In the past several months, I've spent a lot of time inside large enterprises advising on AI strategy — sitting with business heads in the morning and tech leaders in the afternoon.
The pattern is remarkably consistent.
Business heads tell me tech is the bottleneck. They want access to tools, they want to move, and they feel their tech partners are either too cautious or too slow. Tech leaders tell me the business doesn't understand what it's asking for, doesn't have the skills to deploy AI safely, and doesn't have the capacity to do this well even if given the tools.
The truth is they are both right.
But ultimately this isn't about "who is right." It's about aligning the entire organization around a framework that acknowledges the need for flexibility but does it in a manner that appropriately manages the risks.
Why GenAI Breaks the SDLC
Traditional enterprise software gets built through a familiar cycle. Business writes requirements, usually imprecisely. Tech interprets them, often incorrectly. Development, testing, and rework follow, sometimes over many months.
That process was imperfect but workable, because the software was deterministic. Inputs were defined, outputs were predictable, and "done" had a clear definition.
GenAI doesn't behave that way.
GenAI tools aren't written in the traditional sense. They're prompted, evaluated, and iterated. Behavior is probabilistic. Requirements aren't specified in advance — they emerge through use. Testing isn't a phase; it's continuous monitoring.
The skill tech has spent decades refining — translating fuzzy intent into precise code — is less critical in GenAI. And the skill business has historically been weakest at — describing what it actually wants in plain language — now matters more than ever.
In GenAI, natural language is the specification. The old SDLC doesn't just slow GenAI adoption. It produces the wrong output.
Five Actions for Organizational Alignment
The business-tech AI divide is real, but the argument happening inside most enterprises is the wrong argument. Here's what organizations should do instead:
1. Give Every Employee Access to AI Tools, and Train Them on Use Cases Relevant to Their Job
Not just a pilot group or a champions program. Every employee.
Value doesn't emerge from procurement or central strategy decks. It emerges when an analyst realizes she can automate a task she does every Monday. Her manager won't identify that opportunity. She will — but only if the tools are in front of her and the training is specific enough for her to recognize what they're good for.
Create forums to share best practices, and upgrade the organization's collective skills.
2. Create a Risk-Based Tiering of Use Cases
Not every AI use case requires technology engagement. A marketing team drafting copy in a sanctioned model is a different risk profile than an underwriting decision engine.
Define clearly which tiers business can build and operate on its own using end-user AI capabilities, and which require tech partnership. Publish the criteria. Apply them consistently. The goal is to stop treating every AI request as a uniform problem.
3. Redesign the SDLC to Allow Business-Led Prototyping
Let business users describe intent, experiment, and produce working prototypes before tech is engaged. When tech does engage, they start from a functioning artifact rather than a written requirements document.
The prototype becomes the specification. The translation step that consumed months of the traditional SDLC becomes a conversation about something that already exists.
4. Require Documented Evaluation, Full Cataloguing, and Independent Monitoring
This is the condition that makes the rest of this responsible.
GenAI testing must be documented and auditable. Every AI project in the enterprise must be centrally catalogued. Production AI tools should be monitored by independent models for accuracy, drift, and bias over time.
Business-led development without this governance isn't agility. It's risk accumulation.
5. Treat This as a Capacity Reallocation, Not a Loss of Territory
Over time, more tools will be built by the business rather than by tech. Inside technology organizations, this is sometimes read as a loss of scope. It isn't.
It's a reallocation of tech's time toward the work only tech can do: infrastructure, platforms, evaluation frameworks, identity, data, security, and the governance systems above. That work is where most of the enterprise value in AI will be created — and in most organizations, it's under-resourced because the tech team is consumed by a translation layer that GenAI is making obsolete.
Business gets to build more. Tech gets to focus on what matters more. The framing that this is zero-sum is the mistake.
The Real Question
The question isn't whether business or tech is right about who should be doing what. The question is whether either side is willing to stop running a software playbook that GenAI has already outgrown.
Organizations that figure this out early won't just move faster on AI adoption. They'll redraw the line between who builds what — and that line will determine their competitive position for the next generation of technology.
GenAI changes who gets to build the tools. The companies that recognize this first will be the ones that build the future.
The divide is real. The solution requires both sides to think differently.
But the payoff is transformational.