The big story in developer productivity in 2024 was the widespread adoption of generative AI for software development.
But was GenAI a game-changer for developer experience — or will it be in 2025?
It’s complicated.
The state of developer experience in 2024 was marred by a mismatch between leadership expectations and what developers actually want and need. Developers complained — and have been complaining for over a decade — that their biggest impediments to productivity are technical debt and documentation.
Instead of large-scale strategies to fix this, enterprise leadership is planning to increase spending on AI again in 2025. Yet developers still aren’t seeing even marginal gains from 2024’s heavy investment in AI-generated code. In fact, according to the 2024 DORA report, speed and stability have actually decreased due to AI.
This past year has been “companies fixing the wrong problems, or fixing the right problems in the wrong way for their developers,” as Andrew Boyagi, head of DevOps evangelism at Atlassian, told us.
The New Stack spoke to more than a dozen developers and developer advocates to learn what they actually want in 2025, in terms of improving developer experience and productivity. Unsurprisingly, AI was at the top of their minds.
Take their responses as a warning to IT leadership: don’t waste your money and your tech talent on feeble initiatives that nobody wants. And consider their advice about just where AI can truly help developers.
Here’s what our experts see coming for DevEx and productivity in 2025:
New Security Risks Emerge for AI
Just about everyone we spoke to is waiting for the other shoe to drop on the security of AI. There’s more code being created than ever, they noted, with fewer humans in the loop to review it, making it only a matter of time before more vulnerabilities infiltrate the tech industry and its code.
In 2025, generative AI “will continue to put more pressure on dev teams to secure their software,” Brian Demers, developer advocate at Gradle, an open source build system, told The New Stack. “The technology has the potential to introduce greater security risks and vulnerabilities in the software development process due [to] either poorly generated code or the increased amount of code.
“Cases like the xz utils backdoor will only become more common, as developer teams have more on their plates. As a result, fast and frequent vulnerability responses will become critical and developers will need access to better tooling to manage and validate AI-generated code to avoid new risks.”
Certainly, compared with our 2024 predictions, developers have become even more cynical and frankly concerned about the return on investment of AI in software development.
“In 2025, companies will learn what happens when their codebases are infiltrated with AI generated code at scale,” Winston Hearn, senior technical product manager at Honeycomb.io, told The New Stack.
“Everyone was excited about using [GitHub] Copilot and similar tools to improve developer productivity, but no one asked what happens when a significant amount of code was generated and not fully understood or reasoned about by humans, with the full context of a system and the desired goals,”
Maybe the increasing embrace of AI code generation will work out fine, Hearn allowed. But it’s more likely, he said, that “companies are about to learn that improving velocity of shipping by sacrificing developer knowledge, experience and context of the system will lead to bigger incidents with slower resolution times, as there will be fewer experts on hand to quickly understand and resolve issues.”
Observability Will Need to Shift Further Left
Another struggle the industry will continue to face is that AI, and the large language models trained for it, are generally locked boxes. Not only will there be even more code generated in 2025, it will be harder to trace when there’s a bug or vulnerability. This will be the impetus, Demers predicted, for organizations to finally address observability in the developer toolchain.
“As distributed systems, microservices and AI-driven code generation make development environments more complex, pinpointing issues like bottlenecks, test failures and errors will become more challenging without robust observability tools,” he said. “Greater visibility into these processes will be integral for maintaining efficiency and delivering quality software.”
AI also presents an opportunity to address risks, such as assessing the risk introduced by a proposed change in the code. This is already being explored at Meta, where the importance of a change and the centrality of the developer is automatically considered to assign the level of code review needed.
Building at Scale Will Get More Complicated
AI is changing the way — and where — developers work. And that will only intensify in 2025.
“AI will enable developers to write code much faster, which means more deployments, more bugs, more incidents, more security outages, more tickets, more infrastructure — essentially more of everything,” said Zohar Einy, CEO of Port. “Developers will need an effective way to manage code operations, while ensuring compliance with organizational requirements.”
Generative AI will affect not only how developers do their work but also where infrastructure investment will go, Trisha Gee, lead developer advocate at Gradle, told The New Stack.
“Companies are about to learn that improving velocity of shipping by sacrificing developer knowledge, experience and context of the system will lead to bigger incidents with slower resolution times, as there will be fewer experts on hand to quickly understand and resolve issues.”
—Winston Hearn, Honeycomb.io
She particularly sees a demand for investment in building and testing infrastructure in order to maintain code quality at this scale. Quality control and new strategies for code management, she predicts, will become a big priority in 2025 to help developers troubleshoot code they didn’t create.
This is actually an area where AI might be able to help, Laura Tacho, CTO of DX, told The New Stack.
