The surprising ways AI helps strong dev teams and hurts weak ones, according to Google

DamnwellMedia/iStock/Getty Images Plus

Follow ZDNET: Add us as a preferred source on Google.

ZDNET's key takeaways

• Nearly all developers now rely on AI tools.
• AI amplifies strengths and magnifies dysfunction.
• High-quality platforms are a must for AI success.

Google released its 2025 DORA software development report. DORA (DevOps Research & Assessment) is a research program at Google (part of the Google Cloud organization). DORA explores the capabilities and factors that drive software delivery and operations performance.

This year, the DORA project surveyed 5,000 software development professionals across industries and followed up with more than 100 hours of interviews. It may be one of the most comprehensive studies of AI's changing role in software development, especially at the enterprise level.

Also: 10 ChatGPT Codex secrets I only learned after 60 hours of pair programming with it

This year's results are particularly relevant because AI has infiltrated software development to a rather extreme degree. The report shows some encouraging notes but also showcases some areas of real challenge.

In writing this article, I've gone through the 142-page report and pulled five major observations that cut through the hype to reveal what's really changing in software development.

1. AI is now widely used in development

According to survey respondents, somewhere between 90 and 95% rely on software development for work. The report mentions 95% in the intro and 90% later in a detail section, but regardless of which number you choose, nearly all coders are now using AI. According to the report, this is a 14% jump from last year.

The median time spent interacting with an AI was two hours per day. There's a bit more nuance to this, though. For example, only 7% of respondents "always" report using AI when faced with a problem to solve. The largest group, 39%, report "sometimes" turning to AI for help. But what struck me is that a full 60% use AI "about half the time" or more when trying to solve a problem or complete a task.

Eighty percent of programmers reported an overall increase in productivity, but only 59% reported that their code quality improved. Another key metric is this: 70% of respondents trust the AI's quality, while 30% don't.

Also: I got 4 years of product development done in 4 days for $200, and I'm still stunned

Let me share a personal thought on this. I just finished a massive coding sprint made possible by the AI. The code that came out was almost never right on the first run. I had to spend a lot of time cajoling the AI to get it right. Even once the work was done, I went back to do a full QA sweep, where I found more errors.

My conclusion is that there is no way I could have gotten anywhere near the amount of work done I just did without AI. But there's no way in heck I'm going to trust any code the AI writes without doing a lot of review, validation, and testing. Of course, that's not much different from how I felt when I was a manager and delegated coding to employees or contractors.

2. Think of AI as an amplifier

This was one of the more fascinating results coming out of the study. The DORA team contends that AI has become an amplifier. Essentially, AI "magnifies the strengths of high-performing organizations and the dysfunctions of struggling ones."

That makes so much sense. If you read my most recent article on "10 ChatGPT Codex secrets I only learned after 60 hours of pair programming with it," I pointed out that AIs make big mistakes quickly. One malformed prompt can send an AI off to wreak some major destruction. I had the experience where Codex decided to delete a large chunk of one of my files, and then immediately checked in those changes to GitHub.

Also: I did 24 days of coding in 12 hours with a $20 AI tool - but there's one big pitfall

Fortunately, I was able to roll those changes back, but I saw a massive amount of work vanish faster than I could take a sip of coffee.

Essentially, the more effective and organized a team is, the more AI will help. The more scattered or haphazard a team is, the more AI will hurt. In my case, I have really good revision control practice, so when the AI ate my homework, I was able to get it back because of controls I had put in place before I ever gave the AI its first access to my codebase.

3. Seven team archetypes in the AI era

So who wins and who loses? The DORA team identified eight factors that determined a team's overall performance.

1. Team performance: Effectiveness and collaborative strength of a team
2. Product performance: Quality and success of products being produced
3. Software delivery throughput: Speed and efficiency of the delivery process
4. Software delivery instability: Quality and reliability of the delivery process
5. Individual effectiveness: Effectiveness and sense of accomplishment for individual team members
6. Valuable work: Degree to which individual team members feel their work is valuable
7. Friction: How much gets in the way of individuals trying to get their work done
8. Burnout: Feelings of exhaustion and cynicism among team members

Then they measured these factors against respondents and their teams. This helped identify seven team archetypes.

