Stop the Busywork
A no-code guide to building specialized AI assistants for any use case.
Disclaimer: The views and opinions expressed in this post are my own and do not reflect those of my employer, Asana. This content is independent from my role at Asana and designed for educational use.
Welcome back to Information Processed! If you’ve been following along, you already know we’re building a community built around demystify artificial intelligence for tech-forward professionals who want to build custom AI solutions—with zero lines of code.
In my previous posts, we’ve touched on my journey into AI and a recap of 2024.
Today’s post takes a more hands-on approach: I’ll show you exactly how to create a specialized AI assistant for a single, real-world use case, and then scale it with no-code automations.
Start With the Pain
We live in an era where we’re constantly told, “AI can do anything!” Yet, if you’ve tried using AI to automate parts of your job, you’ve likely noticed that not all tasks are well-suited for current models. Often, it’s hard to tell which tasks fit the bill.
Whether you’re in marketing drafting blog posts, in sales generating proposals, in HR writing policy documents, or in customer support summarizing tickets—the steps to decide if AI can handle your pain points are largely the same.
In this post, I’ll share a four-step flow I use to determine if a given task can be offloaded to AI:
Identify a high-pain, repetitive workflow.
Validate quickly if the AI can handle it.
Optimize by adding relevant data if needed.
Automate only if the frequency and structure justify it.
To keep things concrete, I’ll weave in an example of how you might apply this framework to drafting a blog post—but remember, the exact same logic could apply if you’re generating end-of-month finance summaries, writing help center articles, creating internal reports, or something else entirely.
Part 1 - Identify a Painful, Repetitive Use Case
First, ask yourself: “Which parts of my job do I find painfully repetitive?” You’re looking for tasks that check these boxes:
Frequent: You do them multiple times a week or month.
Repetitive: Often the same structure or format (e.g., follow a template or standard approach).
Text/Language Heavy: Generating or refining written content, summarizing info, or responding with consistent messaging.
Manual or Tedious: You’d rather focus on more strategic or creative work.
Examples across departments might include:
Marketing: Repeatedly drafting social posts or blog intros.
Sales: Generating standard proposals or post-call summaries.
Finance: Creating monthly or quarterly performance briefs with the same structure.
Customer Support: Writing similar response templates or summarizing tickets into internal notes.
Example: In marketing, you might be creating multiple blog posts per month. The format is usually the same: an intro, a few points or paragraphs, and a conclusion—very repetitive if you’re churning out content constantly.
Not All Tasks Are AI-Friendly
Even if something is repetitive, it might not be suitable for AI if:
The task relies on esoteric or specialized knowledge that’s not publicly available (and you’re not prepared to provide it in a knowledge base).
There are strict legal requirements or compliance steps that vary drastically every time.
The process is too unstructured (no consistent format or style).
But for most textual tasks that follow a broad pattern—like summarizing documents, writing short responses, or drafting blog posts—AI can often help.
Part 2 - Validate Feasibility Quickly & Cheaply
MVP Testing in a Chatbot
Don’t build an elaborate automation yet. The next step is to test a minimum viable prompt in a user-friendly AI chatbot—like ChatGPT, Claude, Gemini or DeepSeek—using real or representative data.
Write a Simple Prompt: Outline your goal, format, tone, and any must-have elements.
Feed It Real Inputs: Provide an actual scenario or partial data you’d use in your real workflow.
Evaluate the Output: Is it correct? Could you polish it with minimal effort?
Here is a universal prompting framework we’ve developed internally at Asana that can be shaped around any use case.
Objective: “Your job is to do XYZ”
Situation: “You are doing this for XYZ (Person, Company, Department, Team)”
Task: “I will provide you XYZ information and you will turn that into XYZ”.
Specific Context: “Here is a framework/template of how to do this well:”
“Framework / Template / Examples”
If you think about how you would onboard a brand new colleague to the same assignment, it’s the same content / concepts here.
Here’s another resource the incredible EDU team at Asana has put together with some tips and an example of a prompt for naming incoming requests with a framework.
Example: Let’s say you want a 600–800-word blog post introducing a new Asana product feature. You jump into ChatGPT and paste a prompt like:
Objective: "Your job is to write a 600- to 800-word blog post highlighting our new product feature X."
Situation: "You are working for Asana’s Product Marketing team."
Task: “I will give you the details of the new product and you will turn that into a personalized blog post”
Specific Context: ”Here is the template for our Blogs: TEMPLATE.”
Then you add a couple of real bullet points about feature X so the AI knows what it’s talking about. Now you see what it spits out.
If the first draft is off-base, try refining your prompt. Maybe you need more detail on the style, or you see the AI leaving out an essential CTA.
Don’t stress if it takes a few attempts to get a satisfying draft. This is normal for prompt engineering. The real question: “Does this reduce my manual workload enough to be worth it?”
Possible Issues:
AI writes too much fluff or goes off-topic.
AI lacks domain-specific detail.
AI uses the wrong tone for your brand.
Tweak your instructions, feed in more context, and retest. Once you’re hitting a draft that’s around 70%–80% polished, that’s your sign the AI can handle this use case (even if there are more improvements needed).
Part 3 - Optimize With Data (If Needed)
Why Data Matters
Sometimes AI’s training knowledge (or your short bullet points) isn’t enough. If your content relies on:
Internal documents: e.g., HR policies, product specs, or financial figures.
Proprietary facts: e.g., your company’s unique sales metrics or customer success data.
Brand guidelines: e.g., style guides, tone-of-voice docs.
