The logic behind software purchasing has shifted. With AI now autonomously writing, reviewing, and refactoring code, the gap between adapting a system to a company's processes and forcing operations to fit a generic product has narrowed—and become more strategic.
The data supporting this shift is clear. A Cornell University study showed that AI agents intentionally alter 26.1% of commits to improve software structure, primarily in maintainability (52.5%) and readability (28.1%). Concurrently, Jellyfish recorded a jump from 51% to 82% in enterprise adoption of AI agents throughout 2025. For IT and operations leaders, the implication is direct: if software engineering has become faster, more iterative, and automated, the competitive advantage shifts to architecture, data, and the solution's adherence to real-world processes.
AI Has Reduced the Cost of Customizing Software
The most significant change isn't just that AI generates code. It's that AI has become an active participant in continuous software improvement.
According to a study reported by Information Management Portal, based on Cornell University research, AI already performs intentional refactoring—it doesn't just deliver functional code, but alters the codebase to make it more sustainable over time. This changes the economic calculus of customization.
AI-powered tools now automate tasks that previously consumed hours of technical teams, such as:
- suggesting functions;
- completing boilerplate code;
- translating between programming languages;
- identifying logical errors;
- accelerating test generation;
- supporting documentation;
- implementing routine functionalities.
This operational gain helps explain the speed of agentic engineering adoption. According to Jellyfish, 51% of companies used AI with agents at the beginning of 2025; a few months later, this percentage reached 82%. In the same period, there was a 1.16x reduction in code review time and increased use of workflows where AI itself creates, confirms, or opens reviews.
In practice, this reduces some of the historical friction of custom software: repetitive effort. When AI absorbs a growing share of operational work, human teams can focus their energy on process design, integration, business rules, and governance.
Edy Silva, Developer Relations at Codeminer42, summarizes this role change: “Instead of executing operational tasks, developers now act more strategically, coordinating and supervising different AI agents throughout the process. This new dynamic demands an even more solid technical mastery, as the quality of results depends on the ability to guide AI with precision.”
Codeminer42 itself reported a three to fourfold productivity gain with agentic engineering. This data is less important as a universal promise and more as a trend indicator: customization is no longer automatically synonymous with slowness.
AI Exposes the Limitations of Generic Products
If the cost of building has fallen in many cases, the limitations of generic software have become more visible.
Standardized platforms still make sense when the priority is deployment speed, when the use case is generic, or when the company lacks a technical team to build and maintain its own solution. This point is clearly articulated in Sprinklenet's framework on the build vs. buy decision.
The problem arises when AI capabilities cease to be accessory and begin to affect results, margins, operational quality, or differentiation. In these scenarios, the inadequacy of the generic product tends to appear in very concrete ways:
- the company possesses unique data that creates a competitive advantage;
- the case requires specific domain precision;
- there is a need for total control over model behavior, updates, and deployment;
- regulatory requirements demand transparency regarding architecture and training data;
- AI capability is central to the product or service delivered.
Jamie Thompson, from Sprinklenet, synthesizes this point: “Custom AI solutions deliver the greatest value when your competitive advantage relies on proprietary AI capabilities that competitors cannot replicate.”
This shift helps explain an apparent market contradiction. On one hand, enterprise use cases deployed via third-party solutions grew to about 76%, compared to a near 50/50 split two years prior. On the other hand, only 29% of executives are achieving significant ROI with AI initiatives.
The plausible interpretation, based on the briefing data, is that adoption does not equate to value capture. Buying AI-powered software has become easier. Transforming this into productivity, revenue, and operational advantage still depends on process adherence and governance.
The Debate Is No Longer “Build or Buy” but Value Architecture
The decision between custom software and a generic product is not binary. The reference material itself points to a spectrum of possibilities: fine-tuning open-source models, customizing configurable platforms, and building on AI infrastructure services.
This point is crucial to avoid a common mistake: treating custom software as synonymous with building everything from scratch.
In many cases, the most efficient architecture combines different layers:
- third-party AI infrastructure;
- base models already available on the market;
- customization based on internal data and workflows;
- integration with legacy systems;
- proprietary business rules;
- specific operational governance.
This approach matters because the low-code/no-code market surpassed US$37 billion in 2025, according to data cited by AngelHack DevLabs based on Fortune Business Insights. This growth shows the pursuit of speed and autonomy. But it also highlights a tension: as AI enters critical processes, the discussion moves beyond just the build interface to encompass control, integration, sustainment, and total cost of ownership.
