The changing economics of enterprise software

For years, organizations have been forced to make a series of rational compromises about their technology decisions. With agentic AI and the convergence of practices around agentic engineering, several of these compromises no longer apply.

Agentic engineering is the practice of software engineering redesigned for delivery to be executed by AI agents, where every discipline from requirements and development to testing is restructured around context curation, output verification, and human-governed autonomy.

These practices collapse the cost, time, and risk curves that defined enterprise software decisions for decades, invalidating many of the assumptions that still shape IT strategy today.

But the technology motions and decisions of the past solved for a different set of challenges than those faced today, each offering compromises of their own. Vendor platforms won out over custom builds to minimize risk, all at the cost of flexibility and differentiation; agile adoption took hold for its speed and rapid feedback loops, deprioritizing the solution and architectural completeness that were meant to follow; consolidation projects kept getting deferred because the upfront investment in untangling interconnected systems exceeded the current ongoing run costs, leaving operational drag in place. Through all of this, institutional knowledge eroded as people moved on, because capturing it properly was never a priority.

None of these were mistakes, but rather pragmatic calls based on real constraints such as the cost of building well, the time to gather and validate requirements, the sheer difficulty of modernizing while keeping the lights on. However, those constraints have shifted, and the strategies built around them deserve re-examination.

The shift that changes the math

Enterprise technology strategy has been traditionally shaped by risk management. Building software right was risky, with long timelines, scarce talent, and fragile institutional knowledge. 

Agentic engineering changes the calculation behind each of those trade-offs:

• Requirements that previously took months of manual efforts to gather can now be rapidly extracted from legacy codebases, serving as a starting point for modernized systems which preserve up-to-date documentation.
• Detailed designs for new systems, often abandoned in favour of agile architecture techniques, can now go deeper again by offloading the time-consuming schematics or in-depth documentation.
• Business-as-usual changes can now be prototyped in hours, enabling richer and faster discussions to define requirements and produce structured, more informed asks for changes.
• Large volumes of code can be produced in minutes as opposed to days, once clear context and direction are set.

These are structural changes that have deep implications for the unit economics of understanding, changing, and maintaining software—a complete paradigm shift.

Three opportunities that emerge

With this shift in technology transformation economics, value opens up in three key areas.

The path forward

Agentic engineering doesn’t get switched on overnight. Organizations have to grow into it, building agent-compatible architectural foundations for software, establishing engineering patterns, and refining the methodology that supports the rapid iteration loop all while developing the institutional muscle and trust to operate at progressively higher levels of autonomy. Those who start now will build the capability iteratively with each cycle and project reinforcing the next. Most of our clients are picking either narrow pilot domains, domains where institutional knowledge has high risk of decaying, or domains where there is a strong business to drive change of differentiation or simplification.

The evidence is already emerging. Many organizations have adopted AI-assisted development but remain stuck at code suggestions and small-scale experiments, generating fragments of solutions faster without changing the underlying delivery model. Closing the gap isn’t a tooling problem but a methodology and approach problem. It requires architectural discipline, new ways of working, and organizational commitment.

Those who defer will face a widening gap. Not just because competitors are modernizing faster, but because the capability to operate this way is itself a learned competency that takes time to develop. Waiting doesn’t push back the payoff here. It pushes back the organizational learning needed before the payoff is even possible.

Deloitte is redefining the methodology, engineering approaches, and delivery model with our most forward-thinking clients. We’ve developed repeatable approaches to how applications and systems should be structured and delivered for different business archetypes, and we’re testing and refining them in live engagements.

Knowing which technologies to choose, which architectures fit which domains, and how to sequence decisions for a given business context provides an edge. But a lasting edge comes from what surrounds technology: a methodology that keeps pace with the speed of change and can be taught, governance that earns trust, and the domain knowledge deep enough to make the right calls in this new paradigm.

This piece started with a simple observation: the compromises organizations made over the last decades were rational, but the constraints behind them have changed. Industry leaders agree that agentic AI is transformative in enterprise software delivery, and those who start early and design deliberately will reap the benefits.