A SOFTWARE ARTIFACT FOR THE RAPID OPTIMIZATION OF LLMS IN THE

SOFTWARE DEVELOPMENT LIFE CYCLE

UN ARTEFACTO DE SOFTWARE PARA LA OPTIMIZACIÓN RÁPIDA DE LLMS EN EL

CICLO DE VIDA DEL DESARROLLO DE SOFTWARE

1

2

3

Méndez Morales Vania Linette , Gómez Zea José Manuel , Jesús Magaña José Ángel ,

4

5

Hernández Cadena Alejandro , Javier Baeza Teresa de Jesús

https://doi.org/10.61117/ipsumtec.v9i1.457

1

3

4

5

Tecnológico Nacional de México /I.T. de Villahermosa, m24301262@villahermosa.tecnm.mx, https://orcid.org/0009-0003-5303-1898

Tecnológico Nacional de México /I.T. de Villahermosa, jose.gomezz@villahermosa.tecnm.mx, https://orcid.org/0000-0001-7474-5601

Tecnológico Nacional de México /I.T. de Villahermosa, joseangeljm@villahermosa.tecnm.mx, https://orcid.org/0000-0002-3010-6661

Tecnológico Nacional de México /I.T. de Villahermosa, alejandro.hc@villahermosa.tecnm.mx, https://orcid.org/0000-0003-2699-2696

Tecnológico Nacional de México /I.T. de Villahermosa, teresa.jb@villahermosa.tecnm.mx, https://orcid.org/0009-0007-7483-7072

Resumen-- La inteligencia artificial (IA) desempeña un papel

fundamental en el desarrollo de software moderno, transfor-

mando significativamente el diseño, la escritura, las pruebas y

el mantenimiento del código por parte de los desarrolladores.

En la actualidad, programadores de distintos niveles han inte-

grado herramientas basadas en IA en diferentes fases del ciclo

de vida del desarrollo de software (SDLC), desde la generación

de código hasta la implementación. Este estudio analiza el im-

pacto de estas tecnologías en la práctica profesional, identifica

las herramientas más utilizadas y propone las mejores prácticas

para la adopción responsable de la IA, con el objetivo de opti-

mizar su implementación de manera eficiente y ética. Como

parte de este estudio, se desarrolló un artefacto metodológico

para guiar la formulación estructurada de indicaciones, que fun-

ciona como un modelo para mejorar la precisión y la utilidad de

los resultados generados por la IA. Este artefacto se validó a

través de tres casos de uso de prueba de concepto (consultas

SQL, desarrollo de backend e implementación en AWS), lo que

demostró su potencial como base de conocimientos para los

equipos que buscan incorporar herramientas de IA de forma sis-

temática en sus flujos de trabajo.

Abstract-- Artificial Intelligence (AI) plays a key role in mod-

ern software development, significantly transforming develop-

ers’ design, writing, testing, and maintaining their code. Cur-

rently, programmers at various levels have integrated AI-based

tools into different phases of the software development life cy-

cle (SDLC), from code generation to deployment. This study

analyzes the impact of these technologies on professional prac-

tice, identifies the most used tools, and proposes best practices

for the responsible adoption of AI, aiming to optimize its im-

plementation efficiently and ethically. As part of this study, a

methodological artifact was developed to guide the structured

formulation of prompts, functioning as a model to enhance the

precision and utility of AI-generated outputs. This artifact was

validated through three proof-of-concept use cases (SQL que-

ries, backend development, and deployment in AWS), demon-

strating its potential as a knowledge base for teams seeking to

incorporate AI tools systematically into their workflows.

Keywords-- Artifact, artificial intelligence, software develop-

ment, prompt, programmers, tool.

INTRODUCTION

In the era of AI, software developers began to integrate multiple

AI-based tools into various phases of the Software Develop-

ment Life Cycle (SDLC) [1,2]. These applications include

Palabras clave-- Artefacto, desarrollo de software, herramien-

tas, indicaciones, inteligencia artificial, programadores.

Página 87

Reception: 05/02/2026 ǀ Acceptance: 27/04/2026 | Publication: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

automatic code generation, assistance in writing technical doc-

umentation, review, analysis, and error detection; enabling

them to optimize development time, automate tasks, and in-

crease the quality of final products. This trend represents a new

paradigm in software engineering, where the traditional human-

centered method is being renewed to collaborate with intelli-

gent agents [3].

the optimize creation of requests through prompts. This artifact

was adapted through proof-of-concept testing, of which three

implementations are presented in this article, corresponding to

SDLC phases: data modeling, code generation, and production

deployment of applications. The results were significantly sat-

isfactory, achieving 90% effectiveness in integrating AI-based

solutions into the SDLC.

Currently, tools such as GitHub Copilot, Amazon CodeWhis-

perer, Tabnine, DeepCode, Amazon CodeGuru, IntelliCode,

and even ChatGPT and Gemini represent a new form of collab-

oration between humans and language models [4,5]. These in-

novative tools stand out for reducing code errors, increasing

production speed, and generating robust solutions according to

various studies [6].

The article is organized as follows: Section 2 presents the mo-

tivation for this study; Section 3 describes related work that de-

fines the state of the art; Section 4 explains the methods and

materials used for data collection and analysis, as well as the

design of the artifact for prompt optimization; Section 5 pre-

sents the analysis of results and the practical implications of the

proof-of-concept tests; and finally, Section 6 summarizes the

conclusions, recommendations, and proposals for future studies.

As a special case, ChatGPT, Copilot, and Gemini have demon-

strated in recent years that they are versatile tools that can help

developers not only with programming tasks but also with un-

derstanding requirements, generating examples, optimizing

code, and explaining complex concepts in natural language

[4,5,]. However, despite their growing adoption, there are cur-

rently no formal guidelines to guide developers make optimal

use of them, nor universal standards that guarantee the quality

and ethics of the results obtained with these technologies. In

consequence, adopting these tools and obtaining the greatest

benefits depends on technical, organizational, and human fac-

tors [7].

MOTIVATION

Software development is among the professions and industries

revolutionized by AI tools. These tools include solutions to

modernize the software engineering process in a hybrid way;

that is, programmers stand in the need to incorporate innovative

methods to improve their productivity. In this context, artificial

intelligence emerges to transform the software engineering pro-

cess, but its adoption requires more than just using support ap-

plications, it involves developing a new collaborative work cul-

ture that employs ethical solution in the generation of manage-

ment methods; formalizing frameworks and workflows, organ-

izing knowledge management in development environments,

and understanding how, when, and which tools to apply within

the SDLC [8].

In this context, this article aims to analyze the best ways to in-

tegrate Artificial Intelligence into the software development life

cycle, considering both its capabilities and current limitations.

The purpose of this reflection is to contribute to the formulation

of a strategic and sustainable approach that allows maximizing

the potential of AI without compromising the integrity of the

software engineering process or the quality of the products gen-

erated. To achieve this, 150 active software engineering profes-

sionals were surveyed, including those working in the public

sector, private sector, freelancers, programming teachers, IT

professors, and professionals in the technology sector in general.

The survey results made it possible to identify the most used

tools in the software development life cycle. Based on this anal-

ysis, a method and a software artifact were designed to guide in

From an educational perspective, universities being the main

institutions that train programming professionals, need to un-

derstand methods for adopting AI tools that teachers can inte-

grate into their daily teaching. These methods are often difficult

to identify because they are empirical and software companies

adapt them during their processes. On the other hand, it is ob-

served that engineering students use AI tools inappropriately

when solving programming problems. This situation justifies

the need for a formal method or guideline that allows them to

use the tools correctly while also improving their learning.

