AI Is Moving Faster Than the Systems Built to Manage It

The most interesting AI development this week isn’t simply that the technology is getting smarter. It’s that organizations are beginning to discover what happens when AI can produce work faster than their existing systems can manage it.

Several announcements Tuesday illustrate the same emerging challenge from different directions. AI is accelerating the work, but the surrounding infrastructure, permissions and management processes still have to keep up.

For business leaders, that’s becoming a much more practical consideration than another impressive benchmark score.

When AI Outgrows the Tools Around It

On Tuesday, Anthropic published an unusually revealing account of its internal software development operations.

In May, the company averaged approximately 2,500 code changes merged into its systems each day. By September, that number had grown to more than 30,000.

The increase was driven substantially by AI-assisted software development.

The problem wasn’t producing code anymore. It was managing the extraordinary volume of changes.

Anthropic’s existing version control infrastructure, built around Git, became a bottleneck. The company ultimately developed its own system, called Scion, and completed the migration in five months.

Anthropic acknowledges that its development environment is unusual. Most businesses aren’t going to encounter anything resembling this scale anytime soon.

But the underlying lesson extends well beyond software engineering.

When technology dramatically increases the speed of one activity, the constraint often moves somewhere else.

A company might use AI to produce proposals in minutes rather than hours, only to discover that its approval process still takes three days. Another might automate financial analysis while leaving the reconciliation and decision-making process largely unchanged.

The productivity improvement is real, but the full benefit remains trapped inside an operating model designed for slower work.

Anthropic’s experience offers a useful reminder that adopting AI and modernizing a business are related activities, but they aren’t necessarily the same thing.

Enterprise AI Is Getting Closer to the Actual Work

Also Tuesday, Workday announced significant changes to how AI agents can interact with its human resources and financial management systems.

The company introduced broader access to its Trusted Agent Platform, allowing AI agents to work directly with business information and processes without requiring someone to navigate the traditional software interface.

Importantly, Workday says those agents remain subject to existing employee permissions, approval requirements and audit controls.

A related partnership with Anthropic will allow users to work with Workday’s Sana agents through Claude, bringing AI reasoning together with the business rules already maintained inside Workday. That integration is currently in early access, with broader availability expected later this year.

The distinction matters.

For the past several years, much of enterprise AI has operated alongside existing software. Employees ask questions, generate documents or analyze information, then return to their business applications to complete the work.

Increasingly, AI is being integrated directly into those applications and the processes they support.

That creates opportunities to reduce administrative work, but it also raises an important operational question.

When an AI agent can actually change a business record, who has authorized that change?

Workday’s announcement suggests that established enterprise software providers increasingly recognize that permissions, approvals and accountability need to travel with AI capabilities.

The model might generate a recommendation, but the organization’s existing rules still determine whether that recommendation can become an action.

That’s a meaningful step toward making AI useful in day-to-day business operations.

Regulators Are Asking Similar Questions

The same concerns are beginning to appear in financial regulation.

Reuters reported Tuesday that banking supervisors in the United States and Europe are intensifying their examination of AI systems, particularly autonomous agents operating within financial institutions.

American regulators are asking banks how AI agents are monitored, what authority they possess, and whether their actions can be audited or stopped.

European supervisors are also emphasizing cybersecurity recovery and the ability to restore operations following serious disruptions.

These inquiries have not yet translated into a broad new set of prescriptive requirements. Regulators are still evaluating the risks.

Nevertheless, the direction is revealing.

The question is moving beyond whether an organization can use AI effectively. Increasingly, it is whether the organization can demonstrate control over what AI does.

That principle applies just as readily to a manufacturing company, hotel operator or professional services business as it does to a bank, even if the consequences and regulatory requirements differ considerably.

Meanwhile, the Models Keep Improving

Google added another development Tuesday with the broader release of Gemini 4 Argon, its latest flagship AI model.

Google reports improvements in complex, multi-step tasks involving coding, finance, legal analysis and scientific research. The company also expanded access through its consumer applications and developer tools.

Those capabilities deserve attention, although Google’s performance claims will need to be evaluated against independent testing and actual business applications.

The larger point is that increasingly capable models are becoming available to more organizations.

That makes the surrounding business systems even more important.

A better model may complete an assignment faster or more accurately. It doesn’t automatically resolve unclear authority, inconsistent information or an inefficient approval process.

Those remain management responsibilities.

Strategic Perspective

Tuesday’s developments point toward an important transition in business modernization.

For the first phase of AI adoption, organizations concentrated on what the technology could produce. Documents, analysis, software and customer communications became faster and easier to generate.

The next phase will increasingly involve how that work moves through the organization.

That includes the systems AI can access, the decisions it can make, the approvals it requires and the ability to verify its actions.

There’s a practical opportunity here for business leaders.

Rather than evaluating AI solely by how much time it saves on an individual task, consider what happens to the work after AI finishes its part.

If the output still waits in the same queue, requires the same manual handoffs or gets trapped in an unchanged approval process, the organization may be capturing only a fraction of the potential benefit.

The companies that get the most from AI may not be those using the most sophisticated models.

They may be the ones willing to modernize the operations surrounding them.