APIX 2026 survey: IT teams and AI-written code

DB1 Itênex. Collected in May 2026, in São Paulo, Brazil.

The summary

The market does not reject AI, but it asks for control. In May 2026, we surveyed 103 IT professionals at APIX 2026. Overall, 95% trust AI-generated code: 76% with reservations and 19% fully. When we asked where they would use that code, 76% answered only with review before production. The top concern is security and data exposure, cited by 43%.

95% trust AI-generated code, with or without reservations
76% would only use it with review before production
43% name security and data exposure as their top concern

How the survey was run

The survey was run in person at the DB1 booth during APIX 2026, an API and integration event in São Paulo. Participants answered a three-question quiz by QR code and gave their name, work email, job title and company. Internal test records were excluded.

The final sample has 103 IT professionals from companies in the financial, industrial and services sectors.

Limits: this is a sample of people who stopped at a technology vendor's booth, not a probability sample of the market. Read the numbers as a portrait of this audience.

What the professionals answered

Do you trust AI-generated code?

Only 5% do not. Most trust it, but with reservations.

Answer	Respondents
I trust it with reservations	76%
I fully trust it	19%
I do not trust it	5%

Where would you use AI-generated code today?

Three in four ask for review before production. One in ten would go straight to production.

Answer	Respondents
With review before production	76%
Only in prototypes or tests	11%
Directly in production	10%
In no scenario	3%

What concerns you most about using AI in code?

Security, control and maintenance add up to 89% of answers.

Answer	Respondents
Security risks and data exposure	43%
Lack of control over what is generated	24%
Future maintenance problems	22%
Risks in critical systems	11%

Three maturity levels

The answers produced a score from 0 to 9, and each participant fell into one of three levels. Six in ten are still at the initial level.

Level	Respondents	How they use AI	What is missing
Essencial (score 0 to 6)	60%	With caution, in early evaluation.	A structured process to review generated code.
Potencializado (score 7 to 8)	25%	With judgment and structured review.	Taking the review process to the whole team.
Crítico (score 9)	15%	At the center of the development process.	Managing technical debt and maintenance of generated code.

How the scale works

The scale assesses three practices, each from 0 to 2 points, based on public engineering and security references. The first two questions are not scored: they compare your answer with the survey.

Practice	0 points	1 point	2 points	References
Clear rules	No. Each person decides alone	There is guidance, but not everyone follows it	Yes. They are written down, and the team knows and follows them	DORA · NIST AI RMF
Review with a record	No mandatory review	Someone reviews it, but there is no record of who approved	In small, versioned changes, with recorded approval	DORA · OWASP LLM05
Security and measurement	We do not check systematically	With tests or tools used case by case	With automated security tests and tracked indicators	NIST SSDF · NIST AI RMF

Levels by total points: Initial, 0 to 2. Structured, 3 to 4. Governed, 5 to 6.

This scale is a self-assessment created by DB1 Itênex, with our reading of these references. It is not a certification, has no statistical validation and is not endorsed by these institutions. Its levels differ from the three levels measured in the survey, so the percentages from the survey do not apply to it.

What changes from one role to another

The groups are small and the counts are approximate, and they do not add up to 103. Read this table as an indication, especially the rows with fewer respondents.

Role	Respondents	Top concern	How they would use it
Systems and IT analysts	about 28	Security (48%)	82% only with review
IT coordinators and managers	about 25	Security (40%) and lack of control (30%)	76% with review, 14% in production
Solution architects and tech leads	about 18	Future maintenance (35%) and security (32%)	70% with review, 20% in production
Developers and software engineers	about 15	Lack of control (38%)	80% with review
Product managers and product leaders	about 8	Critical systems (37%) and security (37%)	25% in production

What other sources say

International research points the same way: usage is growing, and so is the concern with quality and security.

76% of developers use or plan to use AI tools, and those already using them went from 44% to 62% in one year. Stack Overflow Developer Survey 2024
GitHub Copilot reached 20 million all-time users in July 2025, and Microsoft says 90% of the Fortune 100 use the tool. IT Brief, on Microsoft's quarterly results
29.5% of AI-generated Python snippets in GitHub projects had security weaknesses detected by static analysis. Fu et al., ACM TOSEM, 2025
68% of developers spend more time resolving AI-related vulnerabilities. Harness, State of Software Delivery 2025

What this means

What holds adoption back is not the technology. It is not knowing what AI can and cannot do in each context, and not being able to prove later who approved what. Reviewing everything by hand works at first, but it does not keep up with the volume. That is why we work with these answers:

What the survey showed	How we respond
24% worry about lack of control over what is generated	Harness: defines what agents can and cannot do in your context.
76% require review before production	Human gates: AI proposes, a person approves, and the approver has a name.
43% name security as their top concern	Human review with security tests in the stage itself, before production.
22% worry about future maintenance	Continuous monitoring in production, to keep technical debt under control.

See how this works in the Agentic SDLC consulting

How to cite

DB1 Itênex. The technology market and AI-written code: a field survey at APIX 2026. São Paulo, May 2026. https://www.db1.com.br/en/pesquisas/ia-em-codigo-apix-2026

Frequently asked questions

Do IT professionals trust AI-generated code?

In the DB1 Itênex survey at APIX 2026, with 103 IT professionals, 95% said they trust it: 76% with reservations and 19% fully. Another 5% do not trust it.

Would IT professionals put AI code straight into production?

Few would. In the APIX 2026 survey, 10% would use it directly in production, 76% only after review, 11% only in prototypes or tests and 3% in no scenario.

What is the top concern about using AI in code?

Security and data exposure, cited by 43% of the 103 respondents. Next come lack of control over what is generated (24%), future maintenance (22%) and risks in critical systems (11%).

How was the survey run?

In person, at the DB1 booth during APIX 2026, in May 2026, with a three-question quiz by QR code. The final sample has 103 IT professionals and is not a probability sample.

Want to go deeper?

Get the full survey report, with the analysis by role and what changes when AI agents have rules.

The technology market and AI-written code.