I’m announcing something special this Friday. If you want early access, click here to join me tomorrow at 5 PM for a free AI office hour session (download file for Apple Calendar). If you’re struggling to make AI work for you, this session will help you work through your problems.
—has taken our professional lives by storm. The more I’m using it in my day-to-day tasks, the more I feel I’m solving problems that I didn’t actually have in the first place. It is just too easy to go down a rabbit hole of “What if I just write a quick tool for this?”, and then find yourself 2-3 hours of fable prompting later when you have built a tool that you’ll likely only use once. You snort that peak productivity like cheap cocaine. Only, to feel an emptiness at the end of the day that is hard to describe.
It’s the Pareto principle all over again. I produce about 20% amazing workflows, automation, and apps for the 80% junk that I end up throwing out or not even finishing after a full afternoon of attempts to fix it, rewrite it, or shelve it after I can’t quite get it to the polishing level that I’m happy with. On top of that, the 20 different agent tasks running on my computer feel like a busy conversation at the watercooler, out of which I can really only pick up one or two strands of intelligible conversation. My feeble mind is clearly the bottleneck here. Sometimes this fills me with frustration about how all these AI influencers on YouTube and social media are automating away their solopreneur businesses (I call BS!). Other times, it gives me hope for humanity and that not all of our jobs are going to be automated away much too soon.
All of this hits an aspirational problem that I think many of you share with me. We always find something that we would like to do more of, but that we do not regularly do yet. You might find that your AI automations are competing with an existing habit instead of attaching to one. For example,, if you already read papers on Sunday mornings with your double espresso (or Latté if you’re like me), you will likely abandon a workflow that requires you to first pipe the PDFs through three different tools before you can even begin reading them. The cost is much higher than the benefits that you get from this task. Why? Because now you have an AI output review queue. And believe me, this is one of the insidious tidbits of AI work. Sure, it produces tons of output in minutes, but I still have to review it. And I’m a slow reader.
So, you face two choices: Take your time (and accept that reviewing AI work is now part of your job) or just do it from scratch without AI. For many creative tasks, it ends up taking the same time. Just with AI, you’ve added work to remove work. “Mischief, thou art afoot.”
If you feel like your AI work costs your more to maintain it returns on your time investment, it’s probably time to check your workflows. You shouldn’t have to invest 20 minutes a month to tune a workflow that saves you only 10 minutes. So, today I’m giving you a checklist of four questions that you can run before you build anything with AI to see if the balance sheet stays positive:
1. Does it solve a recurring problem I actually have?
Any problem that you face over and over again, whether it is during a day, a week, or a month, is worth considering automating. So, the first step is to just watch for activities that you notice you are repeating regularly in your work. Those are the most valuable ones to automate with AI.
Let’s say you’re a policy researcher who briefs a committee every other week, writing those briefs is a recurring problem. A workflow that helps you structure or draft them is of true value to you. If you’re thinking, “I should probably read more broadly in adjacent fields,” that is just aspirational. Here, you’d wait to build the workflow until your reading becomes a recurring commitment or habit.
Here’s a simple reflection question that you can ask yourself at the end of a week or a month: Did this exact problem slow you down in the last week/month? If you have to think hard about when it last came up, it’s not recurring enough to automate.
I think the trap that many of us fall into is that we are immaculate at imagining valuable work, which means that we can always see an ideal version of how we’d do things. We could imagine a better knowledge base, a more optimal scanning of the recent literature, or a more elegantly organized research archive. However, we have to remember that a workflow doesn’t become useful because it supports the future person that we’d hope to become. It becomes useful only if it can support what we’re currently repeatedly doing.
So, you always have to start with the work that you’re currently doing.
2. Does it attach to my existing behaviour?
As humans, we operate our daily lives on habits, repeatable structures that shape how we do things. The best way to wire a workflow into your workday is to attach it to a trigger that is part of something that you already do regularly.
Let’s say I already know that I’m opening my email at 8 AM and then pull up the same analysis environment. Or I always draft a one-page summary before I have an important client conversation. I would consider these behavioural anchors and I could turn them into triggers for my behaviour. Workflows can be triggered well if they’re linked to these existing anchors and this is much easier to do than establishing a habit or a new behaviour from scratch.
So, here’s another simple reflection question you can ask yourself: What did you do this morning that you do most mornings? The answer to that will give you an opportunity for creating a new workflow or automation based on your habit.
For example, suppose you routinely annotate articles before adding them to your reference manager. An AI workflow that turns your annotations into three reusable components—a claim, a limitation, and a possible implication for your current project—fits such a habit. You don’t need to adopt it as a separate practice because you’ll just do it where you already work on a paper.
