ChatGPT subscription, or more than a few minutes to set up. Each one took me some iteration to get working well — I’ll share what that looks like so you can skip the parts that didn’t work.
1. Weekly planning in under 10 minutes
Every Sunday evening (or Monday morning if Sunday doesn’t work), I paste my loose list of tasks into ChatGPT with this prompt:
“Here’s everything on my plate this week: [paste list]. I have roughly [X hours] of focused work time available, and [list any standing meetings or fixed commitments]. Help me figure out: what actually needs to happen this week vs. what can wait, a rough order that protects the highest-priority work from getting pushed by urgent-but-less-important things, and anything that looks overloaded or unrealistic. Be direct — I’d rather know now that the week is too full than discover it Wednesday.”
What used to take 20-30 minutes of staring at a list now takes about 8-10 minutes including adjusting the output. The model doesn’t know my priorities better than I do — but having a draft plan to react to is much faster than building one from scratch. And the “be direct” instruction helps — without it, the output tends to be optimistic in a way that doesn’t survive contact with Monday.
2. Meeting notes → action items
This one I use almost every day. After a meeting, I paste in my rough notes (or a transcript if I recorded it) and ask:
“Extract the action items from these notes. Format them as: task / who owns it / deadline if mentioned. Also flag anything that was discussed or decided that doesn’t have a clear owner — those tend to fall through.”
The output isn’t always perfect — sometimes it flags things that aren’t really action items, or misses something that was implied rather than stated directly. But fixing a near-complete list takes 2 minutes. Writing it from scratch from memory takes 10, and I still miss things. The net time saving over a week is significant.
The flag for unowned decisions is something I added after noticing how often things got agreed in meetings and then didn’t happen because no one had explicitly taken responsibility for them. Getting that list in front of people after the meeting reduces a surprising amount of dropped balls.
3. Drafting emails that need a specific tone
This isn’t “write my email for me.” It’s closer to: here’s the situation, here’s what I actually want to say, here’s who I’m sending it to — help me say it without the parts I always struggle with.
My prompt format: “I need to send an email to [person/role]. The context is [2-3 sentences]. I want to [goal — ask for X, decline Y, follow up on Z]. The tone should be [direct but not cold / warm but not sycophantic / professional but not stiff]. Keep it under 150 words and don’t open with ‘I hope this email finds you well.'”
I always edit the output — I’m not copy-pasting AI emails verbatim. But starting from a draft that’s 80% there is faster than starting from a blank page, especially for messages where the tone is tricky. Sensitive conversations, pushback emails, requests to people senior to me — these are the cases where it helps most.
If you work on this specifically, the post on writing human-sounding emails with AI goes into more depth on the editing process that keeps the output from sounding like a template.
4. Turning research into something actionable
When I need to get up to speed on something quickly, I paste in an article, report, or document section and ask:
“Summarize this in plain language. What are the 3-5 most important things I need to know? What would I be wrong about if I stopped reading here and assumed your summary was complete?”
That last question is the one that makes this more useful than a basic summary. It specifically asks for the nuances and edge cases that summaries typically erase. The model is prompted to surface what it’s leaving out, which often includes important caveats.
One thing I always do: I keep the source open and skim-check the summary against it for anything I’m going to act on or share. Takes 2 minutes. AI summaries are usually accurate but occasionally something gets dropped or shifted. Verification is cheap insurance.
5. Building reusable SOPs for anything I do more than once
Any time I finish a task that I’ll probably do again, I spend 5 minutes turning it into a reusable process. I tell ChatGPT what I just did and ask:
“Turn this into a repeatable SOP. Write it as a numbered checklist with just enough detail that someone who hasn’t done this before could follow it without asking me questions. Include any common mistakes or edge cases you can infer from what I described.”
I save these in a simple Notion doc. Over time, it becomes a library of processes I can hand off, delegate, or just follow myself when I can’t remember how I did something six months ago. The “include edge cases” instruction is important — the checklist is most useful when it includes the things that can go wrong, not just the happy path.
6. Brainstorming without the blank-page problem
Brainstorming with AI is most useful when you’re stuck, not when you want it to generate everything. The workflow: give it the context and constraints, ask for options, then pick and iterate.
“I’m trying to [goal]. Here are the constraints: [list]. Give me 8 different approaches — include some obvious ones and some less obvious ones. For each one, give me a one-line version and a one-line reason why it might not work.”
I rarely use any of the options directly. But seeing 8 directions at once usually helps me figure out which direction I actually want to go — which is the hard part of brainstorming. The “reason it might not work” piece adds friction that forces the model to think critically rather than just listing things that sound good.
7. Weekly review in 15 minutes
End of week, I paste a quick brain dump: what I finished, what didn’t get done, anything that felt off or harder than expected, any wins worth noting. Then:
“Based on this, what patterns do you notice? What should I consider adjusting for next week? Are there recurring problems in here that I’m working around rather than solving?”
The value here is less about getting useful answers from AI (though sometimes it does surface something I missed) and more about having to articulate what happened clearly enough to paste it. The act of writing it down in a way that makes sense to an outside reader is where most of the insight comes from.
The “recurring problems you’re working around” question has been particularly useful. I’ve caught a few patterns this way — tasks I kept moving to next week, a type of work I was consistently underestimating — that I wouldn’t have noticed just from the feeling of the week.
What makes these work
None of these workflows are magic. They work because they’re consistent — same structure, same trigger, same place to save the output. The AI handles the mechanical part fast, and I spend my time on judgment calls.
The other thing they have in common: I’m always bringing real input. My actual task list, my actual meeting notes, my actual research needs. Not “help me be more productive in general” but “here’s the specific thing I’m working with right now.” The specificity is what makes the output usable.
