Managing agentic AI’s speed, scale and sprawl: Insights from Think 2026 

The faster new services and features go from ideation to real-world deployment, the harder it is to maintain oversight and organizational alignment. Agentic coding is a powerful tool, but tools are not systems—and systems are what drives enterprise evolution. A tool provides velocity and momentum; a system ensures that tools and teams move together in the right direction and stay on course.

At Think 2026, IT and business leaders from Disney, BNP Paribas, New York Life, Warby Parker, EY and many more joined IBM leaders to explore ways to build an agentic enterprise at speed and scale without sacrificing control or cohesion. “The question comes down to: how deeply is AI embedded in your business processes?” said Arvind Krishna, IBM Chairman and CEO, in his opening keynote. “Is it a part of the enterprise? Or is it something on the side?”

On April 28, we announced the global availability of IBM Bob, an AI-first development partner for the entire software development lifecycle (SDLC). All week at Think 2026, enterprise partners across a wide array of industries—some of whom have been working with Bob since last year—talked about how they’ve used Bob to build systems tailored to their specific needs.

IBM built Bob in response to what enterprise teams have consistently asked us for: a system that transcends isolated code snippets and can coordinate cost, code changes and delivery organization-wide—without losing the speed that makes agentic AI so valuable. “Bob is not just a coding assistant,” said Krishna. “When we say software development, we mean architecture, planning, code generation, testing, security.” 

Bob is model-agnostic and built to work with the deployment environments and technology stack that you already operate in. Research from the IBM Institute of Business Value (IBV) found that 82% of executives expect to their AI capabilities to rely on a multi-model approach in 2030. “We do believe that the best strategy is to mix, and to be sure that at some point you are not locked in with one vendor, with one channel,” said Jean-Michel Garcia, CTO, BNP Paribas, in a conversation about agentic orchestration and governance with Andy Baldwin, Senior VP, Consulting Offerings and Growth for IBM, on Tuesday. “That’s why we’re experimenting with Bob.”

During the conference’s IBM Partner Plus keynote on Monday, Zachary Greenberg, CEO, Nexar, explained how that openness is helping his company innovate. IBM Bob helped Nexar “take this incredibly [valuable] data pipeline we have and actually build agents around it that we can then deliver directly to customers.”

Here at IBM, over 80,000 employees—over a quarter of our total workforce worldwide—are already using Bob, enjoying a 45% gain in productivity on average. It’s robust enough for software engineers yet intuitive enough for business leaders, which enables organizations like Nexar (and IBM) to get everybody operating in the same system. “Everyone, from engineers to product managers all the way to the finance team, is leveraging [Bob] to help accelerate our business meaningfully,” said Greenberg.

AI might simplify software development, but enterprise complexity isn’t going away. Companies are still dealing with legacy systems that can’t be replaced, costs that can’t be ignored and governance that can’t be bypassed. “If you look at software systems, about 60% of work is migration, modernization and maintenance,” said Sunderasan in a Spotlight Session on modernizing the SDLC with agentic AI. “New code development is only about 15%.”

Traditional software development, and therefore the traditional SDLC, is inherently deterministic. Code provides explicit instructions, and the same input will always yield the same output. But AI agents, like the large language models (LLMs) that power them, are inherently probabilistic. “You need a layer in between AI and your application that tells you how to take the probabilistic and make it trustworthy,” Sunderesan explained.

IBM Bob was built to deal with this reality. In keeping with DevSecOps principles, it shifts security left and implements human-in-the-loop review at every critical step: governance and security are integrated into every step, rather than bolted on at the end—which the audience witnesses first-hand in a video demonstration of IBM Bob modernizing a legacy codebase.

Following the demo, Michael Kwok, VP, IBM Bob, and Canada Lab Director, IBM, convened a panel of executives whose organizations were early testers and adopters of IBM Bob.

Chris Aiken, Chief Product Officer, EY, appreciated Bob’s powerful adaptability in a highly regulated environment. “The ability to customize the behavior of Bob through the use of modes was really critical for us,” he explained. Masanori Unno, VP, NIC Partners, described Bob’s role in helping their often change-averse clients to modernize and accelerate onboarding. Christophe Boulange, Cloud Director, BNP Paribas, noted that Bob helps the venerated bank keep up with the pace of innovation they’re seeing from emerging competitors in the fintech and neobank space.

Blue Pearl CEO Saireshan Govender began testing the beta version of Bob after IBM TechXchange in October. Their legacy Java code base had about 127 deprecated APIs. “The initial project plan was about nine months, with 14 Java developers, to modernize that code base,” he recalled. “Bob was able to modernize that code base in three days.” The turnaround was so quick that Blue Pearl’s technical developers weren’t sure if they could trust the results—but after a few days of review, they were sold. “The return on investment is speaking for itself at the moment,” said Govender.

“I lived a sport that was very individual,” tennis legend (and co-founder of Agassi Sports Entertainment) Andre Agassi told Andy Baldwin during Tuesday’s keynote on orchestration and agentic AI governance. “It was a sport that didn’t have dynasties to help pass on best practices and to make the growth of it exponential.” Tennis players can make it on their own, but a company’s agentic AI initiatives need institutional knowledge and coordination to succeed in the long term. 

True agentic transformation is a team sport: An assortment of isolated AI projects all moving in their own directions might each achieve some individual success, but they won’t move your company forward. And as AI grows more autonomous and building powerful AI agents gets quicker and easier, tracking AI sprawl across your organization gets harder: IBV research shows that only 18% of organizations maintain a current and complete AI inventory, and 68% of executives worry their AI initiatives will fail due to lack of deep integration.

IBM Bob is therefore designed to integrate seamlessly with watsonx Orchestrate (and vice versa), which provides a centralized control plane for agents built in Bob (or elsewhere). As a team, they drive speed and scale without sprawl. Following Baldwin’s conversation with Agassi, IBM’s Neel Sunderesan and Sanah Pallithotungal demonstrated exactly how Bob and Orchestrate work together: Orchestrate reports a real-time problem, suggests a fix and provides Bob with the context it needs to immediately build new agents that address the issue.

To learn more about orchestrating, accelerating and governing your agentic systems, watch the full keynote.

On the second day of Think 2026, IBM announced six key enhancements to watsonx Orchestrate. Collectively, these capabilities provide a single operational layer to centralize, manage, monitor and govern all your AI agents across frameworks and environments.

Building AI agents represents only about 20% of the agent development lifeycle (ADLC): most of the ADLC comprises testing, deploying, operating and monitoring agentic systems in production. Whether you’re building custom third-party agents, building agents in Orchestrate or using pre-built options from the Orchestrate agent catalog, Orchestrate’s control plane provides real-time insights and alerts across your entire agentic ecosystem, centralizes guardrail enforcement, identity/credential management and audit logging, and facilitates instant action when the need arises.

AI agents can move faster than conventional manual review cycles, so companies need oversight that can keep up. IBV research from April 2026 found that, on average, organizations committed to orchestration-led governance:

• Were 13x more likely to be scaling their AI practice

• Experienced 30% fewer irregularities (which cost a $20B company about $140M annually)

• Enjoyed 20% greater ROI

• Found 169% greater transparency

• Benefited from 132% stronger data-privacy protection

In an enterprise, you don’t win by creating the most agents, nor do you win inside one silo. You win by running them safely, reliably, and measurably—at scale and across systems.