AI coding tools have moved from experiment to everyday workflow. A Google Cloud survey says 90 percent of tech professionals now use AI tools at work, up 14 points from last year.
That does not mean teams have fully solved how to use them. The picture that emerges is more complicated: AI is helping people ship and work faster, while trust, reliability, and learning remain open problems.
AI is now part of the workday
The study covers a broad group of tech professionals, including developers and product managers. Most respondents say AI is now a regular part of how they work, with a median of two hours a day spent using these tools.
Dependence is also becoming normal. The survey says 65 percent are heavily reliant on AI. Within that group, 37 percent describe their dependency as "a moderate amount," 20 percent as "a lot," and 8 percent as "a great deal."
Those numbers point to a practical shift in software development. AI is no longer only a side tool for isolated tasks. It is becoming part of planning, coding, review, and delivery workflows across teams.
For leaders, that creates a new management problem. If many workers are using AI every day, organizations need to understand not just whether AI is present, but how it changes output, review habits, team learning, and software stability.
Productivity is rising, but confidence is not
The strongest argument for AI adoption is productivity. More than 80 percent of respondents say AI makes them more productive, and 59 percent report better code quality.
The survey also found that teams using AI are shipping more software and apps. That is a change from last year, when no clear link between AI and productivity was visible.
Still, the same report shows that trust has not kept pace. Only 24 percent say they have "a lot" or "a great deal" of trust in AI. At the other end, 30 percent trust AI outputs "a little" or "not at all."
The report calls this the "trust paradox": people keep using AI even when they are not fully confident that its answers can be relied on. In software development, that matters because a tool can feel helpful while still creating work that must be checked carefully.
Reliability is the main pressure point. The source article notes that keeping software stable remains a major challenge. That means AI-assisted development cannot be judged only by how quickly code appears or how much work feels automated.
The learning problem is becoming harder to ignore
The DORA report includes a critical view from Matt Beane, a professor at UC Santa Barbara. Beane argues that AI can give teams a short-term boost while weakening long-term skill development.
The concern is especially important for junior developers. Traditionally, they learned from senior engineers through pair programming and direct problem solving. According to Beane, AI can interrupt those learning channels.
As automation takes over more work, junior developers may be pushed away from the practical experience that helps them grow. Beane found similar patterns in 31 professions, where AI interrupts traditional learning paths and makes it harder for new people to build crucial skills.
The issue is not simply whether AI helps a team move faster today. It is whether the team is still producing the human expertise it will need later.
Beane suggests that teams should balance productivity with skill growth. One proposed fix is to track how developers use AI, connect that data to version control, and measure real learning outcomes.
AI reflects how teams already work
The Google Cloud study describes AI as both a "mirror and multiplier" inside organizations. In teams that are already well organized, AI can increase efficiency. In teams with structural problems, it can make those weaknesses easier to see.
The report groups teams into seven archetypes, ranging from "harmonious high-achievers" to teams caught in a "legacy bottleneck." Google Cloud also introduced the DORA AI Capabilities Model, which lays out seven technical and cultural factors that shape effective AI adoption.
This framing is useful because it avoids treating AI as a simple upgrade. The same tool can have different effects depending on the team using it. Good workflows may become faster, while unclear processes may become more exposed.
For software organizations, that means AI adoption is not only a tooling decision. It also tests engineering culture, review discipline, collaboration, and the ability to keep software reliable while development speeds up.
Other signals point to the same disconnect
The source article notes that other studies show a similar gap between AI usage and AI trust. The Stack Overflow Developer Survey 2025 found that 84 percent of developers use or plan to use AI tools, but only 33 percent trust the code these tools generate.
The leading complaint in that survey was that AI code is often "almost right, but not quite." That phrase captures a central problem for everyday AI coding: outputs can be useful enough to adopt, but flawed enough to demand careful human review.
A randomized METR study found another complication. Experienced open-source developers using AI assistance took 19 percent longer on average, even though they felt like they were working faster.
At the same time, AI coding systems are improving. At the ICPC World Finals 2025, an OpenAI system solved every task, outperforming both top human teams and Google's own model. Neither system is available to the public yet.
The risks are also evolving. New research from 2025 found that Deepseek often produces unsafe code when handling politically sensitive topics. The source article also notes that about one in five AI-generated code snippets includes fake libraries, a technique called slopsquatting, which can create openings for supply chain attacks.
The lesson is not that AI coding tools should be ignored. The lesson is that their value depends on review, team structure, and the ability to protect learning while gaining speed. AI-assisted software development is now mainstream, but confidence still has to be earned.