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 88

From a technological perspective, self-taught junior program-

mers and future professionals in the software industry often

don’t have full access to the methods used by some companies

to integrate AI into software development across different

SDLC phases. This highlights the need for studies focused on

the appropriate use of AI tools.

although with notable limitations in understanding deep seman-

tics.Another relevant study [10], compares several automatic

code completion tools based on language models, including Co-

pilot, CodeWhisperer, and Tabnine. Their study revealed sig-

nificant differences in accuracy, contextual comprehension, and

compatibility among different programming languages. This

comparison showed that although these tools share a common

purpose, their effectiveness varies considerably depending on

the development environment, type of task performed, and pro-

grammer experience level.

From an industrial perspective, freelancers and small, medium,

and large software companies around the world, continue inte-

grating innovative AI methods and techniques into their work-

flows based on their own experiments. However, those who

have not yet done so may lose time, limit quality development,

or even lose money while trying to perfect this collaboration

with LLMs. For this reason, this study is justified.

On the other hand, in 2023 [11], presented a research study fo-

cused on analyzing the most frequent errors made by language

models in code generation. The study concluded that although

tools like Copilot or ChatGPT can be helpful, their outputs must

be critically evaluated to avoid structural and logical mistakes.

Additionally, organizations such as the OECD [12] and IEEE

have proposed classification frameworks aimed at ensuring

transparency, traceability, and accountability in AI-based sys-

tems, which are useful for evaluating the role of tools such as

ChatGPT within the software development life cycle.

Finally, this study aims to contribute to the development of

methodological strategies for programmers, researchers, and

software engineering companies that facilitate the progressive

and sustainable integration of AI tools into the SDLC. It seeks

to help them manage prompt requests across various LLMs they

use, while also enabling them to maintain a knowledge base of

prompts applied at different development phases, with the pur-

pose of reusing these artifacts in multiple solutions.

In 2023 [13], presented a study focused on identifying best

practices and AI tools integrated into the different phases of the

software development life cycle (SDLC), including design au-

tomation, code review, and implementation. The research high-

lighted that incorporating AI techniques improves productivity

and software quality through automated analysis and intelligent

support during system development. In 2024, the study “Heu-

ristics applied in artificial intelligence, a systematic review” [14]

aimed to analyze the role of heuristic methods in optimizing ar-

tificial intelligence processes. This work identifies how these

strategies contribute to improving the efficiency and accuracy

of intelligent models in various areas of technological develop-

ment. This approach is relevant to the present study, as it coin-

cides with the search for mechanisms that allow for a more ef-

ficient and structured use of artificial intelligence within the

software development life cycle. In 2024, Sajja and Thakur pre-

sented the study “Integrating Generative AI into the Software

Development Lifecycle: Impacts on Code Quality and Mainte-

nance” [15], in which they analyze the incorporation of gener-

ative artificial intelligence tools throughout the different phases

RELATED WORKS.

Traditionally, the field of software development has been tradi-

tionally dominated by human logic, experience, and structured

reasoning. However, with significant technological advance-

ments in recent years and the exponential emergence of AI, this

discipline has begun to integrating automated models to assist

programmers with simple tasks.

Despite this remarkable accomplishment of Goostman's devel-

opment team and the milestone in passing the famous Turing

test, the reaction of the scientific community was not unani-

mous. One of the most significant contributions comes from

researchers [9] who introduced Codex, the foundational model

behind the GitHub Copilot. This model, trained in vast amounts

of source code, demonstrated the ability to automatically gen-

erate code in various programming languages, addressing tasks

ranging from simple to complex. The study evaluated its effi-

ciency through competitive programming tests and found that

Codex solved approximately 37 % of the presented problems,

highlighting its potential as a developer assistance tool,

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 89

of the software development lifecycle. Their research highlights

how AI can support processes such as code generation, auto-

mated documentation, and early error detection, helping to im-

prove the efficiency and quality of the final product. The au-

thors also emphasize the challenges related to dependence on

these tools, validation of the results obtained, and interpretation

of the solutions generated.

established theoretical frameworks [1,2]. Figure 2 presents a

schematic representation of these phases highlighting where AI

can have a major impact or is already being actively applied.

Taken together, these studies show that the use of AI in software

development is both viable and an expanding reality. However,

these studies agreed that its implementation must be accompa-

nied by validation criteria, human review practices, and ethical

considerations. This article differs from previous works in that

it not only reviews current AI applications, but also proposes

concrete strategies for their optimal use, addressing both their

potential and limitations from a critical, evidence-based per-

spective.

Figure 2. Most well-known phases of software development.

Source. Own elaboration (2026).

Subsequently, the study’s research variables [16, 17] were es-

tablished and organized into five thematic categories for analyt-

ical purposes: participant profile, technical specialization, use

of AI tools, risk perception, and regulatory considerations, as

shown in Table 1.

MATERIALS AND METHODS

This study was structured into four phases, as shown in Figure

1. In the Investigate phase, a set of items was designed for the

data collection instrument. These items were developed based

on the phases of the SDLC, as well as the dependent and inde-

pendent variables of the case study. The Analysis phase allowed

us to identify the maximum and minimum values in the col-

lected data, thus facilitating a more precise interpretation of the

results. Subsequently, the Implementation phase consisted of

experimenting with the initially identified AI tools. Finally, in

the Generate phase, the best practice model was developed for

the appropriate use of Artificial Intelligence across the different

phases of the software development life cycle.

Table 1. Description of the study variables applied.

CATE-

VARIABLE

PURPOSE

GORY

Age Group, Geo-

Characterize the partici-

Participant graphical Location, pants based on their back-

Profile

Professional Status,

Years of Experience

ground, experience and

context.

Technical

Analyze how the area of

Specializa- Area of Development technical focus relates to

tion

the use of AI

Frequency and confi- Understand actual practices

AI Tools dence in AI generated and validation of AI-gener-

Code

ated code.

Identify key ethical, tech-

nical, and professional con-

cerns

Risk Per- Perceived risks asso-

ception

ciated with AI use

Assess the perceived need

for legal frameworks and

regulatory guidelines.

Regulatory Option on intellectual

Perspective property legislation

Figure 1. Defined phases of the study methodology.

Source. Own elaboration (2026).

Source: Own elaboration (2026)

Investigate Phase

.

During the research phase, the most representative phases in the

software development process were modeled, based on

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 90

Finally, the items that made up the results-obtaining [18, 19]

device were written, that is, the applied questionnaire. To en-

sure coherence and validity, the questions were inspired by in-

struments previously validated in related research studies [20,

21]. Each question was carefully designed to be clear and un-

derstandable to participants.

The questionnaire included closed-ended and multiple-choice

questions [22], enabling quantitative analysis of the data [23].

Standardization of the artifact also ensured that each item was

directly linked to a specific variable and corresponded to a par-

ticular phase of the software development process, thus facili-

tating subsequent analysis and interpretation.

Figure 3. Geographical location.

Source. Own elaboration (2026).

Implementation Phase

Analysis Phase

Based on the analysis carried out in this phase, a standardized

instrument was designed to optimize the use of artificial intelli-

gence tools in software development.

In the second phase, a data collection instrument was used (sur-

vey). This collection was carried out online through Google

Forms [24, 25], ensuring access to a diverse sample of partici-

pants both nationally and internationally.