By contrast, a weekly automated research intelligence briefing sent to a dedicated inbox may look sophisticated, but unless you read this, it won’t be useful to you. You have to ask yourself, if this is something that you’d check, or if it’s competing with everything else that’s demanding your attention currently. Just thinking you’ll read a brief because it’s created is unlikely to happen. It’s much better to work it into a ritual that you already have (e.g., reading a newspaper).
3. Does it produce an immediately usable output?
Good workflows shouldn’t take more than five minutes to clean up and put into action. The best output is output that you can act on right away. Otherwise, if you have to spend too much time re-editing everything, you end up resenting this. And then over time, this will just not naturally fit into your work, and you’ll just start skipping it. So, there’s no efficiency or productivity gain here.
For me, five minutes is really the maximum I would invest into editing it. With the ideal being closer to 30 seconds of a quick review where I just scan the output, maybe tweak a line, and then use it right away. That, to me, is an actually efficient output.
Let’s imagine you always write a one-paragraph framing note before presenting quantitative findings. Then, you should build a workflow that generates a first-draft version of that note from your results and a few keywords. Reviewing this will be manageable and feel familiar because you already know what the purpose of the paragraph is. This will also help you to understand what’s wrong in seconds.
Or just imagine that your clients always come to you with the same questions repeatedly. Things about what a specific finding means, how you would justify a method, or what they could actually do with it. In these cases, you’d want to design a workflow that turns your research report into exact responses to these question types (think of it like an FAQ). It would still be based on your research report, it would just deliver the answers that they’re looking for much quicker.
Problematic workflows will deliver formats to you that you’ve never actually used before, or they will use a voice register that you might not use in your own writing. These would need to be adjusted, and you’d have to improve the output quality just so that you don’t feel resistance to the artifacts you get.
4. Have I run this manually three times before automating it?
I’d almost say that this is the most important rule because to really understand a process, you have to run it by hand first. And I would say three times is a good minimum threshold to see how the process is actually done, so that you can describe it in detail. This helps set the quality bar for the automation you’re trying to build. It lets you understand what steps are essential to this workflow and what criteria determine the quality of the output that you get.
The first manual run shows you what the task is. The second shows you what is annoying about it. The third shows you what matters most in the output.
Going through this more labour-intensive run of the process three times will allow you to not guess about the task but have a solid understanding of it and your automation won’t miss the point of the task.
The goal for any good automation is to model how you actually think through a task. And for this to succeed, you do have to go through the process of thinking through that task first. Otherwise, you might get something plausible and it might look right, but, at the end of the day, it’ll still feel wrong, and you won’t actually use it.
Keep, Redesign, or Retire
But even as you’re building these AI workflows (e.g., skills) and automations (e.g., routines), you have to keep an eye on the ones that actually provide benefit for you. That means running a weekly tally of workflows that you’re actively using. So you have to make three decisions here: whether you want to keep the workflow, redesign it, or retire it. This is the best way to keep your AI work manageable as it likely grows.
Keep an automation/workflow if it runs as expected, produces usable output within five minutes of review, and you would notice its absence immediately if it broke.
Redesign it if you use it occasionally but skip it more often than not, if the output usually needs heavy editing, or if you have started working around it. Redesigning means returning to the four questions and rebuilding from scratch. Don’t just tune the prompt. Rethink the workflow.
Retire it if you can’t remember the last time you ran it, if you are running it only to justify having built it, or if the maintenance cost is higher than the return.
As a simple rule of thumb, you should ask yourself after you’ve run a workflow/automation three times without producing anything that you’d end up using, now should be the time to change or delete it.
Why even trim workflows, you ask? Because in theory there’s lots of storage available and a lot of the times you have more tokens than you need for your AI work. But there’s a mental overhead when you constantly run a system that’s only half managed, so do yourself a favour and clean it up.
The next time you catch yourself mid-rabbit-hole, thinking “What if I just write a quick tool for this?”, stop and ask the four questions. Recurring problem. Existing anchor. Usable output in under five minutes. Three manual run-throughs before you automate. Any idea that makes it past this test deserves your afternoon, and the tool you build will still be useful next month, saving you time while everyone else reviews an inbox full of AI briefings that nobody asked for. The rest of your ideas can stay ideas. Close the tab, pour the Latté, and read your papers the way you always have. The high of building something you use every week beats the cheap stuff.
Build better AI workflows that help you thrive in your own work.
Bonus
This week, Premium Write Insight + AI Research Stack members get the complete workflow audit kit: a Review Queue Balance Sheet spreadsheet, printable Automation and Keep, Redesign, or Retire guides, a downloadable Claude skill that interviews you about a repeated process and turns it into a reusable skill, and five paste-ready prompts for committee briefs, reviewer responses, paper annotations, recommendation letters, and quantitative framing notes. You also get 10 practical resources for deciding which systems deserve a permanent place in your week. Join Premium, run the audit, and reclaim the afternoons you now spend building workflows you abandon before your week ends.
