AI Agents Get Cloud Access Despite Nearly Universal Developer Concern

The cloud access contradiction is now measurable. A survey found that 97% of developers have concerns about AI agents entering cloud environments, yet 79% grant them access anyway. Mandatory platforms deliver twice as often, yet teams are dropping them — a neat reminder that output and trust remain different currencies.

LocalStack released the 2026 State of AI in Cloud Development report, conducted by Dimensional Research, after surveying more than 300 developers and team managers. Respondents worked in software development at organizations with 1,000 to more than 10,000 employees and were responsible for shipping applications to public cloud platforms.

The access pattern reaches across development, testing and staging. The survey found that 79% of respondents give AI agents access to cloud environments for at least one of those stages, and 18% said agents always have access.

Production is not off-limits either. Twenty-six percent said AI agents can access production environments, while 55% said agents can access staging — exactly the sort of arrangement that turns a productivity experiment into an incident review.

More code, less confidence

The survey also shows why teams keep granting access despite the risks. Sixty-five percent of developers generate at least twice as much code since adopting AI, but they estimate that only 61% of AI-generated code reaches production.

That gap reflects a testing and validation problem, not a shortage of keystrokes. Twenty-six percent of developers said AI-generated code has significantly more bugs than code created by humans, while only 6% of development team managers said the same.

Respondents cited leaked credentials, misconfigurations, rising cloud costs and failures in production among their concerns. Fifty-eight percent said leadership expects them to control token and cloud costs without giving them the information needed to do so.

“These responses align with what we hear from our customers every day,” said Waldemar Hummer, LocalStack co-founder and CTO. “AI is generating code faster than teams can confidently ship it, making testing and validation the new bottleneck.”

Dimensional Research Principal Diane Hagglund said the research extends understanding of AI’s effect across testing, staging and deployment, not only code writing. Hummer said LocalStack helps developers and AI agents use fast, safe, isolated cloud environments without the risks and costs of traditional cloud infrastructure.

AWS becomes an operating-model test

The same pressure is visible in Asia Pacific, where enterprises are reshaping AWS environments to support expanding AI use while meeting local regulatory requirements. The 2026 ISG Provider Lens AWS Ecosystem partners report for Asia Pacific says companies are using AWS to modernize technology estates, strengthen data foundations and prepare for production-scale AI deployments.

Companies are deploying GenAI within AWS environments for knowledge assistants, customer service automation, software engineering and industry-specific applications. Amazon Bedrock is now available in more locations across the region, helping organizations address data residency, latency and control requirements.

Cloud growth brings a less glamorous question: what does each workload cost? CIOs and CFOs are scrutinizing AWS unit economics, idle resources, licensing and workload placement, pushing FinOps, cloud optimization, automation and managed AWS operations higher up the priority list.

Michael Gale, partner and regional leader of ISG Asia Pacific, said effective AWS strategies require country-level capabilities with a shared operating backbone. Srinivasan P N, ISG senior lead analyst and lead author of the report, said enterprises are evaluating whether their operating models can absorb AI growth and regulatory change.

The report evaluates 42 providers across AWS Professional Services, AWS Managed Services, AWS Data Lifecycle Management Services and AI-Powered Business Transformation with AWS. Accenture, Capgemini, Cognizant, DXC Technology, HCLTech, PwC, TCS and Wipro are named Leaders in all four quadrants.

NTT DATA is named a Leader in three quadrants, while Infosys and Tech Mahindra are Leaders in two each. ITC Infotech is named a Leader in one quadrant; LTM is recognized as a Rising Star in three quadrants, while Hitachi Digital Services, Kyndryl, NTT DATA and Tech Mahindra receive Rising Star recognition in one quadrant each.

MHP is the global ISG CX Star Performer for 2026 among AWS ecosystem providers, earning the highest customer satisfaction scores in ISG’s Voice of the Customer survey, part of the ISG Star of Excellence™ program.

Underneath both reports sits a quieter management problem: ownership. Luis Aburto, Founder and CEO of Scio, says companies often build applications and never revisit who should maintain them long-term, leaving the core engineering team responsible because it knows the code best.

That arrangement confuses familiarity with fit. The engineering resources needed to maintain a mature application usually differ from those needed to build it, just as the controls needed to validate AI-generated code differ from the controls needed to produce it.

The pattern is plain enough: AI expands output, cloud environments absorb the workload, and leadership demands cost control. Teams still need clear ownership, isolated environments and validation before faster delivery becomes an expensive way to ship uncertainty.

Stay connected via Google News

Follow us for the latest AI News