In the context of AI-assisted software development, the way de-

velopers formulate their requests, also known as prompts

[26,27], is a determining factor for obtaining useful, accurate,

and coherent results. Unlike other automated tools, language

models such as ChatGPT, Gemini, and Copilot (to name a few),

do not infer user intentions; instead, they interpret the instruc-

tions contextually. Consequently, vague, incomplete, or poorly

structured prompts may lead to inaccurate or inapplicable re-

sponses. By contrast, clear, detailed, and contextualized

prompts significantly increase the effectiveness of the language

model’s response [28].

The questionnaire was developed based on the variables de-

fined in the research phase (participants’ profile, technical spe-

cialization, AI tools, risk perception, and regulatory perspec-

tive), which are directly related to the study’s objectives.

Each group of questions was designed to characterize the par-

ticipants’ profiles, analyze their practices in using artificial in-

telligence (AI) during different phases of the software life cycle,

and identify perceptions of risk, trust, and regulatory perspec-

tives regarding the use of generative AI tools.

The questions included closed-ended and multiple-choice items,

with frequency, single, or multiple selection scales, depending

on the type of variable evaluated. The results obtained from the

survey can be found in the results section.

In this regard, a set of recommendations and criteria are pre-

sented to improve communication with language models in a

standardized artifact, thereby facilitating efficient integration

into different phases of software development.

The instrument was applied to a total of 200 people divided into

two groups:

Rather than functioning merely as a documentation format, this

tool acts as a strategic asset that helps translate development

needs into technically detailed instructions for a language

model.

•

150 active professionals in the software development

field, with work experience in various sectors and re-

gions of the country.

The implementation of this artifact aims to ensure that, during

system development, the control of the phases where AI was

used and how it was applied is documented, in order to promote

a responsible, ethical, and verifiable use of AI within software

engineering processes.

•

50 programming students from the state of Tabasco,

Mexico.

Responses from the second group came from various states,

highlighting the state of Tabasco (55.5 %) and Mexico City

(10.5 %) having the highest participation (Figure 3).

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 91

Concrete and structured formulation of the request to be sent

to the linguistic model based on the above attributes, consid-

ering a detailed and step-by-step structure for best results.

The structure used in the User Stories format [29, 30] employed

in agile methodologies was taken as a reference for the design

of this format. This model served as the basis for defining the

fields of the artifact, allowing us to clearly record the stage of

the software life cycle where it is being implemented, the tool

used, the expected outcome, the context of the case, and the

prompt applied.

Source. Own elaboration (2026).

This structure made it possible to translate complex needs into

understandable, specific, and functional prompts [5, 31] for the

AI in use. Its implementation aims to reduce ambiguity in inter-

actions with AI models while also improving developer produc-

tivity and fostering continuous learning when working with in-

telligent systems. To validate the usefulness of the designed ar-

tifact, three real use cases related to common software develop-

ment tasks were applied.

The goal of the tool is to guide developers in formulating effec-

tive prompts and document the process through a series of ques-

tions that allow them to properly structure their requests to the

language model. Among the main questions that guide its ap-

plication are:

Proof of concept 1.

•

What am I doing?

The first case illustrates how AI can be optimized to generate

data queries in a database in PostgreSQL involving specific data

from several interrelated tables. This scenario involves an aca-

demic system that manages student payments across different

educational programs. The objective is to obtain a functional

SQL query that retrieves payment information for students, dis-

playing only those payments whose status is "Validated" or

"Under Review."

What do I need to accomplish with AI?

What attributes or details should the expected solution

include?

•

What environment, technologies, or languages am I

using?

What architecture, pattern, or file structure does the so-

lution require?

What exactly should the expected result do?

This test aimed to assess the accuracy of the model in handling

table relationships, applying conditional filters, and generating

an output format aligned with real system requirements.

Table 3. Artifact for generating SQL queries.

Table 2. Generated artifact for the creation of structured prompts

ARTIFACT FOR THE PROMPT

Artifact

ID

Author

General

Context

Phases of

SDLC

Specific

Context

Data

#

Case

Study

Date

#

ARTIFACT FOR PROMPT

Name

DD/MM/AAAA

Artifact

ID

Author

001

Case

Study

Date

A001

What am I working on?

Vania Méndez

06/04/2025

Where am I in the development process?

What do I need to perform?

General

Context

Phases

I’m developing a database that keeps student

payments

Data Base

of SDLC

Specific

Context

Relevant input structure (attributes, variables,

data type, etc.).

Stack of technologies being implemented.

I need to retrieve students enrolled in a specific

academic program, including only payments

that are validated or in process.

Source

Techno-

logy Stack

Data

Schema for tables student, payment, pay-

ment_status, payment_type and program

PostgreSQL

Element that I require to generate with the IA

according to the development environment

used (files, classes, methods, functions,

JSON, XML, etc.).

Methods to integrate, functions, etc. In this

section the user must be very specific as to

what is required, explaining in detail what the

AI is required to do.

Source

Techno-

logy

Output

Format

Stack

Functional

Require-

ment

Output

Format

Functio-

SQL Query

The query must be returned to the student’s

nal Re- code, full name, program name, payment

Language Model Instruction

quire-

ment

amount and date, only for payments where the

status is "validated" or "under reviews".

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 92

Language Model Instruction

Techno-

logy

Stack

Output

Format

Fun-

ctional

Requi-

rement

Spring Boot as development environment, Post-

greSQL as data access technology, JPA and li-

brary MultipartFile.

Repository, Service and REST Controller clas-

ses.

Repository with method to find student by code,

Service that saves student info and stores up-

loaded image via MultipartFile, Controller with:

POST to register a new student and GET to re-

trieve students by code.

I have a PostgreSQL database with the following tables:

•

Student (id, code, name, lastname, picture, gen-

der_id, program_id)

Payment (id, student_code, date, type_id, status,

amount, file)

Payment_status(id, status)

Payment_type (id, name)

•

Program (id, code, name, description)

Language Model Instruction

I need a SQL query to retrieve students enrolled in a specific

program, including their payments that are either "Validated"

or "Under Review". The result must show the student code,

full name (concatenation of name and lastname), program

name, payment type, payment amount, and payment date.

The query must use JOINs to reflect the key foreign relation-

ships, exclude payments with any other status, and be filtered

by program.

I am developing a school tracking system with Spring Boot;

the information is stored in a PostgreSQL database and JPA

is used for data access. I already have the entity “Student”

with the structure: id, code, first name, last name, program,

gender and photo of the student. The program field is a for-

eign key to the program entity and when storing a student,

the program code must be entered.

The gender field is a foreign key to the gender table. The ID

field is a self-incrementing field in the database.

Source. Own elaboration (2026).

I require: The creation of the project repository with a

method to search for students by code.

Proof of concept 2

The second functional case (Table 4) was developed for the

coding phase, where we sought to solve a common need in

backend projects, the registration of a user with file upload in

an application developed in SpringBoot.

A service that saves the information of a new student based

on the entity structure created, the student image will be

saved in the project path resources/images and you must im-

plement MultipartFile. A controller with a POST endpoint to

receive the student's data based on the created service, and a

GET endpoint to search for students by their code. The code

should be ready to be tested in Swagger.

The general context involved the development of a school

tracking system, with data persistence in a PostgreSQL data-

base and access through JPA [32], where the creation of the

necessary code sections to list the registered students and the

registration module of a new student following the SpringBoot

structure is required. Based on this requirement, the artifact is

structured to capture the request in a clear and complete manner,

Source: Own elaboration (2026).