In Sprinklenet's framework, custom solutions require a greater initial investment in data engineering, model development, and infrastructure. In return, they can deliver a better total cost of ownership over a three to five-year horizon, especially when the alternative would be operating on expensive premium tiers of SaaS platforms.
Additionally, there's a less obvious financial dimension. When AI becomes a central part of operations, the cost isn't just in the tool's license, but in how much it forces the company to circumvent limitations, duplicate work, export data between systems, or maintain parallel processes.
Therefore, the most useful question for IT and operations leaders isn't “which tool has AI?” but “where does standardization begin to create bottlenecks that a custom architecture would solve more efficiently?”
Speed Without Control Increases Risk — and Reinforces the Need for Custom Design
Receba os próximos artigos por e-mail
Conteúdo novo de Draivv direto na sua caixa de entrada. Sem spam.
The acceleration of development does not eliminate a central problem: functional code is not, by definition, secure code.
The study cited by iMasters became known as “Asleep at the Keyboard” precisely for this reason. AI often delivers software that works, but without the necessary security rigor. Among the highlighted risks are:
- prompt injection;
- insecure output handling;
- training data poisoning;
- disclosure of sensitive information;
- supply chain risks.
This point repositions the discussion about custom software. Customization isn't just about adapting interfaces or workflows. In critical contexts, it allows for the incorporation of controls, validations, and governance compatible with the real risk of the operation.
When a company relies on specific transparency requirements, control over model behavior, or domain precision, a generic solution can limit precisely what matters most: predictability. The Forbes article on AI as infrastructure directly describes this movement by pointing to the need to operate “within critical systems, under specific business rules, with predictability and control.”
At the same time, the answer isn't to remove humans from the process. It's to reposition them. “The future of software engineering is not artificial, it is hybrid and focused on real results,” says Professor Arielton Gomes Nunes, a lecturer in the Systems Analysis and Development course at Centro Universitário Ateneu.
This statement helps separate productivity from blind automation. If AI accelerates generation, review, and execution, the value of the technical team shifts to supervision, architecture, and quality. In custom software, this means using AI to reduce mechanical effort without sacrificing proprietary security, integration, and sustainment criteria.
The Market Already Treats Custom Software as an Operational Response, Not an Exception
Market signals point to a structural change, not a punctual adjustment of tools.
According to the report cited by Deloitte’s State of AI in the Enterprise, 66% of organizations reported productivity improvements with AI, and over 40% of OpenAI's revenue already comes from enterprise clients. This data reinforces that AI has moved beyond peripheral experimentation to integrate into corporate budgets, operations, and strategy.
At the same time, Siemens Software warned that inaction could compromise companies' survival in the AI era. This warning is relevant because it shifts the discussion from technological enthusiasm to competitiveness. It's not just about adopting AI, but about deciding where it should be proprietary infrastructure, a configurable layer, or an outsourced service.
There are also practical examples of custom application in an enterprise context. IEL Ceará developed a custom technological solution for Sindpan, a case cited by FIEC. Without detailed operational metrics in the available material, the example still illustrates an important point: in environments with specific needs, the adopted response was customization, not standardization.
This movement does not invalidate SaaS. It redefines its role. Generic products remain suitable for horizontal needs, rapid deployment, and non-strategic use cases. But when AI begins to touch critical processes, proprietary data, and specific business rules, the discussion changes category.
What's Next for IT and Operations Leaders
The next milestones for this agenda are already set in the market.
First, the maturity of agentic engineering should continue to be monitored by metrics of productivity, code review, and software quality, as more companies move from assistive use to workflows where AI executes steps with greater autonomy.
Second, regulatory advancements in security and privacy are likely to increase pressure for transparency, traceability, and control over models and data—precisely one of the factors favoring custom architectures in critical cases.
Third, the balance between low-code/no-code, SaaS platforms, and custom development should become more sensitive to total cost of ownership. With the low-code/no-code market above US$37 billion and AI reducing some development effort, the comparison is no longer just CAPEX versus subscription but now includes sustainment, integration, and governance.
Finally, the market should continue to separate adoption from results. The data that only 29% of executives achieve significant ROI with AI suggests that the next phase will not be about experimenting with more tools, but about aligning software architecture, revenue goals, and governance.
Generic software will not disappear in the AI era. But it is no longer sufficient where operations depend on integrated intelligence, control, and real adherence to processes. As AI reduces the cost of building and increases the cost of improvising, custom software ceases to be a technical exception and becomes a strategic decision.
Related Content
Next Step with Draivv
Applying AI with results begins with choosing the right problem, data viability, and a clear business metric. Discover the AI for Business Diagnostic to transform scattered opportunities into a prioritized application roadmap.