…you might integrate that data into your instructions for AI or use a retrieval mechanism so the AI can automatically reference relevant info of large data sets.
This step is optional—if your use case is low volume, the knowledge needed is generally understood (these LLMs are incredibly smart), or you can easily copy/paste a snippet, you might skip it. But if you’re constantly pulling the same specialized data or brand rules, a data integration could be worth it.
Basic vs. Advanced Integration
Manual Option:
Copy/paste key information into your instructions.
Great if low frequency or your reference is short.
Retrieval-Augmented Generation (RAG):
You store documents in a system that automatically fetches relevant snippets and provides it to the AI.
Useful if you produce lots of text requiring specialized references (like technical support docs).
I could spend an entire post on optimizing your use cases with data - but I’ll keep it lightweight for this initial introduction to the topic.
Part 4 - Decide on Automation to Scale (Or Not)
Criteria for Automation
Finally, if you’ve proven the AI can handle your scenario (and optionally integrated data if needed), you might ask: “Should I automate this fully with a no-code workflow?”
Criteria that justify automation:
High Frequency: The task (or blog creation, or proposal drafting) recurs multiple times a week/month.
Predictable Steps: The process is fairly standard each time (e.g., generate draft → review → finalize).
ROI: Automating it will save you or your team measurable hours, or reduce error rates significantly.
Example: If your marketing team churns out 3–5 new blogs weekly, the overhead of copy-pasting into a chatbot and reformatting in your CMS could get old fast. At this volume, building an automated workflow might be a game-changer.
Building a No-Code Workflow
Example: Suppose you want an automated system that triggers an AI draft whenever a new “Content Request” is created in Asana (or any AI automation tool).
Trigger: A new task or form submission labeled “Blog Request.”
AI Step: The system feeds your validated prompt (and optional data references) to the model, generating a first draft.
Human Review: You or another reviewer checks the draft, makes minor edits, and approves it.
(Optional) Additional Steps:
Translations (if you publish in multiple languages).
Post a final link to Slack/email.
Here is a walk through of how easy it is to automate one of these use cases in a no-code automation tool like AI Studio. (Built from Scratch and tested in 4 minutes - with a minute of wasted time).
When Not to Automate
Rare One-Offs: If you only need to do it once every six months or it’s never the same.
Highly Nuanced Processes: Each instance is too unique to benefit from a structured workflow.
Sometimes a manual prompt with a short copy/paste is just fine—especially if you’re not repeating the process at scale.
PS: Agents will probably be the long term answer for these custom one-offs uses of AI.
Pulling It All Together: An End-to-End Flow
Here’s what it might look like for you to apply these steps—whether you’re drafting a blog, generating a repeatable internal report, or writing monthly status updates.
Identify: You realize you’re spending hours each week rewriting similar text—whether it’s marketing blogs or proposal templates.
Validate: You do a quick test in ChatGPT, providing your real scenario. After a few prompt refinements, you see the AI’s output is decent—maybe 80% finished with minimal extra editing.
Optimize: If your text needs deeper references (brand guidelines, specialized data), you embed those or set up a retrieval system. This ensures accuracy.
Automate: Because the process is frequent and fairly standard, you build a no-code workflow to handle the entire chain: from new request to AI draft to review/approval to final output.
Example:
You confirm the AI can draft a blog about your new product updates effectively.
You feed it a brand voice guide for consistent tone.
You integrate the entire thing in a no-code workflow so each new blog request is auto-drafted. All you do is edit and publish.
Common Pitfalls and Final Thoughts
Overhyping AI:
AI can drastically speed you up, but it’s not an instant cure-all. You still need to review outputs and maintain integrity.
Skipping Validation:
Never jump into building a big automation without testing the AI’s capability in a chat environment first.
Excessive Data:
If you feed AI too much irrelevant info, it can produce muddled outputs. Use data wisely.
Model Changes:
AI providers update their models regularly, which can shift output quality. Be prepared to revisit or tweak prompt techniques over time.
Embrace Iteration
Adopting AI for repetitive tasks is all about experimenting. You’ll refine prompts, test partial data integrations, then maybe scale up automation. The learning curve is normal—especially when you’re among the first in your team or industry to try no-code AI.
Next Steps:
Brainstorm Pain Points: Which tasks are you sick of doing repeatedly?
Do a Chat Test: Put a short prompt into ChatGPT (or similar) using real or sample data. Evaluate the draft.
Refine: Tweak instructions and see if you can reach a “usable” level of output with minimal editing.
Consider Data: If you constantly need special references, embed or integrate them.
Automate (if relevant): If the process repeats often and you want full efficiency, hook it into a no-code workflow to handle most steps automatically.
Conclusion
Whether you’re drafting blogs every week, repeatedly writing proposals in sales, preparing internal reports in operations, or generating monthly briefs in finance—the same four-step logic applies.
Start small by testing a feasible use case in a chatbot.
If the AI handles it well, integrate any extra data you need.
Then—and only then—decide if automation is worth your time, based on frequency and consistency.
When used wisely, AI can free up hours a week, reduce errors, and make your job more strategic (and less copy-paste). The real key is to approach it systematically. If you follow this framework, you’ll know exactly which tasks make sense to automate—and you’ll avoid sinking effort into those that don’t.
Thanks for reading—I have so much gratitude for our growing AI learning community. We are in this journey together…
Share in the comments if you’ve built something awesome or have run into any roadblocks.
Also, if there any of the areas you are interested in learning more about, please let me know.
And always - if there is anyone you know interested in learning more about AI this year - please let them know about our Information Processed community.