Organizations can now build software faster than they can agree on what to build. AI makes a feature that once took weeks possible in days. A product manager can create a prototype without waiting for an engineer, while a support team can automate a workflow on its own. Working software can emerge from almost anywhere in the organization. That sounds like an unambiguous win, until everyone starts building.
Implementation is no longer the filter
For decades, engineering capacity acted as a natural constraint. Ideas were abundant, while turning them into working software required scarce time and specialized skills. Most ideas never made it past a document or a conversation because implementation was expensive. That cost encouraged teams to prioritize and align before committing significant engineering effort.
AI changes that equation. When an experiment takes an afternoon, you no longer need a roadmap slot or a dedicated team to try it. That’s wonderful because more people can test their ideas with real users instead of debating them in the abstract. It also means that an organization can produce far more working solutions than it can reasonably adopt. The bottleneck moves from building solutions to deciding which solutions become the organization’s solutions.
AI changes the economics of alignment
Imagine two teams facing a similar problem. They can spend time understanding each other’s work, negotiating priorities and agreeing on a shared direction. They can also build separate solutions. When implementation takes months, the cost of duplicate work gives both teams a strong reason to align. When each team can build its version in two days, another meeting suddenly looks expensive.
So, what will a team do when building is cheaper than agreeing? It will often build. That’s a rational response to the incentives in front of it. Implementation costs have dropped dramatically, while alignment still requires context and trust. It still takes conversations between people with different goals.
This shift can make a well-run organization look fragmented. Teams move quickly and solve real problems for their immediate customers. Each decision makes sense locally. Yet the organization gradually accumulates several answers to the same question, each with its own assumptions and maintenance burden. Nobody chose the fragmentation: it simply emerged from a series of reasonable decisions.
Parallel exploration is healthy
Preventing duplicate work altogether would waste much of what AI makes possible. Parallel exploration helps teams challenge assumptions and discover approaches that a central plan might miss. Two teams solving the same problem may learn that they were solving different problems all along. They may also discover that one approach works better than the other. You want that learning before committing the whole organization to a standard.
The challenge appears after the experiments produce useful results. How do you recognize that several local solutions now represent one shared need? Who compares them, and who has the authority to choose a direction? How does the chosen solution absorb what the other experiments learned? Organizations need a way to answer these questions without turning every experiment into a governance project before it starts.
Exploration and convergence happen at different times. Early on, variety creates learning. Later, the same variety creates friction for everyone who needs a dependable answer. The skill is knowing when the value of another experiment falls below the cost of another option. Too early, and you standardize before you understand the problem. Too late, and every team has something it can no longer afford to abandon.
The audience eventually becomes the customer
Several overlapping solutions may coexist quietly while each team serves its own audience. The cost becomes visible when someone crosses those boundaries. Another team looks for a capability and finds three. A customer encounters different ways to solve what appears to be the same problem. An AI agent recommends whichever approach happens to be documented best, rather than the one the organization wants people to use.
Who pays for that inconsistency? At first, it’s the people trying to navigate it. They compare options and ask around to learn which guidance still applies. Later, the organization pays through duplicated maintenance and support work, while customers encounter conflicting experiences and leaders see initiatives pulling in different directions. By the time the full cost becomes visible, several teams already depend on their own solutions.
Software makes this pattern easy to see, but the same economics apply throughout an organization. Marketing teams can create overlapping assistants. HR teams can automate onboarding in different ways, while consulting teams build competing accelerators for similar customer needs. These outcomes can come from capable teams responding quickly to local needs. And AI makes each response cheaper, so organizations accumulate them faster.
Convergence becomes a capability
As experimentation gets cheaper, convergence becomes more valuable. The more ideas you can explore, the more important it becomes to recognize which ones deserve broader adoption. That requires visibility into what teams are building and a credible way to compare the results. It also requires teams to let go of a working local solution when another option serves the organization better. That last part in particular is hard, because working software creates ownership remarkably quickly.
Perhaps one day we’ll measure how well organizations do this. DevOps gave us deployment frequency and mean time to recovery because those measures exposed capabilities that mattered to delivering software. AI may give us another: time-to-convergence, the time it takes to recognize overlapping solutions and align around the one that best serves the organization and its customers.
A short time-to-convergence lets teams learn before divergence becomes expensive. Healthy organizations will keep room for local experiments and make the path from a successful experiment to a shared capability clear. They will know who can make that decision and what evidence matters. They will also show how the chosen solution earns adoption.
Can convergence keep pace with creation?
Large organizations have always balanced team autonomy with company-wide coherence. AI changes the economics of that balance. More ideas can become a working solution, and more teams can solve problems independently. Each local optimization becomes easier to create, and the ability to converge now needs to grow at the same rate.
Building faster is a remarkable opportunity. It gives organizations more chances to learn and more ways to serve their customers. But the value comes from turning the best experiments into capabilities that others can discover and reuse with confidence. The question for organizations shifts from how quickly they can create to something harder: can they agree just as fast?