If you’re trying to figure out where to start: pick the workflow that solves the most annoying recurring task you have, and do just that one for two weeks. Get the habit before you add anything else. The goal isn’t to implement all seven of these at once — it’s to make one of them automatic, then decide if you want more.
For more on building AI into a daily practice rather than using it occasionally, the guide on why AI productivity systems fail after 7 days is worth reading — it covers the specific reasons people set these things up and then stop using them, which is a different problem than figuring out what to set up.
Why These Workflows Actually Stick
A lot of automation guides tell you to set up complex systems that look impressive and then get abandoned within two weeks. The seven workflows I’ve described are different for one reason: they each solve a specific, recurring pain point in a way that’s immediately obvious the first time you use them. You don’t need to convince yourself the ROI is worth it — you feel it on day one.
That immediate value is what makes them stick. Habits that are reinforced by obvious results are the ones people maintain. If you start with the email drafts workflow because writing emails is something you actively dread, you’ll feel the relief the first time you go from 20 minutes of staring at a blank reply to a good draft in 2 minutes. That feeling is what gets you to open the tool again tomorrow.
The opposite is also true. Workflows that require a lot of setup before they pay off — complex automations, elaborate prompt systems, integrations between multiple tools — are the ones that people abandon. Start with the simple, immediate-payoff workflows first. Build from there.
How to Add New Workflows Over Time
Once you have two or three of these workflows running smoothly, you’ll start to notice other repetitive tasks in your work that might benefit from AI. The right way to add a new workflow is to identify a specific task you do repeatedly that follows a pattern, figure out the prompt that handles the pattern, and use it for two weeks before evaluating. Don’t try to build everything at once.
The pattern for evaluating whether a task is a good candidate for automation: is it repetitive? Does it follow a similar structure each time? Would a good first draft — even one that needs editing — save meaningful time? If yes to all three, it’s worth trying a workflow for it.
A few additional tasks that commonly make good AI workflows, beyond the seven I’ve covered: weekly status report writing, creating agendas for recurring meetings, first-pass categorization of feedback or survey responses, turning voice memos or rough notes into structured documents, and generating talking points for presentations. Each of these has the same property as the core seven — repetitive structure, high text volume, benefits from a good template prompt.
The Bigger Picture: AI as Infrastructure, Not a Magic Trick
The framing I find most useful for AI automation is infrastructure rather than magic. Infrastructure is something you set up once, maintain occasionally, and rely on every day without thinking much about it. Electricity is infrastructure. Your internet connection is infrastructure. These aren’t exciting — they’re just reliably useful in the background of everything else you do.
That’s what good AI workflows become. You stop thinking of them as AI — you just think of them as how you do certain tasks. Email replies involve a quick prompt. Meeting notes involve pasting into a template. Weekly planning involves a few minutes with ChatGPT on Friday afternoon. The “AI” part fades into the background; the result is just that you’re consistently faster and more organized at a set of tasks you used to find tedious.
Getting to that point takes a few weeks of deliberate habit-building. The payoff compounds indefinitely. Seven workflows, 15 minutes each week to build them initially, and you’ve permanently changed how you work on a set of tasks that probably represent a significant fraction of your work time.
For anyone just getting started, the most important thing is to pick one workflow from this list today — whichever one addresses the task you find most tedious — and use it for the rest of the week. Not all seven at once. Just one, consistently, until it feels natural. Then add another. That’s how AI automation actually sticks, and that’s how it eventually becomes infrastructure. You can find more guidance on how to get started with AI without getting overwhelmed if you want a broader foundation before building workflows.
Making Each Workflow More Powerful Over Time
The first version of any workflow prompt is rarely the best version. The way you improve it is through accumulated iteration: every time you use a prompt and find yourself editing the output more than expected, you update the prompt with a new constraint or clarification. This creates a feedback loop where your prompts get more specific and accurate over time, and the outputs require less editing.
For example, when I first started using AI for email replies, my prompt was generic: “Draft a professional reply to this email.” After a month of editing outputs, I’d learned a lot about what I actually wanted. My current prompt includes: “Match the tone of the original email — if it’s casual, be casual. Keep it under 150 words unless more detail is explicitly needed. End with a clear next step or question.” Those three additions eliminated most of the editing I was doing. They came from noticing, over 30 or 40 uses, what I kept changing.
The same improvement process applies to every workflow. Build in a small habit of reviewing your prompt once a month and asking: what do I still find myself fixing? Then add a constraint that addresses it. After six months, your prompts will be substantially better than when you started — and the time you spend editing will be substantially lower.
When Not to Use These Workflows
There are situations where even the most well-designed workflow isn’t the right tool. High-stakes communications — anything with significant professional, legal, or emotional consequences — should not be drafted primarily by AI and lightly edited. The risk of getting the tone wrong or missing something important is too high, and the cost of getting it wrong is too high. For those situations, AI might be useful for a thinking partner or a structural outline, but the actual writing should come from you.
Similarly, tasks that are fundamentally about relationship and judgment — a difficult conversation, a sensitive piece of feedback, a nuanced negotiation — don’t benefit from AI automation. The value in those situations comes from your specific knowledge of the person and context, which AI doesn’t have. Use it for the operational and repetitive work; reserve your full attention for the things that require genuine human judgment.
Finally, any task where accuracy is critical and verification is difficult — financial calculations, medical information, legal analysis — should use AI only with careful verification of every output. These are areas where AI’s tendency toward confident-sounding errors is most dangerous. The workflows I’ve described are appropriate for the kind of work where a good first draft is valuable and where errors are detectable and fixable. Know the limits and stay within them.