Software developer working on code across a laptop and a desktop monitor — A programmer at work. Photo: Emilija Manevska/Moment RF/Getty Images via CNN Newsource.

AI coding tools are now part of daily work for most developers. Yet a new Stack Overflow survey says six in 10 of them try to avoid using AI when they can.

That sounds like a contradiction. I run delivery for client projects across India and the USA, and to me it is the most honest picture of AI in software development this year.

In this post I break down what the survey found, why developers feel this way, and what agencies and their clients should do about it.

What the Stack Overflow Survey Found About AI Coding Tools

Stack Overflow shared its 2026 Developer Survey results with CNN ahead of the full release this week. The survey covers about 30,000 software developers in 169 countries. CNN's Lisa Eadicicco reported the headline numbers on October 5.

Here are the figures that matter most for teams that build software:

65% of respondents use coding assistants or agents.
Six in 10 say they try to avoid using AI.
A little over 74% say AI is useful for implementing code.
About 72% say it helps with debugging.
Only 34.4% find it useful for other work, such as communication and design.
Just 6.6% trust AI with important decisions.
About 78% say source attribution matters when they check AI-generated code.

So developers use these tools a lot, but mostly for narrow tasks. They do not hand over judgment.

Job happiness is slipping

The survey also shows rising unhappiness. Nearly one in three developers said they are not happy in their current role, up from 28.4% last year. About 45% called themselves "complacent," and only 22.3% said they are happy.

The top reasons were burnout, tech fatigue, economic uncertainty and unclear paths to promotion. Stack Overflow CEO Prashanth Chandrasekar told CNN that the pace of AI has made some of these worries worse.

"People have to completely retool their jobs, and so it is quite unnerving for folks to know exactly how much of what they're doing will be automated," he said.

How Developer Use of AI Coding Tools Has Grown

The 2026 numbers make more sense next to earlier years. Stack Overflow's data team published a look back at past Developer Survey findings on September 30, written by Erin Yepis.

Year	Developers using AI tools	Notes
2023	44%	First year Stack Overflow asked about AI tools
2024	62%	72.0% held a favorable view of AI tools
2025	79%	Favorable view fell to 59.7%; almost half used AI daily
April 2026 pulse	59% use AI agents	Up from 31% agent use in 2025

Use keeps climbing while enthusiasm keeps cooling. Negative views of AI tools rose from 6.4% in 2024 to 20.4% in 2025.

The tool mix is shifting too. In 2025, ChatGPT (82%) and GitHub Copilot (68%) led among out-of-the-box assistants. In Stack Overflow's April 2026 pulse survey, Claude Code use rose from 41% to 55%.

Stack Overflow Developer Survey retrospective artwork — Stack Overflow published a retrospective of its Developer Survey data ahead of the 2026 results. Image: Stack Overflow blog.

Why Developers Push Back on AI Coding Tools

The pushback is not about fear of new tech. The developers CNN spoke to have used AI for years. Their problem is context.

Software engineer Laura Housh started using AI for coding in 2024. Two years later she called it a love-hate relationship. "I've tried to use it for coding, and I just can't get behind it," she told CNN.

Audrey Eschright, an engineer with 20 years of experience, made a sharper point. Handing code writing to AI can leave a developer out of the loop on the solution. "Writing code is how you solve problems," she said.

Stack Overflow's own data backs this up. In its 2025 survey:

20.0% said AI made them less confident in their own problem-solving.
16.3% said it was hard to understand how or why AI code worked.
61.3% said they would still ask a person when they want to fully understand something.
75.3% said they would turn to a human when they do not trust AI's answers.

Bar chart from the 2025 Stack Overflow Developer Survey showing when developers expect to need human help over AI — When developers say they will still need people over AI. Source: Stack Overflow 2025 Developer Survey, data licensed under ODbL.

The pressure to keep up

There is also fatigue. Housh said she watches AI videos and reads platform docs in her own time and still struggles to keep up. "It's taxing because you have a lot of noise saying, 'Do this, do this, do this, or you'll get left behind,'" she said.

Chandrasekar added that the bar has been raised for juniors. Entry-level hires are now expected to work as if they have three years of experience. Senior developers, meanwhile, may feel the skills they built are less valued.

