Practical AI Workflows You Can Try This Week

Every day, my social feeds are full of folks announcing that some new model or tool "changes everything." For a while, I tried to keep up with it all. I'd read the threads, save the demos, and still feel like I was falling behind.

What I eventually figured out: that whole approach is a bit backwards.

The people getting the most out of AI right now aren't the ones who can name every beta feature published on the daily Claude Code changelog.

Instead, they're the ones using AI on something real -- a meeting they had to prep for, vibecoding internal tools, improving customer support.

That’s because practical AI fluency doesn't come from scrolling hundreds of daily news items and announcements. It comes from using AI in real-world scenarios that actually matter to you.

So today, instead of getting hypnotized by another viral "this changes everything" post, I want to show you:

• How to pick one real task and give AI the first shot

• Which practical AI workflows you should try this week

• How to turn a one-off successful AI experiment into a repeatable system

—@dharmesh

Give AI the First Shot

If I had to compress everything I've learned about practically integrating AI into one sentence: before you do a task the way you always do it, give AI a shot at it first.

It sounds simple. But it's most of the difference between people who continuously get better at AI, and people stuck at "I tried ChatGPT once."

Any time you sit down to do something tedious at work, whether that's writing an email you were dreading, prepping for a meeting, or researching a company, pause and ask: how might I use AI to help me with this? Then try it before your muscle memory kicks in and opens the usual applications.

Sometimes, it won't nail it perfectly. That's fine.

The first pass still saves time, and every attempt teaches you where these tools are strong and where they still need you.

You can do this today.

You don't need to study prompting or memorize frameworks for any of this. And it doesn’t matter whether you do it with ChatGPT, Claude, or Gemini. What matters is getting reps on real work, because that’s how your AI intuition is developed.

With that as the starting point, below are a few practical workflows worth trying.

Practical Workflows To Try This Week

Quick distinction -- “beginner” may undersell these initial baseline exercises. Advanced users should not skip these early tips if they haven’t completed them already. This is a high ROI setup work that only takes a minute.

Beginner: onboard the intern

Think of AI less like a search box and more like a smart intern that needs context, feedback, and a real assignment.

• Set Custom Instructions. Share your role, company, audience, tone preferences, and default formats. Include the kinds of answers you like and the kinds you do not. The goal is simple: make every first draft less generic.

• Meta-prompt one recurring prompt. Take one prompt you use often and ask AI to improve it. Show it the prompt, explain what keeps going wrong, and ask it to identify missing context, ask clarifying questions, and return a stronger reusable version.

• Create a reusable context packet. Make one short doc with the background AI usually needs: goal, audience, examples, constraints, files, and desired output. Before it starts, ask AI to restate the assignment so you can catch bad assumptions early.

Intermediate: use AI as a sparring partner

Now that your intern has the context it needs, it becomes more like a junior employee. We certainly want to review its work, but it can often surprise us by surfacing thoughts that we may not have considered ourselves on a first pass.

• Generate a company research brief. Use when you need a fast read on a company. Give it the website, your role, and the questions you care about. Ask for a customized report, not a generic summary.

• Use AI as a critic panel. Use when an idea needs pressure before people see it. Give it the idea and the decision. Ask what a customer, investor, marketer, engineer, and skeptic would each worry about.

• Ask AI to edit a webpage. Give AI the page (via sharing a link, copying and pasting, or uploading a file), intended buyer, and desired action. Ask it to find fuzzy promises, missing proof, confusing sections, and lines that could be sharper. Then ask for specific replacement copy.

Advanced: design the system around the work

By this point, your working relationship with the AI intern has flourished. You are not just prompting anymore: you are designing a system that helps them do their best work.

• Run a simple agent eval. Give two or three models, agents, or prompts the same real task. Then ask AI to act as the evaluator: create a scorecard, compare the outputs against your success criteria, identify hallucinations or weak reasoning, and recommend which one should become your default for that job.

• Create a feedback loop. For one repeated workflow, add a required “post-run review” to the prompt. After AI finishes the task, have it list what it was confident about, what information was missing, where the output may be weak, and what should be added to the prompt, examples, policy doc, or knowledge base before the next run. You still approve the changes, but AI generates the improvement log.

• Decompose a difficult workflow. Give AI one of your messiest workflows and ask it to turn the work into an operating map: inputs needed, steps involved, tools required, decisions to make, failure points, and handoffs. Then have it label each step as deterministic, AI-solvable, automatable, or human-owned.

Start small. Early on in your journey to master AI, the goal is to build momentum and the AI intuition of knowing where AI works and where it doesn’t.

Pick the workflows that most closely map to the work you’re already doing. And if you need step-by-step instructions to help you get started, just paste the relevant workflow into your AI of choice and ask for help. There’s no shame in that -- I do it all the time.

How to Turn 1 Win into 10 Wins

What turns a few one-off workflow experiments into AI fluency is the habit of saving, improving, and retrying what works.

Save the wins so you don't rebuild them. When a workflow works, don't just admire it and move on. Save the prompt, the context, and the steps, so next time you run it rather than reinvent it. Bonus points: convert a successful workflow into your first skill.

Spend your AI time intentionally. One way to think about the balance is what I've called 60/30/10: spend roughly 60% on workflows that work today, 30% improving them, and 10% on experiments that might not pan out. The idea is that AI fluency grows when you balance reliable workflows with deliberate improvement and a little exploration.

Keep a "not yet, but soon" list. Some things AI can't do well for you today. Write those down. The capability curve is steep enough that "it failed when I tried it" has a short shelf life. Set a reminder to retry the important ones in three to six months.

So pick one workflow. Give AI the first shot at something you were going to do anyway. Keep what worked, and circle back on the rest later.

If this nudges you to try something useful, let me know in the poll below. I'd love to hear about it.

—Dharmesh (@dharmesh)