Proof of concept 3.

The third case study focused on evaluating the use of artificial

intelligence as a key assistant in the complete configuration and

deployment of a backend application developed with Spring-

Boot on an AWS web server.

detailing the required structure for each requested file.

Table 4. Artifact for a service in Spring Boot.

ARTIFACT FOR PROMPT

This case represents a realistic scenario in which a programmer

needs to deploy a project in the cloud, starting from covering

everything from domain configuration to production deploy-

ment and verification using Swagger.

Artifact

ID

002

Case Study

A002

Author

General

Context

Phases

of

SDLC

Specific

Context

Bryan Scott

Date

05/04/2025

Table 5. Artifact for web application deployment.

I am developing a school payment control sys-

tem with Spring Boot.

Coding / Backend Implementation

ARTIFACT FOR PROMPT

Artifact

ID

003

Case Study

A003

Author

Joseph McA- Date

06/04/2025

llen

I need to implement backend logic to register a

new student, save a photo of them, and query stu-

dents using code.

Student Entity whit fields: id, code, first name,

last name, program (FK to program Entity, stores

program code), gender (FK to gender Entity) and

picture saved in resources/images

General

Context

Phases

of

I have a backend application developed in Spring

Boot 3.4.6, with Maven 4.0.0.

Deployment

Data

Source

SDLC

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 93

Specific

Context

I need to configure the complete deployment of

the application in AWS, from domain configura-

tion to EC2 setup and final Swagger test.

The application is a .jar file located in a local

folder. Uses PostgreSQL 16 as DB and needs to

be deployed behind NGINX in an EC2 instance

with Amazon Linux 2023.

Spring Boot 3.4.6, Maven 4.0.0, Java JDK 21,

PostgreSQL 16, AWS EC2, Amazon Linux 2023,

NGINX, WinSCP

Step-by-step instructions for, AWS domain and

DNS configuration, EC2 provisioning, Software

installation, Project upload and execution,

NGINX reverse proxy Domain linkage and

Swagger UI access

The system must be deployed successfully and

made accessible via a custom domain name

(chatgpt-prompt-model.com) through HTTPS

and tested via Swagger UI.

effectiveness for optimizing effective prompts for the imple-

mentation of AI tools during system development.

Survery Results

Data

Source

Participant Profile

The graphs in figure 4 show the age groups and professional

status of the people surveyed. The graph (a) shows that the larg-

est percentage of respondents (30 %) were in the age group of

18 to 23 years, followed by those aged 34 to 40 (23 %) and those

aged 24 to 28 (19 %). This suggests that the use of AI tools is

particularly prevalent among young adults.

Techno-

logy

Stack

Output

Format

The graphic (b) in Figure 4 shows that 29 % of respondents

were developers in the private sector, 25 % were programming

students, 14 % worked as freelancers, and 11 % were develop-

ers in the public sector.

Fun-

ctional

Requi-

rement

Language Model Instruction

I have a backend application developed in Spring Boot 3.4.6,

with Maven 4.0.0 and I need you to give me the steps for the

configuration of its deployment contemplating:

The detailed steps for creating a domain in AWS, which will

be named chatgpt-prompt-model.com, integrates the config-

uration to obtain the elastic IP, enable port 8080 and type “A”

records, one with www and the other with the IP address of

the project.

The detailed steps for the creation of a virtual machine in an

EC2 instance with Amazon Linux 2023 where java with JDK

21, PostgreSQL in its version 16, NGNIX as a reverse server

must be installed. The configuration must contemplate the

creation of remote keys, enable access protocols by https,

network configurations to connect from anywhere (SSH), in-

tegrates the steps for the configuration of the domain in

NGNIX. The project folder is located locally, integrates the

configurations to convert it to a .jar and the instructions to

upload it from winscp to the virtual machine. To test the op-

eration of the deployment, it should allow me to connect to

Swagger UI to test my API.

(a)

Source: Own elaboration (2026).

Results

This section presents the results obtained during the study,

(b)

which are divided into two main sections.

Figure 4. (a) Age range of participants; (b) Sector where the partici-

pants are located.

The first part presents the results of the survey conducted

among students and professors in the development area (estab-

lished in the analysis phase), aimed at identifying the degree of

adoption, perception, and use of artificial intelligence tools in

the different phases of the software life cycle.

Source. Own elaboration (2026).

Technical Specialization.

Figure 5 shows us the areas in which the surveyed developers

were focused: 42.9 % were Full Stack developers, 21.8 % were

backend developers, and 12.9 % specialized in database devel-

opment and administration.

The second part shows the proof-of-concept tests of the stand-

ardized artifact, developed to validate its functionality and

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 94

(9.65% each). This indicates a growing but still limited use of

AI at this stage.

The graphic (b) in Figure 6 demonstrates that most respondents

(84.21%) have adopted ChatGPT as a tool during the coding

phase, followed by GitHub Copilot (57%) and Gemini (33.3%).

Only 4.39% indicated they don’t use AI tools for code genera-

tion.

Figure 7 shows the AI tools used during the error detection

phase, where 64.91% of respondents use ChatGPT in this phase,

followed by GitHub Copilot (40.35%) and Gemini (14%).

Figure 5. Developers' area of expertise.

Source. Own elaboration (2026).

AI Tools.

Below are the results of the survey questions regarding the most

used tools in each established development phase. Iti’s im-

portant to notice that this section included multiple-choice an-

swers, so that users could select more than one response.

Figure 7. Most commonly used tools during the error detection

phase.

Source. Own elaboration (2026).

Figure 8 shows that 79.6 % of respondents do not use AI tools

during the deployment phase. However, among those who do,

13.16 % have implemented the use of GitHub Copilot for

DevOps as a support tool.

(a)

(b)

Figure 6. (a) Most commonly used tools during database schema

generation and design; (b) most used tools in the coding phase.

Source. Own elaboration (2026).

Figure 8. Most commonly used tools in the deployment phase of a

system.

The graph (a) in Figure 6 shows that most respondents (49.12%)

did not use AI tools for database development. Among those

who do, ChatGPT was the most popular (38.6%), followed by

GitHub Copilot (15.79%), DbDiagram.io, and SQL AI Copilot

Source: Own elaboration (2026).

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 95

Figure 10. Most commonly used tools for project documentation.

Source. Own elaboration (2026).

Finally, in Figure 11 we can observe that regarding AI-gener-

ated code requests, 67.5 % of respondents review AI-generated

code before using it, while 27.2 % trust the generated code but

run tests prior to implementation.

(a)

Figure 11. Confidence in the results obtained by AI.

Source. Own elaboration (2026).

(b)

Figure 9. (a) Most used tools during the testing phase; (b) Most used

tools in interface design.

Risk Perception.

All respondents were asked about the perceived risks of using

AI in software development (Figure 12). The most significant

concern (60 %) was the risk of over-dependence on the use of

AI. Other risks mentioned included loss of creativity (16.5 %),

potential production errors (14.5 %), and privacy issues (5 %).

These results indicate that developers critically reflect on the

implications of using AI for software development.

Source. Own elaboration (2026).

Figure 9 section (a) shows that during the testing phase, most

developers (80.7 %) did not use AI tools. Among those who do,

Testa.AI (11.4 %) and Testim (4.39 %) are the most used.

