INDUSTRY TRENDS & EMERGING TECHNOLOGIES
With Artificial Intelligence assistants such as GitHub’s Copilot becoming more powerful every day, so is their presence in our work. How does this impact companies and workers?
1. Defining AI Copilots: A New Era of Digital Assistance
2. The Mechanisms Behind AI Copilots
3. Real-World Case Studies: Transformative Benefits for Organizations
4. Where AI Copilots Still Fall Short
5. Join the Conversation at MTLC 2025
It’s been almost two years since the Generative AI chatbots took the world by storm and started the radical shift we are experiencing now. And yet, here we are, still speculating about its effect, consequences and predictions.
No, Artificial Intelligence tools are no longer a novelty; in fact, they are becoming more and more a part of our everyday lives and jobs. But we are still adapting and discovering novelties around its use. One of them is: What does their mass adoption mean for companies and the developers they employ?
A recent survey by GitHub found that 97% of software engineers reported using AI coding tools at some point in their work. Additionally, 75% of enterprise software engineers are predicted to use AI coding assistants by 2028, according to Gartner, up from less than 10% in early 2023. As these tools evolve from simple chatbots to full-featured copilots, their impact is only expected to grow. As these tools evolve from simple chatbots to full-featured copilots, their impact is only expected to grow.
To better understand the real-world implications of these changes, we turned to the experts at Jalasoft, including Miklos Cari, a member of Microsoft’s Most Valuable Professionals program, which connects technical community leaders with Microsoft to foster innovation and knowledge sharing, and Davor Pavisic, VP of Engineering & CTO. We asked them for their insights on the rise of AI copilots in the workplace.
AI copilots represent the next generation of workplace automation. You can view them as an upgrade to the generative AI Chatbot we’ve grown accustomed to. Instead of just answering questions, they’re embedded into workflows, proactively making suggestions, writing code, organizing information, and more. They come in various forms:
These are designed for focused tasks in specific domains. For example, tools like Harvey AI assist law firms in drafting and analyzing legal documents, while ChatGPT’s Code Interpreter (also known as Advanced Data Analysis) helps users explore datasets and generate visualizations using natural language. In marketing, apps like Jasper create brand-aligned content tailored to different channels with minimal input.
They support individual tasks without full integration into enterprise systems. GrammarlyGO, for instance, enhances writing through suggestions and drafting, while Notion AI offers summarization and brainstorming features inside collaborative notes. Another popular example is Otter.ai, which transcribes meetings and generates key takeaways automatically.
These focus on digital organization and operational efficiency; they help manage the growing complexity of digital workspaces by sorting emails, classifying files, and organizing shared folders automatically. Microsoft’s Copilot OneDrive is a prime example—it assists users in locating, categorizing, and summarizing content stored in the cloud, reducing manual effort and improving file management at scale.
These go a step further, as they are deeply embedded into productivity tools or coding environments. One example is Salesforce Einstein Copilot, which empowers sales teams by offering AI-driven insights and next-step recommendations directly within their CRM workflows.
Specifically, coding environments, GitHub Copilot is a leading example, and it is integrated directly into IDEs like Visual Studio Code. It suggests code completions and even entire functions in real time. These are the ones we are focusing on in this article: According to GitHub’s own reports, developers using Copilot often complete tasks up to 55% faster than without it. “It’s incredibly helpful for generating comments, summarizing code, and maintaining consistency across large codebases. These are areas where Copilot really shines”, explained Miklos Cari.
Importantly, as Cari pointed out, a significant shift occurred in December 2024 when GitHub announced that Copilot would be available for free to all Visual Studio Code users. “This move significantly lowered the barrier to entry and has had a noticeable impact on developer productivity,” he explained. It marked a moment when access to AI support tools became frictionless for a global community of developers.
That evolution continued at Microsoft Build 2025 with the introduction of GitHub Copilot’s agent mode, which Cari described as “a game-changer.” In his words, “Agent mode positions the AI as a true peer programmer—going beyond simple code suggestions to enable more intelligent collaboration, context-aware assistance, and even task automation.”
We know that most workers have embraced generative AI in their workflows, and developers are no exception. But when organizations start adopting these tools at a scale, the conversation shifts. Who benefits the most from these copilots? An MIT study might give us the answer. Let’s find out!
These systems are powered by large language models (LLMs), trained in vast repositories of publicly available code or documents. As users type, the AI predicts what comes next—whether that’s a line of code, a paragraph of text, or an email subject line. Over time, these tools learn from user behavior and improve their contextual accuracy.
Importantly, they don't just autocomplete; they actively accelerate the decision-making process by offering relevant options and anticipating user needs.
(Read our blog post “What is Generative AI” to learn more about the inner workings of Generative AI)
The speculation around Generative AI and efficiency has been around since its widespread adoption some years ago. It’s fair to say that the world is still adjusting, and LLMs are still evolving; therefore, it might be too soon to make any strong affirmations. However, many specialists are already showing some preliminary data. For instance, last year’s IBM data showed that only 42% of the surveyed companies had actively deployed AI in their business. Even today, one year later, experts estimate that the adoption rates have only increased by 60%.
This year, a large-scale field study by MIT researchers offered one of the most comprehensive looks at GitHub Copilot’s real-world impact, focusing on companies like Microsoft and Accenture. A first look at their results is beyond promising:
Their data showed that developers using Copilot saw a 26% increase in weekly tasks completed, with even stronger gains among junior or newer developers. As the researchers note, “Our analysis reveals a 26.08% increase (SE: 10.3%) in completed tasks among developers using the AI tool.”