“I think by the end of 2025, it will just be normal that all code reviews have some element of AI review,” she said, beyond the automatic nudges a lot of teams have in place. “Some teams rely on metrics around code reviews as performance metrics, so it’s important for those teams to keep learning from the data as new tools change workflows.”
Teams Will Be Organized Differently
Team organization will likely change in 2025 as well, as AI becomes a teammate.
“We’re likely to see more AI-native developer workflows emerge where assistants/agents seamlessly integrate into the software development life cycle,” predicted Helen Greul, vice president of engineering at Multiverse.
These specialized AI worker bees, she said, will be able to handle a range of tasks, “from prototyping and feature development to testing, deployment, optimization and hopefully maintenance.”
In a post on the company blog of Hups, which focuses on upskilling talent, “Agile in the Age of AI,” Henrik Kniberg, an agile coach, predicted that teams will soon evolve into human-AI hybrids, each with a pair of programmers plus their AI assistant. Organizations will have to reconsider the coordination of not only AI technology, but of many more, smaller teams as well.
Junior Developers Will Be Most Vulnerable
Junior devs are at risk of being left behind by current AI strategies. Early-career developers are the canaries in the coal mine with all things GenAI.
“The gap between current education models and real-world developer skills keeps growing wider as technology advances, making it harder for early talent to enter the workforce effectively,” Gruel observed.
Most computer science curricula include a Python class or two, but rarely cover cloud migration, brownfield applications, judging the accuracy of AI-generated code, and other true-to-life complexities.
Already facing an unwelcoming job market, entry-level developers feel at risk of being replaced by generative AI. That’s ironic, because AI-generated code needs at least as many extra reviews as a beginner dev does.
“I have never heard a developer say they are spending too much time coding, so I don’t see how applying AI to the coding aspect of a developer’s role can improve developer experience.”
—Andrew Boyagi, Atlassian
Ironically, less experienced developers are also the most likely to embrace AI tools, according to Stack Overflow’s 2024 developer survey — and also the least likely to be skeptical of them.
Some observers are worried that AI coding assistants — and AI systems that track job applications — may negatively change the tech talent pool, especially for juniors.
“With the proliferation of these tools, take-home assessments might allow less experienced candidates to enter the job market,” said Wesley Faulkner, developer relations professional who’s worked at Amazon Web Services, IBM, MongoDB and SingleStore. “While the actual impact on the industry will likely be minimal, media narratives could exaggerate this story, casting doubt on the value of developer contributions.
“This heightened scrutiny could lead to mistrust or misunderstandings about the work developers do, potentially complicating relationships with stakeholders and affecting morale.”
In 2025, Faulkner hopes that the tech industry will let go of the “code monkey” stereotype and finally emphasize the irreplaceable human elements of software development — creativity, problem-solving and strategic thinking.
Tacho, for one, wonders if and how hiring plans change in 2025: “Some people have said that GenAI will absolutely eliminate junior developers — which I don’t believe, but I bet some companies will start experimenting with new hiring strategies because of claims like that.”
This may seem ridiculous — because you can’t grow more senior developers without them starting out as junior. But, as AI inevitably gets better at certain tasks, it could be used as an excuse to simply not fill vacancies. GenAI assistants could also be used as an excuse to pay entry-level devs less because their job is “easier.”
Everyone Will Need to Upskill
Organizations should remember that generative AI tools require training at all levels.
“The adoption of GenAI has the potential to stunt the growth of both junior and senior developers,” warned Gee, but senior developers are more able to spot the flaws in AI-generated code and content. “They will rely on it as a productivity tool, spending their time training the AI instead of training junior devs.”
Mentoring and collaboration with junior engineers must continue to be a clear part of the senior developer job description.
“While GenAI can provide the starting point and boilerplate code to get a junior running in the right direction, they will ultimately be left on the sidelines without the guidance from senior devs,” she said. “Not only does this create a future talent gap, it prevents the senior devs from continuing to evolve and grow as leaders.”
Knowledge sharing is essential for both teacher and learner and how the industry has always grown its talent pool.
“I think it would be cool to not throw all our eggs in the AI basket,” Tajah Hamilton, an apprentice software developer at MONY Group, told The New Stack. “The impact it’s having on the environment is extreme.”
Better, Hamilton said, to learn more from senior developers: “More emphasis on peer-to-peer learning and community building would be really welcomed by me.”
Burnout Will Still Threaten Developers
Despite this rise of platform engineering in 2024 as a way to unburden internal developer customers, burnout remains an unfortunate industry standard.
“In 2025, I hope organizations prioritize sustainable work environments for developers,” Gruel said. “As the pressure for going faster and being ‘always on’ builds up, we should remember it’s a marathon not a sprint in most cases.”
