The advent of agentic artificial intelligence has fundamentally transformed the economics and velocity of software development. In an era increasingly defined by rapid iteration and colloquial terms like vibe coding, the barrier to entry for creating functional software applications has plummeted. Today, non-technical personnel—ranging from project managers to domain specialists—are working alongside seasoned software engineers to ideate, prototype, and deploy digital features at unprecedented speeds. Yet, this democratization of creation has introduced a complex set of operational dilemmas. When virtually any conceptual idea can be rapidly translated into a working feature, deciding what actually deserves to reach the end consumer has become a monumental challenge fraught with strategic, financial, and security risks.
These tensions took center stage during a high-profile panel session at the recent Harness Unscripted event in London. Convening senior technology and engineering leaders from major multinational corporations, the discussion explored how modern enterprises can successfully harness the culture of experimentation enabled by AI while maintaining rigorous quality control, fiscal discipline, and robust pipeline management.
The Evolution of Development: From Scarcity to Velocity
To understand the current state of software engineering, one must examine the rapid technological trajectory of recent years. For decades, software development was bottlenecked by human capital. Writing, testing, and debugging code required specialized linguistic and architectural skills, making engineering hours a tightly managed and expensive resource. The introduction of large language models initially assisted developers with syntax completion and snippet generation. However, the maturation into agentic AI—autonomous systems capable of executing complex workflows, planning multi-step tasks, and self-correcting—has shifted the paradigm entirely.
Organizations are now witnessing a tenfold increase in raw coding output. Yet, as engineering leaders at the Harness Unscripted event pointed out, generating volume is no longer the primary hurdle. During the panel, James Ford, Principal Engineer at price comparison specialist Compare the Market, highlighted that while AI dramatically accelerates the mechanics of coding, the fundamental limits of human oversight remain unchanged.
"The scarce resource is human attention and proper judgment over things," Ford stated during the London session. "We can build stuff faster, but the scarcity is getting human eyes on this 10x that we’re doing."
This sentiment underscores a growing anxiety within the enterprise architecture community: as code becomes cheap to produce, the discipline traditionally associated with software craftsmanship risks erosion. If teams become overly enamored with sheer output metrics, they may neglect the foundational key performance indicators (KPIs) that dictate whether a software update genuinely drives long-term business value rather than mere computational noise.
Managing the Pipeline and Prioritizing Value
With development velocity soaring, the orchestration of the software release pipeline has emerged as a critical battleground. Matt Murphy, Principal Engineer at financial services titan Barclays, emphasized that organizations must implement stringent controls over the release sequence to prevent operational chaos.
According to Murphy, the sheer volume of AI-generated code means that businesses are increasingly overwhelmed by potential features. Without a strict mechanism to govern what enters the production pipeline, companies risk squandering the very efficiency gains they sought to achieve.
"The business has got to decide which customers to prioritize, and which feature provides the biggest bang for the buck," Murphy explained to the Harness Unscripted audience. "You can deliver more software using AI, but you still must get that pipeline to say what we are going to deliver in which order."
Agreeing on these development priorities is rarely straightforward. Aditya Bhardwaj, Head of SaaS Development at enterprise software firm Temenos, noted that organizations evaluating AI-generated features must look far beyond immediate deployment costs. Every application pushed into a production environment carries a long-term operational tail, requiring ongoing maintenance, security patching, and customer support. Consequently, leadership teams are being forced to re-evaluate traditional financial metrics to ensure that accelerated delivery translates directly into measurable business outcomes.
The Double-Edged Sword of Rapid Experimentation
One of the most touted advantages of agentic AI is its capacity to facilitate rapid prototyping and testing. In a traditional development cycle, conceptualizing an idea, building a minimum viable product, and deploying it to a subset of users could take weeks or months. AI compresses this timeline into days or even hours, allowing organizations to embrace a "fail fast" methodology.
Barclays’ Murphy championed this capability, noting that rapid prototyping enables businesses to engage directly with their audience much earlier in the product lifecycle. By gathering immediate customer feedback, development teams can execute multiple iterations in rapid succession, aligning the final product more closely with user demands.
However, this freedom to experiment introduces distinct logistical hurdles. Temenos’ Bhardwaj warned that the ease of running tests can easily lead to over-experimentation. When development teams initiate too many parallel tests, the associated financial and computational costs escalate rapidly. More critically, overlapping experiments muddy the analytical waters, making it exceedingly difficult for data scientists and product managers to isolate which specific variables drove customer engagement or satisfaction.