1. Foundational challenges: Survival mode, gaps everywhere
2. Legacy bottleneck: Constant firefighting, unstable systems
3. Constrained by process: Stable but bogged by bureaucracy
4. High impact, low cadence: Strong output, unstable delivery
5. Stable and methodical: Deliberate pace, consistent quality
6. Pragmatic performers: Reliable, fast, moderately engaged
7. Harmonious high-achievers: Sustainable, stable, top performance

AI, says the report, is a mirror of organizations. Using AI makes the strengths and weaknesses of teams more apparent. But what I found particularly interesting is the idea that the "speed vs. stability" trade-off is a myth.

This is the idea that you can be fast or you can produce good code, but not both. As it turns out, the top 30% of respondents fall into the harmonious high-achievers or pragmatic performers archetypes, and those folks are producing output quickly, and the quality of that output is high.

4. Seven key practices

The report stresses, "Successful AI adoption is a systems problem, not a tools problem." The DORA folks seem to like the number seven. They say the following seven key practices drive AI's impact (for good or bad).

1. AI policy: An organization's clear, communicated AI stance.
2. Data ecosystems: Overall quality of an organization's internal data.
3. Accessible data: AI tools connected to internal data sources.
4. Version control: Systematic way to manage changes to code.
5. Small batches: Breaking changes into small, manageable units.
6. User focus: Teams prioritizing the end users' experience.
7. Quality platforms: Shared capabilities available across the organization.

As you might imagine, the successful teams employ more of these practices. While the unsuccessful teams might have highly productive individual programmers, it's the lack of these fundamentals that seem to bring them down.

They recommend, "Treat your AI adoption as an organizational transformation. The greatest returns will come from investing in the foundational systems that amplify AI's benefits: your internal platform, your data ecosystem, and the core engineering disciplines of your teams. These elements are the essential prerequisites for turning AI's potential into measurable organizational performance".

5. Two factors that influence AI success

Last year, it became fairly big news when the previous DORA report showed that AI actually reduced software development productivity, rather than increased it. This year, the opposite is true. The DORA explorers were able to identify two key factors that turned those results around.

Development organizations are more familiar with AI and know how to work it more effectively than they did a year ago. The study shows that 90% of developer organizations have adopted platform engineering. This is the practice of building strong internal development platforms that aggregate the tools, automations, and shared services for a development team.

Also: The best AI for coding in 2025 (and what not to use)

According to DORA, when the internal platform works well, developers spend less time fighting the system and more time creating value. If you view AI as an amplifier, then you can see how good systems can really improve results. Interestingly, if platforms are weak, AI doesn't seem to improve organizational productivity. Good internal platforms are a very clear prerequisite to effective AI use.

The next factor seems like a buzzword out of a workplace sitcom but is really quite important. It's VSM (or value stream management). The idea is that managers create a map of how work moves from idea to delivery. It's basically a flowchart for operations rather than just bits.

By seeing every step, teams can identify problem areas, like very long code reviews or releases that stall at various stages. The report states that the positive impact of AI adoption is "dramatically amplified" in organizations with a strong VSM practice. For the record, the word "dramatically" appears in the report four times.

The report states, "VSM acts as a force multiplier for AI investments. By providing a systems-level view, it ensures AI is applied to the right problems, turning localized productivity gains into significant organizational advantages instead of simply creating more downstream chaos."

What it all means for software development

There are a few clear conclusions from the report. First, AI has moved from hype to mainstream in the enterprise software development world. Second, real advantage isn't about the tools (or even the AI you use). It's about building solid organizational systems. Without those systems, AI has little advantage. And third, AI is a mirror. It reflects and magnifies how well (or poorly) you already operate.

What do you think? Has your organization been using AI tools in software development? Do you see AI as a genuine productivity boost or as something that adds more instability? Which of the seven team archetypes feels closest to your own experience? And do you think practices like platform engineering or VSM really make the difference? Share your thoughts in the comments below.

You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.