AI Design Pattern Recommender

Is this a strategy pattern, or is it three if statements that would read perfectly well? That question comes up constantly and it rarely gets answered honestly, because pattern names carry authority. Saying "I used the observer pattern" sounds more considered than "I added a callback list", even when they are the same thing.

What is AI Design Pattern Recommender?

It is a free page that matches problems to patterns. You describe the situation, or paste the code that is bothering you, and the response names the patterns that apply, explains the fit, sketches how each would look in your context, and says what it costs.

The last part is what makes it useful. Every pattern is a trade. Indirection buys flexibility and charges you in files, names and hops. A recommendation without that cost stated is just a suggestion to add structure.

Why Use AI Design Pattern Recommender?

Because patterns are learned as a catalogue and applied from memory, which produces two failure modes. The first is reaching for the three patterns you know well whether or not they fit. The second is not recognising that a well known solution exists and reinventing a worse version of it.

A recommendation that names several candidates and compares them addresses both. You see the option you would not have considered, and you see the argument for the boring answer next to it.

Who Should Use It?

• Developers with a design decision to make and nobody senior to ask.
• Reviewers who suspect a change is over engineered but want a second opinion.
• Engineers refactoring a class that has grown several responsibilities.
• Anyone learning patterns who wants them explained against real code.
• Teams settling a disagreement about whether abstraction is warranted.

How Does AI Design Pattern Recommender Work?

You describe the problem, it proposes candidates. The description that produces a good answer has three parts: what the code does now, what keeps changing about it, and what makes the current shape uncomfortable.

1. Open AI Design Pattern Recommender. Free, no account needed.
2. Paste the code, or describe the situation if the code does not exist yet.
3. Say what you expect to change, and how often that kind of change arrives.
4. Name the language, since idiom sometimes removes the need for a pattern entirely.
5. Turn Note Trade-offs on and set Style to Pragmatic.
6. Generate, then read the do nothing option before the recommendations.

Key Features

Advanced Options Guide

Ten controls sit behind the accordion. Style and Depth decide how opinionated and how detailed the answer is.

Example Inputs

The situation matters more than the code. Here is a description with the right parts in it.

TypeScript, Node backend. We have a shipping cost calculator that started as
one function and now has a fourteen branch conditional: three carriers, each
with domestic and international rates, plus weekend surcharges and a legacy
flat rate for two grandfathered customers. We add a carrier roughly twice a
year, and each addition means editing that function and re running the whole
test suite nervously. Small team, four developers, none of us wants a
framework. What fits here, and is a pattern even the right answer?

The last question is the important one. It invites the answer that the branch count is the symptom and the underlying problem is that carriers are data rather than code. Without that invitation you tend to receive the pattern you hinted at.

Example Outputs

With Style on Pragmatic and trade offs on, the response starts with the option most people skip.

Option 0: no pattern, restructure the data
  Carriers differ in rates, not in behaviour. Move the rates into a table or a
  configuration file and the fourteen branches collapse into a lookup plus two
  real rules (weekend surcharge, grandfathered flat rate).
  Cost: a migration for the rate data. Gain: adding a carrier stops being a
  code change at all.
  This is the strongest option if the carriers really only differ in numbers.

Option 1: Strategy
  One interface, one implementation per carrier, chosen by a factory.
  Cost: three to five new files, and a new developer has to open two of them to
  follow a single calculation.
  Gain: a carrier with genuinely different logic, not just different numbers,
  fits without touching the others.
  Choose this if at least one carrier has behaviour the others do not.

Option 2: Chain of Responsibility for the surcharges
  Each surcharge is a step that may or may not apply.
  Cost: ordering becomes implicit and harder to debug.
  Gain: adding a surcharge does not touch the base calculation.
  Only worth it if surcharges are added more often than carriers.

Recommendation
  Start with Option 0. If a future carrier needs different behaviour rather
  than different rates, introduce Strategy at that point, when you know what
  varies. Do not do both now.

What makes this useful is that the strongest recommendation is the one with no pattern in it. Fourteen branches over data that differs only in numbers is a data modelling problem wearing a design pattern costume.

Tips & Common Mistakes

• ✅ State what changes and how often, since that is what a pattern absorbs.
• ✅ Ask explicitly whether a pattern is needed at all.
• ✅ Name the language, because idiom often replaces a pattern.
• ✅ Check whether a library already implements the pattern properly.
• ✅ Apply the pattern when the second case arrives, not in anticipation of it.
• ✅ Make sure the whole team can read the result, not only the person who wrote it.

The mistakes are familiar to anyone who has reviewed a lot of code. Introducing an abstraction for a variation that has occurred exactly once. Naming something after a pattern it does not actually implement, which confuses every future reader. Using a pattern to avoid a data modelling decision, as in the example above. And adding indirection nobody on the team is comfortable with, which turns a design improvement into a maintenance problem.

Comparison Table

AIToolsay gives each job a page that already knows the domain, so this prompt box expects a design problem rather than an open question. The options include a style setting that decides whether you get the textbook recommendation or the pragmatic one, and a depth dial that controls how far each candidate is developed. The model selector lets a second engine assess the same problem when a recommendation feels like it is adding structure for its own sake. It is free and needs no account, so checking a decision before you write the code is a one minute habit. Session history keeps each assessment under the result while you compare options. The rest of the architecture tooling on AIToolsay follows the same shape, so the structure and the refactor around this decision are each a page away.

Frequently Asked Questions

Take the class in your codebase that everyone hesitates before editing, describe what keeps changing about it, and ask whether a pattern is even the answer. Sometimes it is, and sometimes the honest reply reframes the problem entirely. The Telegram community is a reasonable place to argue about that, and the newsletter or push notifications will tell you when new architecture tools arrive.

Let AI Speak.