For interface design section (b) in Figure 8 shows that, the ma-

jority of respondents (71.05 %) don’t use AI tools. However,

among those who do, FigmaAI is the most used (14.9 %), fol-

lowed by Galileo AI and Designs AI (7.02 %).

Figure 10 shows that for the documentation phase in software

development, 55.26 % of respondents don’t use AI tools. On

the other hand, 31.58 % indicated to use ChatGPT and 23.68 %

use GitHub Copilot.

Figure 12. Perception of the risk of AI implementation.

Source. Own elaboration (2026).

Regulatory Perspective.

This section focuses on the ethical implications associated with

AI implementation in software development. [30] It highlights

the need for specific ethical considerations in the design, use,

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 96

and consequences of technologies, particularly in the era of Ar-

tificial Intelligence. Data privacy and equitable access are criti-

cal issues.

The results revealed that the most widely used tool during the

development phases included in the data collection artifact was

the ChatGPT, particularly in the coding and error detection

s0tages.

Regarding intellectual property, one question asked respond-

ents whether they believed there should be laws or legislation

regulating the ownership of the AI-generated code. More than

half of the respondents (52.5 %) agreed that laws or legal defi-

nitions should clarify the legal ownership of AI-generated code.

Meanwhile, 18.5 % believed that code generated by AI should

not have authorship implications, 16.5 % considered copyright

regulation unnecessary, and 12.5 % were uncertain about this

approach (Figure 13).

However, it’s important to clarify that ChatGPT is not a spe-

cialized coding assistant nor a tool created exclusively for de-

velopment field. ChatGPT is an AI-based language model

trained on large volumes of text, including documentation, code,

and technical conversations [33, 34], enabling it to simulate ex-

pert interactions in programming and other fields created by

OpenAI, which ich also known as generative AI [35] because

of its ability to produce original content.

Since its launch on November 30, 2022, ChatGPT has become

the fastest-growing application in history, reaching 100 million

active users just two months after its release [35]. ChatGPT op-

erates using generative pre-trained transformers (GPT), a type

of large language model (LLM). It relies on complex Machine

Learning algorithms [37] that compare the user input with its

pretrained data. The ChatGPT results were predictions based on

the patterns learned the training.

Similarly, the results reflect the need for a regulatory frame-

work for the implementation and use of AI in software devel-

opment, which takes into account the diverse contexts in which

these technologies are adopted.

Figure 13. Consideration of intellectual property laws.

Source. Own elaboration (2026).

Regarding broader regulations on AI use, 38.5 % of respondents

supported AI regulation due to potential risks to software qual-

ity, 20 % favored limited regulation, and 20.5 % believe AI

tools should be treated like any other development tool (Figure

14).

While a significant portion of respondents supported formal

regulation due to potential ethical and software quality risks, it

is essential to acknowledge those who advocate for flexible,

adaptive, and non-obstructive regulation. Allowing innovation

and full exploitation of AI capabilities without unnecessary re-

strictions.

Generate Phase

The experimental tests of the proposed artifact were carried out

using ChatGPT, a tool selected based on the survey results,

which identified ChatGPT as the most widely used artificial in-

telligence across the different phases of the SDLC.

For the implementation of the case studies, the free version of

ChatGPT 4.0 was used, executing each of the prompts designed

in the instrument. Subsequently, to verify the consistency and

stability of the results, the same tests were replicated using the

paid version (ChatGPT Plus, model 4.0).

Figure 14. Consideration of AI use regulations.

Source. Own elaboration (2026).

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 97

In both versions, the results obtained were identical in terms of

content, coherence, and output structure, which allowed us to

validate that the artifact maintains its effectiveness regardless

of free access or subscription to the platform.

implementing JOINS and complying with the specified require-

ments. Figure 15 presents the code fragment generated by the

artifact.

The model accurately identified the associations the data model

entities and applied filtering criteria to retrieve only transac-

tions corresponding to the specific payment types. Additionally,

despite not being explicitly requested, the model concatenated

the first and last name fields to construct the student's full name,

thereby optimizing the query (Figure 16).

Each of these tests focused on a different phase of the develop-

ment process: generating SQL queries, implementing backend

code, and deploying the application in a production environ-

ment. The main objective was to guide users in formulating pre-

cise and complete prompts to optimize LLM model answers in

each case. The following sections provide a detailed description

of each test, including its development, the results obtained, and

the technical considerations identified throughout the process.

Results of concept proof 1.

The first proof of concept aimed to evaluate ChatGPT's ability

to generate SQL queries based on the functional requirements

expressed in natural language. This includes the implementa-

tion and validation of the queries generated using the proposed

data model (Table 3).

The proposed scenario involved retrieving information on pay-

ments made by students to a school, considering only those with

status of “Validated” or “In process.

Figure 16. Implementation of results in PostgreSQL Database.

Source. Own elaboration (2026).

In terms of syntactic accuracy, logical coherence and clarity of

data output, the result is considered satisfactory, demonstrating

how a well-structured instruction can fully leverage the capa-

bilities of the AI model in specialized data querying tasks.

5.2.2. Results of concept proof 2.

The second proof of concept focuses on evaluating ChatGPT's

ability to generate functional backend code, using the Spring

Boot framework, adhering to its conventional file structure. The

scenario involved developing a module for student registration,

which included uploading of a profile image. The required

functionality included the implementation of a REST endpoint

capable of receiving both the student’s data and an image file,

storing the information in a database and saving the file in a

local folder, at a specified path. The clarity of the prompt (Fig-

ure 17), designed following best practices, was the key to ob-

taining an accurate and fully functional response.

Figure 15. Results obtained from the prompt.

Source. Own elaboration (2026).

The formulated prompt (Table 3) enabled the model to generate

a properly structured SQL query, that met the established re-

quirements, respecting the relationship of the tables by

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 98

Figure 19. Student Service archive generated by ChatGPT.

Source. Own elaboration (2026).

Figure 17. Prompt generated for code phase.

Source. Own elaboration (2026).

The generated prompt provided us with the components to gen-

erate the Repository, the Service, the Controller, and how to im-

plement Swagger into our project as requested, in a sectioned

manner. The codes can be seen in Figures 18 to 21. The gener-

ated structure was correct, and the AI understood the structure

of the relationships between the tables as provided, so its results

were generated with minimal errors.

Similarly, the model

shows how to modify the pom.xml file of our SpringBoot pro-

ject to include the necessary dependency to interact with Swag-

ger.

Figure 20. Student Controller generated by ChatGPT.

Source. Own elaboration (2026).

Figure 18. Student Repository archive generated by ChatGPT.

Figure 21. Swagger Implementation.

Source. Own elaboration (2026).

Source. Own elaboration (2026)

The request specifies the intention to perform tests using Swag-

ger. For this reason, the model demonstrates how to modify the

pom.xml file to include the necessary dependency for interact-

ing with Swagger.

Although the generated code was generally functional (Figure

22), several important issues were identified during the valida-

tion process. For example, the model changed the data type of

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 99

the gender entity from into long in the structure to save a student,

which did not match the original structure of the entity, causing

a mapping error. This type of automatic modification is evi-

dence that, although the model can generate robust structures,

it is necessary to carefully review the inferences made.

Figure 22. Backend testing in Swagger.

Figure 23. Prompt generated deployment phase.

Source. Own elaboration (2026).

This proof of concept confirms that ChatGPT is a valuable tool

for accelerating the development of common backend function-

alities, particularly in controlled environments or during the

prototyping phase. However, it is always recommended to thor-

oughly review the generated code, as potential issues such as

data typing, variable names, or error handling evidence the need

for the developer to maintain an active role during the integra-

tion of the generated code.

