The new gold rush: vibe coding, AI engineers, and the testing crisis nobody is ready for

One of the most frequent questions I get asked by customers is: what does the engineer of the future look like?

This pace of change across the software industry is unlike anything the world has seen before. What began as AI-assisted coding, through GitHub Copilot for example, has rapidly turned into something much broader: autonomous engineering agents capable of writing applications, debugging systems, generating tests, refactoring repositories and orchestrating workflows with minimal human interaction.

For startups, founders, and engineering teams this is the new gold rush. It’s never been easier to launch a new product or business. Platforms like Cursor, Windsurf, Claude Code, Lovable and IBM’s Bob are all dramatically lowering the barrier to entry. What once required large development teams, significant investment and months of delivery can now be prototyped in days or even hours. Not only that, but these platforms have fundamentally changed how software creation feels. Development is becoming increasingly conversational, iterative and autonomous.

Then came Devin…

Cognition launched and positioned Devin as an “AI software engineer”, the first of its kind, capable of planning, building, testing and shipping software autonomously. Shortly afterwards, it acquired Windsurf, signalling a major shift in the direction of the market.

This demonstrated the way in which the industry is rapidly converging toward a future where AI-native IDEs (integrated development environments), autonomous engineering agents, testing automation, orchestration layers, governance controls and deployment workflows, all became part of a single intelligent software delivery ecosystem.

Organisations are still validating software like it’s 2018

The barriers to software creation are collapsing in real time. A single founder can now prototype products overnight, automate workflows without large engineering teams and iterate faster than heavily funded startups could only a few years ago. For the first time, entire categories of entrepreneurship are becoming accessible to smaller teams and individuals.

The average developer now submits 76% more code per month than two years ago. This sounds exciting, but it also introduces a serious problem. Most organisations still validate software like it is 2018, whilst AI is generating software like it is 2030. The result is a widening operational gap.

As AI accelerates development speed, it also increases:

• Release frequency
• Code volume
• Infrastructure complexity
• Dependency sprawl
• Governance pressure
• Operational risk

Testing teams have also not scaled at the same pace. According to UiPath, many organisations still operate with:

• Four-to-six-week release cycles
• Testing estates that remain heavily manual
• Testing overhead consuming roughly 30% of IT budgets
• Fragile scripts that regularly break after application changes
• Multiple disconnected testing tools spread across teams

Banks, insurers, trading firms and payment providers cannot tolerate instability in the same way other industries sometimes can. Failed deployments, broken onboarding journeys, payment disruption or unstable customer experiences quickly become operational, regulatory and reputational issues.

The commercial impact is significant too. UiPath references research showing that every additional 100 milliseconds of latency can reduce revenue by roughly 1%, whilst large organisations increasingly report downtime costs exceeding $300,000 per hour. Suddenly, the conversation has shifted from quality assurance to one focused on operational resilience.

New operational risks

The industry is also learning another important lesson very quickly: AI-generated code still breaks. There are already examples emerging across the market of AI coding agents:

• Deleting production resources
• Introducing security vulnerabilities
• Generating unstable infrastructure configurations
• Hallucinating dependencies
• Modifying systems unexpectedly
• Damaging data environments during autonomous execution

Anthropic positions Claude Code as an “agentic coding tool” capable of editing files, executing commands and operating autonomously across repositories. The capabilities are extraordinary, but they also introduce entirely new operational risks when deployed without sufficient governance and validation.

A human review alone no longer scales linearly with AI-generated change, and it’s quickly becoming one of the defining technology challenges of this decade. The real challenge ahead is no longer how to generate software faster, AI has already solved much of that problem, it’s whether organisations can validate, govern and safely release software generated at AI speed.

Testing is more critical than ever

Historically, testing was often viewed as an operational overhead or a late-stage delivery checkpoint. Now, it’s becoming one of the most important control layers in enterprise technology estates, and this is why the testing market itself is evolving rapidly. Traditional enterprise testing platforms like Tricentis and OpenText continue to dominate large enterprise QA estates, particularly across highly regulated environments and complex SAP estates.

Meanwhile open-source frameworks like Selenium remain widely used across developer-led engineering teams. However, many enterprises are discovering that whilst open-source tooling may appear inexpensive initially, operational complexity often grows significantly over time through:

• Framework sprawl
• Maintenance overhead
• Brittle scripts
• Developer dependency
• Governance gaps

As UiPath bluntly states: “Selenium is free to download. It is not free to run.”

AI-assisted testing models

At the same time, platforms like Leapwork and UiPath are helping push the market toward AI-assisted and autonomous testing models.

UiPath’s positioning is particularly interesting because it reflects the broader direction the market appears to be heading. Rather than treating testing as a standalone discipline, platforms are increasingly converging AI agents, orchestration, automation, governance, testing, synthetic data generation and enterprise workflows, into unified operational platforms.

UiPath refers to this as “Agentic Testing”. The concept is relatively simple: AI agents increasingly generate tests, execute validation, self-heal broken scripts, automate regression coverage and support software delivery continuously rather than periodically.

It’s the answer for firms that are trying to scale AI ambition on estates that were never designed for continuous autonomous change. Governance, operational resilience and software assurance can no longer sit separately from development itself.

Where is the market heading?

The organisations making the most progress are recognising something uncomfortable but important: the challenge is rarely ambition, it is architecture.

When development, testing, governance and automation operate as disconnected disciplines, scaling AI safely becomes extremely difficult. When they are integrated intentionally, resilience and trust become by-products of how the organisation operates rather than retrospective controls layered on afterwards.

The firms moving fastest today are not necessarily those experimenting with the most AI tooling. They are the organisations building the operational capability to validate and govern software continuously and safely as engineering velocity accelerates.

This is where the market is ultimately heading. The software industry spent the last decade trying to accelerate development. AI solved that problem faster than almost anyone expected. Now enterprises face a new challenge entirely: making sure what gets generated can still be trusted before it reaches production. As organisations accelerate AI adoption across software engineering and operational delivery, many are discovering that testing, governance and release confidence are becoming critical operational constraints.

At Softcat, our Data, Automation & AI practice works with organisations to help assess AI readiness, modernise testing strategies and evaluate emerging AI-assisted testing and automation platforms responsibly. Click here to find out more.