The risk of burnout — and collapsed productivity — will become worse if developers have all this new technology thrust upon them by their managers instead of being invited to collaborate and experiment with generative AI.
“It’s going to be important to slow down in order to speed up,” Tacho said. “I’m afraid that the pressure to ‘produce, produce, produce’ will lead to organizations adopting tools and approaches that their developers and their cultures aren’t ready for.”
Pressure to Automate Everything Will Increase
Building for the wrong goals — like adopting AI because everybody else is — will also lead to the wrong developer productivity metrics.
“If there is pressure to produce more lines of code or commit code more often, developers will be optimized for volume of work rather than quality,” Tacho said. “Teams will do better if they take a step back and consider what’s really slowing them down,” which may or may not be solved by AI.
“Being able to observe and measure their toolchain will help them to see genuine friction and blockages, and allow organizations to prioritize fixing real productivity issues rather than simply encouraging developers to use tools to crank out more code.”
Just don’t go looking for problems to shove AI at.
“If there is pressure to produce more lines of code or commit code more often, developers will be optimized for volume of work rather than quality. Teams will do better if they take a step back and consider what’s really slowing them down.”
—Laura Tacho, DX
“The entire industry is focused on automation, which usually you apply to solved problems,” Hearn said. “Since DevEx is not a solved problem — or at least solved at a widespread, easily replicable level — the money and power in the system are all focused on automating experiences that are insufficient.”
“AI will become a fixture, hindering future improvements to DevEx because what we automated and invested in will be determined ‘good enough’ due to the sunk cost fallacy. This is a regular pattern in the industry.”
The Developers’ AI Wish List for 2025
But just because developers are skeptical of generative AI doesn’t mean they don’t want it to solve their most mundane and repetitive problems.
“AI can have a huge impact on developer experience if it’s applied as a solution to problems a developer is facing,” Boyagi said. “I have never heard a developer say they are spending too much time coding, so I don’t see how applying AI to the coding aspect of a developer’s role can improve developer experience.
“If developers spend 30% of their time coding, AI could be a good solution to solve the friction points in the remaining 70%.”
Among the areas where the experts we spoke to saw a role for AI in 2025 and beyond:
Documentation and Code Analysis
“Every developer survey has this as a data point,” as a top, if not the top, frustration, noted Lorna Mitchell, a developer experience specialist who’s worked most recently at Redocly and Aiven.
And docs generation is probably the biggest proven AI gain so far, as this year’s DORA report found that a 25% increase in AI adoption translates to a 7.5% improvement in documentation quality.
AI is already great at summarizing even complex topics. That’s why Camille Fournier, author of the “Platform Engineering Guide for Leaders,” would like it applied to better summary tooling for incident management.
This knack for explainability can also be used to understand code quality, Gee said. “I would like AI to be able to identify why particular failures happen, and/or tell me what’s wrong with the code.”
Technical Debt Clean-Up
In 2025, AI can also be applied to cleaning up technical debt, including operational debt, like slow and painful releases, noisy alerts and monitoring, and stale dashboards, suggested Tilde Thurium, senior developer educator at LaunchDarkly.
“Generative AI will be able to use more context from your existing environment to provide more relevant coding suggestions,” they said. “For example, Cursor adds AI into your editor so it can use your source code and docs to formulate a better answer, without having to write the perfect prompt.”
Code Testing
Better testing is another perennial developer request that, like docs and debt, most devs don’t actually want to be involved in fixing. As Tacho said, AI presents both more of a demand for testing and potentially its own solution.
“Testing will become a strategic priority for IT leaders” in 2025, she said. “Engineering leaders are finally starting to recognize that testing is fundamental to delivering quality software and that slow or flawed tests, like flaky tests, can hinder an organization’s ability to reach set milestones.”
“Additionally, as AI-driven development becomes more widely adopted, automated testing will be imperative and can provide companies with a competitive advantage.”
Easier Provisioning of Cloud Infrastructure
If Einy had just one wish for AI, it would be to automate a golden path for the provisioning of cloud infrastructure.
“Today, developers are required to know Terraform and adhere to organizational standards around security and FinOps,” he said. “Being able to provision and manage cloud infrastructure will empower developers to move quickly and innovate without breaking things.”
AI can also enable release automation, Thurium said, because AI can aggregate signals, to more quickly understand when releases are going south in order to automatically roll back.
“I hope generative AI will be applied in ways that make it easier and safer to change existing systems,” echoed Nathen Harvey, DORA lead at Google Cloud.
“We’ll see this starting to happen as AI enables more time in flow state, provides fast, quality feedback on changes, and helps teams learn more about their applications and their users.”
The post Developer Productivity in 2025: More AI, but Mixed Results appeared first on The New Stack.