Contact

Where AI is overkill: tasks that code does better

When a language model only gets in the way: the principle of rule or understanding, twelve tasks with an honest verdict, five questions before adding AI and how to combine a model with code.

Artificial intelligence Updated

In short

AI is overkill wherever the task has a clear rule. Calculating a price with a discount, finding a product by its article number, checking a phone number in a form, counting orders for a month, sending a request to the right department by a field in the form — ordinary code does all this instantly, for free and without mistakes, while a language model does it slower, for money and sometimes wrongly. AI earns its place where there is no rule: free text, letters, documents of any form, pictures, questions in someone’s own words. The best products combine both: the model understands, the code decides and calculates.

The principle: a rule or understanding

Code follows rules exactly: the same input always gives the same result, in milliseconds and for free. A language model works with meaning: it understands a request written any way, but its answer varies, takes seconds and costs money. That is why the first question about any task is simple — can it be written as a rule?

If yes — code. If the input is free text, a scanned document or a picture, and the rule cannot be written in advance — this is AI territory. And very often the answer is “both”: the model turns chaos into structured data, and code takes it from there.

  • There is a rule — code
  • Free text and images — AI
  • Most often — both

Does this task need AI: 12 examples

Typical requests from clients, each with an honest verdict.

  • Calculating a price with discounts

    Code

    A formula. A model can make a mistake in arithmetic; code cannot.

  • Search by article number

    Code

    An exact match in the database — instant and certain.

  • Checking a form

    Code

    Phone, email, required fields — validation rules.

  • “How many orders in March”

    Code

    A database query, not a guess by a model.

  • Routing by a field of the form

    Code

    The person already chose the department — a rule is enough.

  • A weekly report from the database

    Code

    A scheduled query and a template.

  • Sorting letters by topic

    AI + code

    The model reads free text; the code routes by its label.

  • Data from invoices and contracts

    AI + code

    The model extracts the fields; the code checks totals and dates.

  • Catalogue search in plain words

    AI + code

    Meaning finds candidates; filters by price and stock are code.

  • Answers from a knowledge base

    AI

    Questions in someone’s own words — rules cannot foresee them.

  • Product descriptions from specifications

    AI

    Writing living text is exactly what models do well.

  • Recognising products in photos

    AI

    Images have no rule — only a model sees them.

Five questions before adding AI

If the answer to the first question is “yes”, the rest can be skipped — code is enough.

  1. 01

    Can it be written as a rule?

    If a programmer can describe the logic in a day — it is code.

  2. 02

    What is at the input?

    Free text, letters, scans and pictures are AI territory; fields and numbers are not.

  3. 03

    What does a mistake cost?

    Where an error means money or law, the model only prepares and a person or code confirms.

  4. 04

    Is there volume?

    Ten documents a week are faster to process by hand than to automate.

  5. 05

    Can the result be checked?

    If not, nobody will notice when the model starts to be wrong.

The best combination: the model understands, the code decides

  1. An invoice arrives

    The model extracts supplier, items and totals; the code checks the sum and finds the order.

  2. A customer writes

    The model determines the topic and urgency; the rules send it to the right person.

  3. Someone searches the catalogue

    The model turns “a warm jacket for autumn up to 100” into filters; the database applies them.

  4. A manager asks about sales

    The model turns the question into a query; the database counts; the model explains the figures.

  5. A contract is reviewed

    The model finds risky clauses; the lawyer makes the decision.

Signs that AI was added in vain

  1. The model counts money

    Sums, taxes and discounts are calculated by text generation instead of a formula.

  2. The answer changes every time

    Where the business needs one correct answer, variability is a defect.

  3. A chat instead of a button

    The person has to describe in words what one button or filter would do.

  4. Seconds instead of milliseconds

    The interface waits for a model where a database would answer at once.

  5. AI for the sake of the word

    The feature exists for the presentation, and nobody measured what it improves.

  6. A bill without a result

    The model is paid for every month, and the work is still checked by hand.

Questions about when AI is needed

When is AI not needed?

When the task has a clear rule: calculations, exact search, checks, reports from a database.

When is AI really useful?

When the input is free text, documents of any form or pictures, and there are many of them.

Can AI make mistakes in calculations?

Yes: a language model predicts text, it does not calculate. Numbers are left to code.

Is automation without AI still automation?

Yes, and the most reliable kind: most routine work in a company is rules.

How to tell if AI pays off?

Compare the hours of manual work saved with the cost of building and the monthly bill for the model.

What if the task is half rule, half text?

Then both: the model handles the text, the code handles the rule.

Online form

AI where
it pays off

I will go through your task and say honestly where AI gives a result and where ordinary code is enough — it is cheaper and more reliable. Tell me about the task — I answer within one working day.

Or write to [email protected]