This measurement dilemma was echoed by Ford, who detailed the complex mathematics involved in managing multiple concurrent deployments. "You’ve got the complicated math of, ‘Well, how do I push 20 things live at once, and how do I divvy up my traffic between all of them, so we don’t get overlapping groups and the wrong signals out of the measurements as to whether this experiment is working or not?’" Ford said. "Measuring the impact of what you’ve done is the challenge to solve."
Compounding this challenge is the difficulty of associating costs with specific experiments. Bhardwaj pointed out that because AI-enabled projects often yield multifaceted business outcomes—such as a simultaneous lift in Net Promoter Scores (NPS) and operational efficiency gains—disentangling financial inputs from discrete outputs remains an unsolved puzzle for many enterprise accounting and development teams.
Expanding the Circle: Non-IT Professionals and Vibe Coding
Perhaps the most culturally disruptive aspect of the agentic AI era is the inclusion of non-IT professionals in the software creation process. Often referred to in tech circles as vibe coding, this phenomenon allows individuals without formal computer science backgrounds to direct AI agents to build functional applications using natural language prompts.
While this trend has raised eyebrows among purists, progressive engineering leaders view it as a powerful catalyst for innovation. Murphy shared an illuminating anecdote from a recent Barclays company offsite, noting that the most effective rapid prototype during an internal exercise was built not by a career software engineer, but by a project manager. Unencumbered by traditional engineering dogmas about how software "ought" to be written, the project manager leveraged AI agents to deliver a creative and functional solution.
Nevertheless, industry experts stress that a stark divide exists between building a prototype and deploying production-ready software. Bhardwaj cautioned that non-technical creators must remain realistic about enterprise timelines. Even with advanced AI agents assisting in the initial coding phase, getting a tool into production still takes weeks or months because the entirety of the software development lifecycle (SDLC) is not yet fully automated.
Ford reinforced this perspective, drawing a sharp distinction between casual prototyping and professional AI engineering. A prototype, he argued, is merely the starting point of a rigorous journey.
"You’ve got to have all the security checks, all the other elements of quality, all the metrics, and all the robustness and the engineering underneath something before you can put it live to anyone," Ford emphasized. "It needs to go through the system to make it ready to put out with some confidence."
Establishing Guardrails in Regulated Sectors
The debate over speed versus safety inevitably leads to the question of governance, particularly in heavily regulated industries such as banking, insurance, and healthcare. While tech startups and consumer-facing application developers may embrace a philosophy of continuous deployment and high-frequency automated releases, heavily regulated institutions must exercise profound caution.
Murphy noted that while the utopian vision of fully automated, continuous deployment is alluring, it is likely incompatible with the risk profiles of major financial institutions. In environments where compliance failures carry severe legal and financial penalties, human oversight cannot be bypassed.
"Working in a heavily regulated environment, I think we might always have to have a human in the loop," Murphy asserted. "The nirvana state might be, ‘Well, let’s keep continuous deployment,’ but I don’t think we’ll ever get there in a bank."
To mitigate risk, organizations are establishing clear boundaries that dictate the boundaries of automated systems. Critical business logic, irreversible financial transactions, and sensitive data access remain strictly tethered to human review. Automated agents are increasingly utilized to handle the heavy lifting of drafting code, running automated unit tests, and scanning for known vulnerabilities, but the final authorization to ship remains a human prerogative.
Industry Implications and Strategic Outlook
The insights shared at the Harness Unscripted event in London point to a broader maturation phase in the enterprise adoption of artificial intelligence. The initial gold rush—characterized by unbridled enthusiasm and a rush to deploy AI tools wherever possible—is giving way to a more sober, disciplined approach to software governance.
As organizations digest the reality of 10x coding speeds, executive leadership must redefine their operational frameworks. Success in the age of agentic AI will not belong to the companies that write the most lines of code or push the highest volume of features into production. Rather, it will belong to enterprises that successfully balance the creative energy of rapid experimentation with the unglamorous, essential disciplines of pipeline management, cost attribution, and rigorous quality assurance.
Ultimately, the consensus among industry leaders is clear: while artificial intelligence has democratized the ability to build, the ultimate responsibility for what reaches the customer remains firmly human. Ensuring that these technological advancements serve strategic business goals without compromising security or quality will define the next era of enterprise software engineering.