What This Means for Agencies in India and the USA

Agencies sit right in the middle of this tension. Clients read that AI writes code faster and ask why estimates have not dropped. Developers know the real gains are narrower and that review takes time.

The stakes are rising. CNN also cited a January Deloitte report covering more than 3,200 business and IT leaders. In it, 54% expected to move 40% or more of their AI experiments into production within three to six months. More AI code in production means more code someone must own.

For teams in India

Stack Overflow's 2025 data showed India had the highest share of unhappy developers among the countries it compared, at 32.7%. India also had the highest share of fully on-site workers, at 30.5%.

For Indian delivery centers, that is a warning about burnout. Adding AI tool mandates on top of tight schedules can make it worse. Rollouts need training time, not just licenses.

For teams and clients in the USA

US developers reported the highest happiness in 2025, at 28.7%, and the most fully remote work, at 45.0%. US clients often push hardest for AI-driven speed. They also tend to ask the most questions about code ownership and security, which is where trust gaps show up.

A Practical Playbook for Rolling Out AI Coding Tools

Here is the approach we use with client teams. It is built around what the survey says developers actually trust.

Start with the tasks developers already rate highly. Implementation help and debugging scored above 70%. Begin there instead of forcing AI into design or client communication.
Keep humans on every important decision. With only 6.6% trusting AI on big calls, architecture, security and data choices should stay with named engineers.
Track where code comes from. About 78% of developers want source attribution. Tag AI-assisted pull requests so reviewers know what they are looking at.
Review AI output like any other code. NIST's Secure Software Development Framework (SP 800-218) calls for code review and analysis to find vulnerabilities. That applies to AI-written code too.
Budget time for learning. Do not expect developers to learn new agents on weekends. Put training hours into the sprint plan.
Measure outcomes, not usage. Track bugs found in review, rework and cycle time. A high usage number means little if rework goes up.

Comparing review tools? See our breakdown of AI code review benchmarks from GitHub's ReviewBench. It shows how differently models perform on real pull requests.

Mistakes to avoid

Usage quotas. Forcing a set number of AI prompts per day breeds the exact resentment the survey describes.
Cutting estimates before you have data. Gains vary by task and codebase. Measure for a few sprints first.
Skipping junior mentoring. If juniors lean on AI without understanding the output, you build a team that cannot debug its own system.
Ignoring the people side. Burnout and unclear career paths were top complaints. Tools alone will not fix those.

How Clients Should Talk to Their Agency About AI

If you hire an outside team, ask direct questions. Good agencies will have clear answers.

Which AI coding tools does your team use, and for which tasks?
How do you review and test AI-generated code before it ships?
Who owns the final decision on architecture and security?
How do you flag AI-assisted work in pull requests?
Has AI changed your estimates, and what data backs that up?

These questions are part of any solid software development consulting engagement. They protect your codebase and set honest expectations on both sides.

Skills matter as well. Companies are paying more for people who use AI well, as we covered in our look at the AI skills salary premium for 2027. The best hires can explain their AI-assisted code, not just produce it.

What to Watch Next

Stack Overflow is releasing the full 2026 results this week. Its research team has flagged two open questions. Does wider agent adoption lead to real delegation? And does the gap between use and positive sentiment keep growing?

For agencies, the answer to both will shape pricing, hiring and training plans for 2027. The early signal is clear enough. Developers will use AI coding tools where they help, but they want control over the work that matters.

Key Takeaways

65% of developers use AI coding assistants or agents, yet six in 10 try to avoid AI when possible.
Trust is low. Only 6.6% trust AI with important decisions, and about 78% want source attribution.
AI helps most with implementing code and debugging, and far less with design and communication.
Unhappiness is rising, driven by burnout, tech fatigue and unclear career paths.
Agencies should roll out tools task by task, keep humans on key decisions and measure outcomes, not usage.

About the Author

Nitin Kumar is the Lead Project Coordinator at Debugged Pro. He coordinates development and QA teams delivering web platforms, ecommerce systems and AI integrations for clients across India and the USA, and he helps teams adopt AI coding tools without losing control of code quality.