The response generated by the prompt was accurate and fol-

lowed a clear step-by-step procedure, using updated commands

along with security recommendations and best practices inde-

pendent of the requested points. Part of the generated instruc-

tions and results can be seen in Figures 24 and 25.

The code and commands were successfully tested, enabling the

replication of the environment on a real server without critical

errors, in the GitHub repository https://github.com/Va-

niaMendez/ChatGPTPromptModel you can find the file with

the complete results obtained with the prompt and the demo on

AWS.

Results of concept proof 3.

The third proof of concept aimed to assess the effectiveness of

ChatGPT in generating accurate and functional instructions to

enable the deployment of a backend application developed with

Spring Boot on an Amazon EC2 server running Amazon Linux

2023. The objective was to verify whether the model could as-

sist in configuring a production environment from scratch cov-

ering tasks such as installing dependencies, configuring ser-

vices, binding a custom domain, and exposing the system to the

network.

The prompt provided (Figure 23) included essential details such

as the operating system, the need to use the Spring Boot embed-

ded server (no external container such as Tomcat), and the need

to point the custom domain to the server.

Figure 24. Steps for configuring the domain in AWS.

Source. Own elaboration (2026).

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 100

In conclusion, this proof of concept demonstrated that ChatGPT

can serve as a useful knowledge base for application configura-

tion and deployment tasks, offering clear step-by-step guidance

that can act as a starting point for developers with intermediate

experience. Nevertheless, it is recommended that these instruc-

tions be complemented with official documentation, security

practices and testing in staging environments before final de-

ployment.

CONCLUSIONS

(a)

This work demonstrates the rapid advancement of language

models, such as ChatGPT, in emerging roles within software

development tasks, both at the academic level and in profes-

sional practice.

Data analysis indicated that approximately 81% of surveyed

software developers used the tool in critical phases including

coding, logical design, database queries, and technical docu-

mentation. ChatGPT is the most widely adopted AI in the cur-

rent software development landscape. However, 60% of re-

spondents expressed concerns about potential overreliance on

AI, and more than 50% highlighted the need for a regulatory

framework to accompany the integration of these tools into the

software development life cycle, thereby reflecting a collective

awareness of their ethical and professional impact.

(b)

Figure 25. (a) Steps for configuring the virtual machine; (b) Virtual

machine configuration.

Source. Own elaboration (2026).

The graphical analysis of the study variables (Table 1) revealed

that the majority of users were in the 18-23 age group, suggest-

ing a greater openness among younger individuals to adopt

these technologies.

To verify that our backend was deployed correctly, when enter-

ing the domain from a web browser, it should display the Swag-

ger testing panel, as shown in Figure 26.

Based on these findings, a methodological artifact was devel-

oped to guide and facilitate the application of language models

throughout the different phases of the software development

life cycle. In organizational contexts, this type of resource is

often referred to as a knowledge base: it documents and stand-

ardizes processes, practices, and lessons learned so that differ-

ent team members can reuse them. Its use enables the capitali-

zation of tacit knowledge and accelerates procedures, thereby

reducing the learning curve for new collaborators.

The artifact was subsequently tested under three real-world sce-

narios to demonstrate its practical effectiveness:

Figure 26. Application deployed and tested in Swagger.

Source. Own elaboration (2026).

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 101

The first proof of concept focused on generating a multi-table

SQL query. ChatGPT produced an accurate and actionable re-

sponse, proving that it can assist with database tasks, provided

that the prompt is well structured and detailed. It allows com-

plex data to be extracted easily, without the need for extensive

documentation or manual coding.

ensuring that the use of artificial intelligence remains transpar-

ent, justified, and aligned with their technological objectives.

However, it is important to emphasize that these tools are not

recommended for testing tasks, as they present limitations in

understanding complex logical contexts, identifying edge cases,

or generating reliable automated tests. Therefore, the validation

and quality assurance phases should remain the responsibility

of developers.

The second proof of concept involved generating a backend

module for a web application developed with Spring Boot, sup-

porting student registration and image upload. Despite the cor-

rect functionality, an error was detected in the data type of the

sex field, which changed from int to long. This highlights the

persistent margin of error and confirms that developers must

carefully review AI-generated code. This case underscores the

importance of being specific and clear regarding expected at-

tributes, relationships between tables or variables, and desired

behaviors in order to obtain a functional code.

Final Recommendations:

•

Use clear prompts that include technical context and

well-defined objectives.

Always validate the results delivered by the model,

even for simple tasks.

AI is integrated in phases such as design, coding, and

documentation, but not in testing.

We leveraged the benefits of an internal knowledge

base built from effective interactions with AI models.

Provide training for development teams on the critical

and responsible use of generative tools.

The third proof of concept involved requesting a detailed guide

for configuring a custom domain, creating an EC2 instance in

Amazon Linux 2023, installing Java 21 and PostgreSQL 16,

configure NGINX as a reverse proxy, and enable secure access.

It was demonstrated that, if the prompt is not sufficiently spe-

cific, ChatGPT may provide vague or unhelpful responses.

However, by using the methodological artifact as a reference, it

was possible to obtain a detailed and accurate sequence of steps

that facilitated the full setup—including the use of WinSCP to

upload .jar file and access the Swagger interface. This case

highlights that, for infrastructure-related topics, prompt preci-

sion is even more critical.

The methodological artifact presented in this work is the first

step toward the controlled and effective integration of these

models in real-world software development tasks, fostering col-

laboration between human expertise and intelligent systems

within the development workflow.

Finally, it is recommended that the use of ChatGPT and similar

models remain a support tool, not a replacement for profes-

sional judgment. The most effective way to leverage this tech-

nology is to craft detailed, structured, and contextualized

prompts, as proposed in the artifact. It is essential to explicitly

define what is being done, what is required, what attributes the

solution must have, what the technological environment is, and

what behavior is expected. These practices not only optimize

outcomes but also enable a more ethical, efficient, and profes-

sional use of artificial intelligence in the field of software de-

velopment.

The implementation of this artifact in business environments

represents a strategic opportunity to standardize the use of lan-

guage models in the software development life-cycle. Its struc-

tural approach enables development teams to integrate more ef-

ficient, systematic, and collaborative processes when interact-

ing with AI tools, reducing common errors in requirement for-

mulation and improving the traceability of technical decisions.

By integrating this artifact into DevOps environments, tech-

nical backlogs, documentation processes, or code reviews,

companies can not only accelerate productivity but also

strengthen the quality and consistency of the generated software,

Acknowledgements

The authors wish to note that a preliminary version of this work

was previously published as a preprint under the title “Model

for the Adoption of AI Tools in the Software Development Life

Cycle: A Framework for Prompt Optimization in LLMs” [37].

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 102

[2] S, Shylesh, AStudy of Software Development Life Cycle Process

Models (June 10, 2017). Available at SSRN: https://ssrn.com/ab-

stract=2988291 or http://dx.doi.org/10.2139/ssrn.2988291

[3] Treude C, Storey MA. Generative AI and Empirical Software En-

gineering: A Paradigm Shift. 2025 Feb 12 [cited 2025 Oct 18];

Available from: https://arxiv.org/pdf/2502.08108

All authors will be updated at each stage of manuscript pro-

cessing, including submission, revision, and revision reminder,

via emails from our system or the assigned Assistant Editor.