The number of code builds increased by 38%, and code commits rose by 13%. They add, “We find evidence of productivity-enhancing effects of GitHub Copilot: on average, the number of weekly pull requests made by developers increases by 26.08%, the number of weekly commits increases by 13.55%, and the number of weekly builds increases by 38.38%.”
While these gains are promising, especially for onboarding and supporting junior developers, they also raise important questions: How do teams ensure code quality in AI-assisted environments? Well,researchers were unable to evaluate the quality of the work produced with Copilot.
The study also focused on another most-asked question: Who benefits the most?
Researchers found that these benefits are particularly strong for less experienced professionals: As they put it, “Copilot significantly raises task completion for more recent hires and those in more junior positions, but not for developers with longer tenure and in more senior positions.” In other words, junior developers not only use the tool more often, they also see the biggest improvements in their output.
This finding aligns with what many in the industry are already observing. As Jalasoft’s expert Miklos Cari explains: “GitHub Copilot acts like a constant tutor, offering best practices and frameworks at your fingertips, which can significantly accelerate learning, especially for junior developers.”
However, interpreting the data requires some nuance. As Mert Demirer, assistant professor of economics at MIT Sloan, noted: “Inexperienced and short-tenured software developers were more likely to use the tool, and, moreover, their productivity increased a lot more. For those who are more experienced, we actually don’t see much of an effect.”
That doesn’t mean senior engineers are left behind. On the contrary, Cari points out that they benefit in different, more strategic ways: “They can leverage Copilot not just for coding, but for making complex architectural decisions, automating repetitive tasks, and streamlining workflows.” This advanced use of the tool can unlock significant efficiency, but it also poses a risk: if not carefully managed, it may widen the gap between junior and senior developers rather than close it.
But why is it that Senior engineers are more reluctant to use the tool? The reasons range from technical and security concerns to legal, ethical, and even philosophical ones
One of the most pressing concerns is code safety. AI-generated code can introduce vulnerabilities if not carefully reviewed. Studies have shown that tools like Copilot may produce code with security weaknesses, such as insufficiently random values or improper control of code generation. These issues concern senior developers responsible for ensuring the security and robustness of applications.
Another big issue for experienced developers has to do with ethical and legal considerations. Concerns about code licensing, potential plagiarism, and the implications of using AI-generated code without clear attribution are significant.
As one developer put it, "AI doesn’t innovate; it mimics." Copilot, like other LLM-based tools, functions by identifying patterns in existing code and reproducing them. That raises the uncomfortable possibility that proprietary or sensitive logic could be inadvertently reused.
While GitHub has stated that Copilot does not train user code in real time, there's still ambiguity about how that data might be used in the future. This uncertainty becomes even more critical at the enterprise level, where protecting intellectual property and maintaining regulatory compliance is non-negotiable.
In fact, GitHub Copilot has faced criticism (and even a class action lawsuit filed in 2022) for allegedly reproducing code from public repositories without proper attribution. If Copilot-generated code closely mirrors GPL-licensed content, it could expose companies to open-source compliance risks they didn’t bargain for.
As Davor Pavisic, VP of Engineering & CTO at Jalasoft, explains:
"At Jalasoft, we are deeply committed to protecting our clients' information security. As part of this commitment, our developers do not use AI tools in their daily work unless explicitly authorized by the client and in accordance with the conditions they specify."
So, contrary to what many believe, these tools aren’t magic…and they’re certainly not foolproof.
Miklos Cari explains: “Like any generative AI, copilots can hallucinate, generate buggy or suboptimal code, or overlook important context. Having the experience to spot and correct these issues is essential. This makes critical thinking and a solid grasp of development fundamentals more important than ever.”
“There are also tasks that should never be fully outsourced to AI”, he continues. Take architectural decisions, for example—these involve deep system knowledge, long-term trade-offs, and an understanding of evolving infrastructure needs. Similarly, security-sensitive code often requires an awareness of company-specific policies and compliance standards that AI cannot predict or enforce.
In other words, copilots are powerful for companies, which means potentially faster delivery cycles, improved efficiency among less experienced staff, and better utilization of senior developers’ time. But their success depends on thoughtful, informed, human developers. Rather than replacing developers, AI copilots appear to amplify human capability, especially when paired with the right training and adoption strategies.
Conclusion
AI copilots like GitHub Copilot are becoming a foundational part of how modern software is built. The data shows real, measurable benefits, especially for junior developers who gain speed, confidence, and guidance from having AI integrated directly into their workflow. But as with any transformative tool, there’s more beneath the surface.
With the data available to us right now, we can only say that the productivity and efficiency hype seems to be real. However, as workers and AI Models continue to change, we are still uncertain about how these changes might look 10 to 15 years from now.
Curious to learn more about how AI is reshaping software development in real-world teams?
We’re excited to sponsor “When AI Becomes a Mandate,” an event hosted by the Mass Technology Leadership Council on June 20th!
As AI becomes a core business expectation, tech leaders are under increasing pressure to deliver measurable outcomes. While early wins are encouraging, scaling AI across teams and functions remains challenging.
That’s why this event is so important!
Join us for an honest, practical conversation with operational leaders who are making AI work in finance, HR, supply chain, and customer success.
Recibe latest updates and promotion for your organization
