The Future of Software Development: AI Speed, Human Judgment

By Ken Mugrage

It was hard to ignore the fanfare from the recent International Collegiate Programming Contest: machines trump human coders. And while the truth might be the first victim of an eye-catching headline, it’s hard not to admire this display of technological prowess. That said, few enterprise leaders should be planning to dispense with their human workers yet.

The competition — often dubbed the coding Olympics — saw Google’s Gemini 2.5 Deep Think and OpenAI’s GPT-5, along with an experimental model, secured gold medals, a distinction earned by only four of the 139 human teams. OpenAI’s models achieved a perfect score, solving all 12 problems, while Gemini solved 10 of the 12, a performance that would have placed it second among human teams. So, machines 1: humans 0, right?

Maybe not. Firstly, you shouldn’t equate competitive programming success with real-world software engineering. The results, while a testament to the models’ abstract reasoning and problem-solving abilities, don’t negate the critical need for human oversight and engineering rigor. Secondly, the evidence to support a wholesale shift to AI coders just doesn’t stack up. Check out the 2025 State of AI-assisted Software Development report from DORA for more details.

This well-timed deep dive into AI-assisted software development provides a clear, data-backed perspective on why humans are still needed in the loop, especially when AI is used to create complex software systems.

As the DORA report notes, AI’s primary role in software development is that of an amplifier. It magnifies the strengths of high-performing organizations and exposes the weaknesses of struggling ones. For a complex, real-world project, an organization with a solid technical foundation, clear workflows and a healthy data ecosystem can supercharge results through AI adoption. On the flip side, a disorganized, siloed organization will find that AI simply creates “localized pockets of productivity that are often lost to downstream chaos”. If you hand the keys to AI, don’t be surprised when your project ends in the ditch.

Why is this? Is this simply a case of a cabal of techies trying to guard their closed shop? Not quite.

The DORA report mirrors much of what we’ve seen from our teams: real-world software development is messy. It’s filled with ambiguities, unstated requirements, and constantly shifting priorities that require human judgment and contextual understanding.

Meanwhile, the ICPC contest, with its well-defined problems, strict rules and clear success criteria, is an ideal environment for AI to thrive.

The contest results show AI’s incredible speed, but as the DORA report notes, increased throughput doesn’t automatically mean better outcomes. It’s worth digging into DORA’s data to see that while AI adoption improved software delivery throughput, that has been accompanied by an increase in software delivery instability.

In other words, teams are adapting for speed but have yet to adapt to AI-accelerated software development in a way that manages that speed safely. And here’s the rub: AI can generate code at lightning speed, but it lacks the human intuition to anticipate how that code might break an existing system, introduce a security vulnerability, or create technical debt that will haunt the team for years. Humans are needed to provide the checks and balances that prevent this speed from turning into chaos.

Unlock AI’s potential through human-led decisions

So far, so disheartening. It’s all well and good knowing that AI amplifies the efforts of high-performing teams, but what if your team isn’t at that level? How can you leverage AI successfully?

Again, the DORA report highlights some interesting ideas. It identified seven foundational practices that have proven to amplify AI’s positive impact on enterprise performance. These are:

Clear and communicated AI stance. Leaders must set clear AI policies to balance innovation with security.
Healthy data ecosystems. High-quality, unified data is essential for impactful AI.
AI-accessible internal data. Securely connecting AI to internal sources boosts effectiveness.
Strong version control practices. Robust version control mitigates risks from AI-driven code changes.
Working in small batches. Small batches improve performance and reduce AI friction.
User-centric focus. Human empathy ensures AI solutions align with real user needs.
Quality internal platforms. Strong platforms provide the guardrails to scale AI safely.

Put your trust in skeptical developers

The DORA report also highlights a fascinating finding: a notable portion of developers (30%) report having little to no trust in AI-generated code, yet a majority (95%) rely on it. This “trust but verify” approach is a sign of mature adoption. As the report states, developers compare this to the healthy skepticism they apply to other widely used resources, like solutions found on Stack Overflow. This is a crucial distinction. In the real world, human developers are not passively accepting AI’s output; they’re critically evaluating, guiding and validating the work to ensure its quality and correctness. This is especially true for the more complex and nuanced tasks that make up real-world software development.

To leverage the AI’s full potential, then, you want to harness its ability to help you move at lightening pace. The ICPC results are a powerful demonstration of what AI can do. They represent a significant leap forward in AI’s ability to reason and problem-solve.

But you also need the human in the loop. The real opportunity is to build on this base, treating AI not as a replacement, but as an integral part of a powerful, human-led workflow. The DORA report validates the approach we’ve taken over the past 18 months and outlines the roadmap for a successful AI transformation. The future of software development lies in combining the incredible speed and power of AI with the irreplaceable judgment, empathy and experience of human engineers.

Originally published at https://www.thoughtworks.com.