Data Availability Statement: The data that support the findings

of this study are openly available in Data from a study of AI

tools currently used by professional at https://zenodo.org/rec-

ords/, reference number 16975651.

[4] P. Maddigan and T. Susnjak, "Chat2VIS: Generating Data Visu-

alizations via Natural Language Using ChatGPT, Codex and

GPT-3 Large Language Models," in IEEE Access, vol. 11, pp.

45181-45193, 2023, doi: 10.1109/ACCESS.2023.3274199.

[5] J. M. Martínez Gómez, M. E. Higuera Marín, Y E. Aguilar Díaz,

«Enfoque Metodológico Para El Diseño De Interfaces Durante El

Ciclo De Vida De Desarrollo De Software», Gti, Vol. 12, N.º 34,

Pp. 59–73, Feb. 2014

Conflicts of Interest: The authors declare no conflict of inter-

est. Institutional Review Board Statement: This study, was

conducted in strict compliance with ethical principles and ap-

plicable legal regulations regarding personal data protection

and human participation in research. The “NO APPROVAL

REQUIRED” by an ethics committee or institutional review

board is justified given the scope and specific nature of the re-

search, which did not involve the collection of sensitive data or

the exposure of participants to physical, psychological, or social

risks. In accordance with the Federal Law on Protection of Per-

sonal Data Held by Private Parties (LFPDPPP), the collection

of personal data is lawful provided that the principles of legality,

consent, information, quality, purpose, loyalty, proportionality,

and responsibility are complied with. The research was de-

signed and implemented under these precepts, guaranteeing the

privacy and protection of respondents' data. Informed Consent

Statement: Participants in the study were clearly informed in

advance about the purpose of the processing of the information

(exclusive use for academic and scientific purposes) through an

introductory text on the digital survey platform. Participation

was voluntary, and the act of responding to the questionnaire

was interpreted as express consent on the part of the respond-

ents, who, by completing it, confirmed their understanding and

agreement with the terms presented. No images, recordings, or

sensitive data that could compromise the privacy of participants

were collected.

[6] S, Shylesh, AStudy of Software Development Life Cycle Process

Models (June 10, 2017). Available at SSRN: https://ssrn.com/ab-

stract=2988291 or http://dx.doi.org/10.2139/ssrn.2988291

[7] Treude C, Storey MA. Generative AI and Empirical Software En-

gineering: A Paradigm Shift. 2025 Feb 12 [cited 2025 Oct 18];

Available from: https://arxiv.org/pdf/2502.08108

[8] P. Maddigan and T. Susnjak, "Chat2VIS: Generating Data Visu-

alizations via Natural Language Using ChatGPT, Codex and

GPT-3 Large Language Models," in IEEE Access, vol. 11, pp.

45181-45193, 2023, doi: 10.1109/ACCESS.2023.3274199.

[9] Y. Luo, N. Tang, G. Li, J. Tang, C. Chai and X. Qin, "Natural

Language to Visualization by Neural Machine Translation,"

in IEEE Transactions on Visualization and Computer Graphics,

vol.

28,

no.

1,

pp.

217-226,

Jan.

2022,

doi:

10.1109/TVCG.2021.3114848.

[10] Alenezi M, Akour M. AI-Driven Innovations in Software Engi-

neering: A Review of Current Practices and Future Directions.

Applied Sciences 2025, Vol 15, Page 1344 [Internet]. 2025 Jan 28

[cited

2025

Oct

18];15(3):1344.

Available

from:

https://www.mdpi.com/2076-3417/15/3/1344/htm

BIBLIOGRAPHY

[11] Bejarano MH, Baquero-Rey LE, Bejarano MH, Baquero-Rey LE.

AI-Driven Software Development: A Paradigm Shift in Software

Engineering. 2025 May 27 [cited 2025 Oct 18]; Available from:

https://www.intechopen.com/chapters/1206853

[1] J. M. Martínez Gómez, M. E. Higuera Marín, Y E. Aguilar Díaz,

«Enfoque Metodológico Para El Diseño De Interfaces Durante El

Ciclo De Vida De Desarrollo De Software», Gti, Vol. 12, N.º 34,

Pp. 59–73, Feb. 2014

[12] Marar HW. Advancements in software engineering using AI.

Computer Software and Media Applications [Internet]. 2024 Feb

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 103

1 [cited 2025 Oct 18];6(1):3906. Available from: https://sys-

tems.enpress-publisher.com/index.php/CSMA/article/view/3906

[13] Chen, M., Tworek, J., Jun, H., Yuan, Q., De Oliveira Pinto, H. P.,

Kaplan, J., Edwards, H., Burda, Y., Joseph, N., Brockman, G.,

Ray, A., Puri, R., Krueger, G., Petrov, M., Khlaaf, H., Sastry, G.,

Mishkin, P., Chan, B., Gray, S., . . . Zaremba, W. (2021, 7 julio).

Evaluating Large Language Models Trained on Code. arXiv.org.

https://arxiv.org/abs/2107.03374

Asenjo, 14(1),

90-

93. https://doi.org/10.35434/rcmhnaaa.2021.141.905

[22] Hernandez Mendoza, S., & Duana Avila, D. (2020). Técnicas e

instrumentos de recolección de datos. Boletín Científico De Las

Ciencias Económico Administrativas Del ICEA, 9(17), 51–53.

https://doi.org/10.29057/icea.v9i17.6019

[23] Hamed Taherdoost. Data Collection Methods and Tools for Re-

search; A Step-by-Step Guide to Choose Data Collection Tech-

nique forAcademic and Business Research Projects. International

Journal of Academic Research in Management (IJARM), 2021,

10 (1), pp.10-38. ⟨hal-03741847⟩

[14] Zhang, B., Liang, P., Zhou, X., Ahmad, A., & Waseem, M. (2023).

Practices and Challenges of Using GitHub Copilot: An Empirical

Study. Proceedings/Proceedings Of The . . . International Confer-

ence On Software Engineering And Knowledge Engineering,

2023, 124-129. https://doi.org/10.18293/seke2023-077

[24] A. Borgobello, M. P. Pierella, and M.-I. Pozzo, “Ús de qüestiona-

ris en recerca sobre la universitat: anàlisi d’experiències amb una

perspectiva situada”, REIRE, vol. 12, no. 2, pp. 1–16, Jul. 2019.

[25] Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Mi-

chael Collins, Ankur Parikh, Chris Alberti, Danielle Epstein, Illia

[15] Song, D., Zhou, Z., Wang, Z., Huang, Y., Chen, S., Kou, B., ... &

Zhang, T. (2023). An empirical study of code generation errors

made by large language models. In 7th Annual Symposium on

Machine Programming.

Polosukhin, Jacob

Devlin, Kenton

Lee, Kristina

Touta-

[16] Papers, O. D. (22 de 02 de 2022). OECD Framework for the Clas-

sification of AI systems. Recuperado el 02 de 05 de 2025, de

https://www.oecd.org/en/publications/oecd-framework-for-the-

classification-of-ai-systems_cb6d9eca-en.html

nova, Llion Jones, Matthew Kelcey, Ming-Wei Chang, Andrew

M. Dai, Jakob Uszkoreit, Quoc Le, Slav Petrov; Natural Ques-

tions: A Benchmark for Question Answering Research. Transac-

tions of the Association for Computational Linguistics 2019; 7

453–466. doi: https://doi.org/10.1162/tacl_a_00276

[17] Soni, Arpita and Kumar, Anoop and Arora, Rajeev and Garine,

Ramakrishna, Integrating AI into the Software Development Life

Cycle: Best Practices, Tools, and Impact Analysis (June 10,

[26] Murat, P. (2020). Analysis of Multiple-choice versus Open-ended

Questions in Language Tests According to Different Cognitive

Domain Levels. Novitas - Journal Research on Youth and Lan-

guage, 14(2), 76-96.

