Designing Around a Workflow

Forbes Stories — Tue, 08 Sep 2026 03:02:39 +0000

title: "Building AI Tools for Real-World Web Workflows"
published: true
description: "What I learned while building AI-powered tools for writing, SEO, content processing, and digital productivity."
tags: ai, webdev, productivity, seo

Building AI Tools for Real-World Web Workflows

Building an AI-powered application is easier than it was a few years ago.

The difficult part is no longer simply connecting an application to an AI model.

The more interesting challenge is figuring out where AI actually provides useful value inside an existing workflow.

I've been exploring this problem while building AI Mr Ferdy, a web-based project focused on AI tools, SEO workflows, content creation, and digital productivity.

The project started with a simple idea:

AI should make a workflow easier, not make the workflow more complicated.

The Problem With "AI Everywhere"

It's tempting to add AI to every feature.

A text field can have an AI button.

A dashboard can have an AI assistant.

A search box can have an AI suggestion.

A content editor can generate an entire article.

But adding an AI model doesn't automatically create a better product.

The real question should be:

What problem does the AI solve?

For example, if a user already knows exactly what they want to write, generating another draft may not help.

But if the user is stuck at the beginning of the process, generating ideas or a structure can be extremely useful.

That distinction changes how an AI tool should be designed.

Designing Around a Workflow

Instead of starting with the model, I prefer starting with the workflow.

A simple content workflow might look like this:

User Input
↓
Understand the Task
↓
Generate Suggestions
↓
AI Processing
↓
Human Review
↓
Edit
↓
Final Output

The AI model is only one component.

The interface, validation, prompt design, output formatting, and human review are equally important.

This approach has influenced how I've been developing AI Mr Ferdy.

Small Tools Can Be More Useful Than Large AI Systems

There's a lot of attention around autonomous AI agents and complex AI applications.

Those systems are interesting, but not every user needs an autonomous agent.

Sometimes a small utility is enough.

For example:

generating a better prompt

improving a piece of text

creating an initial content outline

organizing SEO ideas

transforming information into a useful structure

A focused tool can solve one problem quickly without requiring the user to learn an entire platform.

This is especially useful for people who work with websites and digital content.

Building an AI Humanizer

One of the concepts I've been exploring is AI-assisted text processing.

An AI humanizer shouldn't simply replace words with synonyms.

That approach often produces awkward results.

Natural writing depends on much more than vocabulary.

It depends on:

sentence structure

context

rhythm

clarity

audience

tone

information density

For that reason, an AI writing workflow should allow the user to review and edit the output.

The objective isn't to pretend that AI wasn't involved.

The objective is to produce text that communicates effectively.

AI and SEO

SEO is another area where AI can provide significant productivity improvements.

But SEO automation can also create problems if the workflow becomes focused only on generating pages.

A useful SEO workflow should start with search intent.

For example:

Search Query
↓
Search Intent
↓
Topic Research
↓
Content Structure
↓
Draft
↓
Human Review
↓
Internal Linking
↓
Technical SEO
↓
Publish

AI can help with several of these steps.

However, the decision-making process still matters.

A generated outline isn't automatically the best outline.

A generated article isn't automatically useful.

And a keyword appearing many times doesn't automatically make a page relevant.

The technology should support the SEO strategy rather than become the strategy.

Why Human-in-the-Loop Matters

One principle that keeps appearing in my AI experiments is the idea of human-in-the-loop systems.

Instead of:

Human → AI → Publish

I prefer:

Human → AI → Human → Validate → Improve → Publish

This extra review step is important.

AI is very good at producing possibilities quickly.

Humans are still responsible for determining whether those possibilities make sense.

This becomes especially important when an application generates content, code, recommendations, or information that users may rely on.

The Technical Side of AI Web Applications

Building an AI web application also involves many problems that have little to do with the AI model itself.

For example:

Input validation

Users don't always provide the information the application expects.

The application needs to handle empty, incomplete, or unexpected input gracefully.

Prompt design

A good prompt should provide enough context without unnecessarily restricting the model.

Small changes in instructions can produce significantly different outputs.

Output handling

AI responses may contain unexpected formatting.

The application therefore needs to handle the output safely and consistently.

Performance

Users don't want to stare at a loading screen while waiting for an AI response.

Streaming responses, efficient API calls, caching where appropriate, and clear loading states can significantly improve the experience.

Error handling

AI services can fail.

Requests can time out.

Rate limits can be reached.

A production application needs to handle these situations instead of assuming every request will succeed.

Building AI Mr Ferdy

These ideas are part of what I'm experimenting with through AI Mr Ferdy.

The project focuses on practical AI and web tools rather than trying to solve every possible problem with one giant application.

Some of the areas include:

AI-assisted writing

AI humanization

SEO content workflows

prompt generation

content creation

digital productivity

web automation

The project is available at:

https://aimrferdy.com/

The GitHub Pages reference site is also available at:

https://aimrferdy.github.io/

What I've Learned So Far

The biggest lesson is that building an AI product isn't primarily about the AI model.

The model is one component of a larger system.

A useful AI application requires:

Good UX
+
Good prompts
+
Good validation
+
Good output handling
+
Good engineering
+
Human judgment

If any of these pieces are ignored, the application can become frustrating even when the underlying model is powerful.

AI Should Reduce Friction

The most useful AI tools I've encountered have one thing in common:

They reduce friction.

They don't necessarily replace an entire job.

They remove a repetitive step.

They provide a useful starting point.

They help organize information.

They accelerate experimentation.

That's the direction I'm interested in exploring.

Instead of asking:

"How much of this workflow can AI automate?"

I think a better question is:

"Which part of this workflow should AI help with?"

That small change in perspective can lead to much better product decisions.

What's Next?

I'm continuing to experiment with AI-powered web applications and workflows.

The areas I'm particularly interested in include:

AI agents

generative AI

AI-assisted development

SEO automation

prompt engineering

developer productivity

human-in-the-loop systems

AI-powered web applications

The technology will continue to change quickly.

The challenge is to build tools that remain useful even as the underlying models evolve.

For me, that means focusing less on the hype around AI and more on the actual workflow.

Build something small.

Test it.

Measure what helps.

Improve the workflow.

That's the approach I'm taking while building AI Mr Ferdy.

DEV Community: Forbes Stories

Designing Around a Workflow