Does my app need AI?

AI is not always the answer. Often, standard code is better — faster, cheaper, and always correct.

The question to ask: can the rules be written down?

Task	Rules exist?	Choose
Calculate the total in a shopping cart	yes, exact	code
Check that a password has 8 characters	yes	code
Sort by date	yes	code
Determine if an image shows a cat	no	AI
Summarise a text	no	AI
Suggest the next step in a process	partially	both

The rule: if the rules can be written down — write code. It will be correct every time, it costs nothing per use, and you can explain exactly why.

If they cannot — there are thousands of ways a cat can look and no list covers them all — then AI is needed.

What it costs to choose AI unnecessarily:

	Standard code	AI
Cost per use	~0	money every time
Speed	milliseconds	seconds
Correct answer	always (if the code is right)	usually
Explanation	exact	approximate
Works without internet	yes	usually not
Maintenance	the code	model, prompt, evaluation

The last row is the one that is underestimated: an AI feature requires ongoing supervision. The model is updated, behaviour changes, and what worked last autumn may not work this spring.

When AI is the right answer:

Signal	Example
Input is messy and varied	free text, images, speech
Rules are too many to write down	language understanding
“Approximately right” is acceptable	suggestions and summaries
Rules change constantly	trends, language usage

When AI is the wrong answer:

Signal	Why
The answer must be exact	calculation, payment, authorisation
The decision has major consequences for someone	must be justifiable and appealable
It must run in milliseconds	too slow
A good formula already exists	no reason to use AI

The most common pattern in real apps is to mix them: AI handles the messy part (understanding what the user wrote), standard code handles the exact part (calculating the price). Neither part does the other’s job.

Assess eight features in a hypothetical app for a publishing house. Write “code”, “AI” or “both” — and one sentence on why.

#	Feature	Your choice
1	Show which titles are published
2	Calculate the publishing fee
3	Suggest a book based on what you have liked before
4	Search by title
5	Search for “an exciting book about space”
6	Remind the day before the book is due back
7	Read an ISBN from a photographed barcode
8	Answer “do you have anything similar to Harry Potter?”

Answer key with justification:

1. Code — a database query.
2. Code — a formula: days × fee.
3. Both — a simple rule (same genre) goes a long way; AI gives better suggestions.
4. Code — text matching.
5. AI — requires understanding what “exciting” and “about space” mean.
6. Code — a date and a scheduled job.
7. Code — barcode reading is a solved task with standard libraries, not AI.
8. AI — requires understanding of similarity between books.

Number 7 is the trickiest. It feels like AI because it involves an image, but barcodes are designed to be easy to read mechanically and ready-made libraries exist. Throwing an image model at the problem would be more expensive and worse.

Number 3 shows the most common pattern: start with the rule, measure how good it is, and add AI only if the measurement shows it is needed.