2023).

Available

at

SSRN: https://ssrn.com/ab-

stract=4918992 or http://dx.doi.org/10.2139/ssrn.4918992

[18] Marcillo, F., Castillo Anzules, M. S., & Begnini, L. (2024). Heu-

rística aplicada en inteligencia artificial, una revisión sistemá-

[27] Fullstory. (03 de December de 2021). What is quantitative data?

How to collect, understand, and analyze it. Obtenido de Fullstory

BLOG: https://www.fullstory.com/blog/quantitative-data

tica. Revista

Científica

Kosmos, 3(2),

81–94.

https://doi.org/10.62943/rck.v3n2.2024.100

[28] Leyva López, Hermelinda Patricia, Pérez Vera, Monserrat Ga-

briela, & Pérez Vera, Sandra Mercedes. (2018). Google Forms en

la evaluación diagnóstica como apoyo en las actividades docen-

tes. Caso con estudiantes de la Licenciatura en Turismo. RIDE.

Revista Iberoamericana para la Investigación y el Desarrollo Edu-

cativo, 9(17), 84-111. https://doi.org/10.23913/ride.v9i17.374

[29] Pillajo Sotalín, Adriana Jovita (2019) Guía Digital Del Uso Del

Formulario De Google Forms Para La Evaluación En Básica Su-

perior Quito Uisrael, Maestría En Educación Mención: Gestión

Del Aprendizaje Por Tic Quito: Universidad Israel 2019, 127p.

[19] N. A. Sajja, N. D. Thakur, y N. A. Mehra, «Integrating Generative

AI into the Software Development Lifecycle: Impacts on Code

Quality and Maintenance», International Journal Of Science And

Research Archive, vol. 13, n.^o1, pp. 1952-1960, oct. 2024,

doi: 10.30574/ijsra.2024.13.1.1837.

[20] Villasís-Keever, M. Á., & Miranda-Novales, M. G. (2016). El

protocolo de investigación IV: las variables de estudio. Deleted

Journal, 63(3), 303-310. https://doi.org/10.29262/ram.v63i3.199

[21] Oyola-García, Alfredo Enrique. (2021). La variable. Revista del

Cuerpo Médico Hospital Nacional Almanzor Aguinaga

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 104

PhD. PhD. RENÉ CORTIJO UISRAEL-EC-MASTER-EDU-

378-242-2019-012

artificial intelligence to boost higher education», LATAM, vol. 4,

n.º 3, pp. 767–784, sep. 2023.

[30] Mora Naranjo, B. M., Aroca Izurieta, C. E., Tiban Leica, L. R.,

ánchez Morrillo, C. F., & Jiménez Salazar, A. (2023). Ética y Res-

ponsabilidad en la Implementación de la Inteligencia Artificial en

la Educación. Ciencia Latina. Revista Multidisciplinar, 7(6), 2054

- 2076. doi:https://doi.org/10.37811/cl_rcm.v7i6.8833

[31] Owusu, P., Raef, A., & Sharaf, E. (2025). Carbonate Seismic Fa-

cies Analysis in Reservoir Characterization: A Machine Learning

Approach with Integration of Reservoir Mineralogy and Poros-

ity. Geosciences, 15(7), 257. https://doi.org/10.3390/geosci-

ences15070257

[39] Belcic,

I.,

&

Stryker,

C.

(2025,

22

mayo).

ChatGPT. IBM. https://www.ibm.com/mx-es/think/top-

ics/chatgpt

[40] Bonteanu, A.M.; Tudose, C. Performance Analysis and Im-

provement for CRUD Operations in Relational Databases from

Java Programs Using JPA, Hibernate, Spring Data JPA. Appl.

Sci. 2024, 14, 2743. https://doi.org/10.3390/app14072743

[41] Méndez Morales, V. L.; Gómez Zea, J. M.; Jesús Magaña, J. Á.;

Javier Baeza, T. D. J.; Hernández Cadena, A.; de la Cruz Álva-

rez, J. Model for the Adoption of AI Tools in the Software De-

velopment Life Cycle: A Framework for Prompt Optimization

[32] Barney, N. (26 de August de 2024). What is language modeling?

Obtenido

de

TechTarget:

in

LLMs.

Preprints

2025,

2025091036.

https://www.techtarget.com/searchenterpriseai/definition/langua

ge-modeling

https://doi.org/10.20944/preprints202509.1036.v1

Table of Contribution Roles

[33] Menzinsky, A. L., Palacio, J., & Sobrino, M. A. (Agosto de 2022).

Historias de Usuario, Ingeniería de Requisitos Ágil. (S. Manager,

Ed.) Recuperado el 25 de 04 de 2025, de SafeCreative:

https://www.safecreative.org/work/2009135322450?0

[34] J. M. Gómez-Zea, J. Ángel Jesús-Magaña, J. de la Cruz-Alvarez,

E. Morales-Romero, y E. Sosa-Silva, «Optimización de la docu-

mentación en proyectos de software ágiles: Buenas prácticas y ar-

tefactos en el marco de trabajo SCRUM», ProgMat, vol. 15, n.º 3,

pp. 51–64, nov. 2023.

Role

Author(s)

Conceptualization

Methodology

Vania Linette Méndez Morales

«equal» - José Manuel Gómez Zea

«equal»

Project Administra- José Manuel Gómez Zea

tion

Supervision

José Ángel Jesús Magaña

Alejandro Hernández

«equal» - Teresa de Jesú Javier

Baeza

Visualization

Cadena

Writing

Draft)

(Original Vania Linette Méndez Morales

Writing (review and José Manuel Gómez Zea «equal» -

editing) José Ángel Jesús Magaña «equal»

[35] K. K. Ruiz Mendoza, «El uso de ChatGPT 4.0 para la elaboración

de exámenes: crear el prompt adecuado: The Use of ChatGPT

4.0 for Test Development: Creating the Right Prompt », LATAM,

vol. 4, n.º 2, pp. 6142–6157, ago. 2023.

Esta obra está bajo

una licencia internacional

Creative Commons Atribución 4.0

[36] Islam Sakib, M. S. (February de 2023). What is ChatGPT? Obte-

nido de researchGate: https://www.researchgate.net/publica-

tion/367794587_What_is_ChatGPT

[37] Ortiz, P., & Ortiz, P. (2025, 22 mayo). Chat GPT: qué es, para qué

sirve y su aplicación en la economía [explicado por Chat GPT].

EDEM Escuela de Empresarios. https://edem.eu/chat-gpt-que-es-

para-que-sirve-y-su-aplicacion-en-la-economia-explicado-por-

chat-gpt/

[38] M. E. Chávez Solís, E. Labrada Martínez, E. Carbajal Degante,

E. Pineda Godoy, y Y. Alatristre Martínez, «Inteligencia artificial

generativa para fortalecer la educación superior: Generative

Recepción: 05/02/2026 ǀ Aceptación: 27/04/2026 | Publicación: 30/05/2026

IPSUMTEC9 ǀ Volumen 9 –N°1 ǀ enero -junio 2026

Página 105