The AI conversation in boardrooms shifted a couple years ago. We’ve gone from debating whether artificial intelligence will transform business operations, to how quickly we can move from pilots to production implementations with measurable impact. The organizations that get this transition right won’t just automate tasks; they’ll create digital workforces that amplify human capabilities and unlock incredible organizational agility.
After the last year of conversations with technical and operational leaders, one thing has become crystal clear: while the gap between flashy AI demos and production-ready multi agent architectures is closing rapidly, it’s littered with well-intentioned expensive failures. The most successful organizations aren’t necessarily the ones with the biggest AI budget – they’re the ones applying the same rigor to AI adoption that they would to any major organizational transformation.
The foundation: Understanding how to get started
Sam Sharaf, Sr. Director, Product Management at Salesforce, recently outlined 10 critical considerations for creating agent-led architectures, and they’re more than just technical checkboxes, they’re the underpinnings of a platform for digital labor. When we talk about agent cards, discovery, communication patterns, observability, security, and scalability, we’re really talking about framing a digital workforce with defined work norms, standards, and guardrails within your enterprise.
In our whitepaper “From Digital to Agentic Transformation,” we go deeper into the need for clean accessible data to support agentic interactions, so in this post we’ll focus more on the organizational approaches and strategies for successfully adopting a digital labor force.
Just like every successful human workforce has roles, communication tools, security clearances, performance metrics, and governance structures, digital workforces require the same foundational elements. However, digital workforces carry the added risks of making mistakes at machine speed, requiring explicit definition of norms and work patterns that human workers would learn from onboarding and the company culture.
This isn’t to say that treating agents like software with a traditional product lifecycle is wrong, but one needs to layer on additional considerations and constraints to avoid common pitfalls when a deterministic software perspective is applied to agentic workflows.
The organizations that recognize this early, and invest accordingly, will see transformational results. Those that treat AI agents as sophisticated chatbots will be disappointed by the impact and scope of transformation.
Strategic planning: Beyond the tech stack
Business outcomes first, technology second: A common mistake executives make is leading with the technology instead of the business outcome. Large language models are impressive and autonomous agents can handle complex workflows, but none of that matters if you can’t articulate exactly what business problems you’re solving and how you’ll measure success.
Start with pressing operational pain points. Where are your teams spending disproportionate time on routine tasks? Where do information silos create bottlenecks? Where would 24/7 availability genuinely transform customer experience? The answers to these types of questions become the foundation of your agents’ “job descriptions,” and like a hiring process, clarity on the jobs to be done for the role, drives the outcomes.
Executive sponsorship isn’t just helpful; it’s essential. AI adoption touches every organizational layer, from data governance to employee training to customer interaction models. Without clear executive mandate, aligned expectations, and measurement frameworks, even technically sound implementations struggle to scale beyond departmental pilots.
Immediate impact areas and what’s next
Thankfully, AI adoption benefits have moved from the theoretical to the measurable. Early AI implementations focused on employee productivity are proving highly impactful across industry sectors. Knowledge retrieval that once took hours now happens in seconds. Research and qualitative analysis that used to require dedicated teams can be delivered with AI capabilities. Content creation, editing, and decision validation are becoming dramatically more efficient.
Customer-facing workflows around service and support have demonstrated equally compelling results. The key difference between successful implementations and disappointing ones isn’t only the underlying technology – it’s how well the organization prepared for the human and process changes that AI enables.
The current wave of agentic implementations are delivering single agent processes for both employees and customers that consume existing software tools and workflows exposed through traditional APIs or AI specific interfaces called Model Context Protocol (MCP) Servers. To ensure these agents are using the right data, specialized vector databases and AI operations platforms ensure answers come from reliable internal and trusted external sources.
By leveraging existing enterprise capabilities and workflows, abstracted with clean interfaces, this first wave of agents can solve problems in consistent ways, reducing the risk of hallucinations and simplifying governance as they begin to perform more complex tasks across agent maturity levels two and three.
We’re just on the cusp of truly agentic architectures (maturity level four) where AI agents will perform complex tasks with other AI agents across domain and company boundaries to get work done in dynamic and novel ways. With solid strategy, a foundational technical framework, and a ready culture, organizations will be adaptable enough to adopt innovations at a staggering pace to get reliable measurable outcomes.
Thinking like a CIO and a COO
Many technical leaders get tripped up approaching AI agents solely as software deployments, when they should be thinking like operational leaders, not just technology platform owners. Because LLMs are trained with human language, they seem to exhibit human-like reasoning patterns, including human-like mistakes. Given current agent capabilities, it’s useful to model agents as digital employees with varying experience levels rather than deterministic state machines.
This means adopting a “hire to retire” lifecycle for agents. Define job descriptions with clear scope and domain boundaries. Evaluate different AI models and agentic implementations for role fitness. Onboard agents with documentation and tools, monitor performance, and provide feedback. Most importantly, prepare to retire agents that no longer meet organizational needs as technology evolves.
Organizations embracing this metaphor (within limits) achieve outsized outcomes because they apply appropriate governance, evaluation, and management practices from day one, accounting for behavioral variability. They plan for evolution and the ability to retire agents and frameworks that no longer fit needs.
Traditional software testing approaches with basic pass/fail metrics don’t map well to AI systems’ probabilistic nature. Instead, you need evaluation frameworks accounting for the iterative, improvement-oriented nature of agent performance.
Worker patterns and organizational design
The current wave of single agents handling isolated tasks is the beginning of the value journey, not the end. The real transformation happens when you leverage multi-agent systems that align with effective human team structures. Instead of super agents reigning over all processes, we anticipate that agent specialization based on functional domains, or process knowledge will both accelerate outcomes due to specialized knowledge and capabilities, as well as improve how employees learn to interact with AI capabilities.
Some agents will specialize in research and analysis. Others focus on execution and action-taking within enterprise systems. Still others can coordinate across teams and manage complex workflows with customers. The deployment structure and agent scope will evolve as will the team structures of the people who interface with them.
From a risk perspective, this requires thinking carefully about trust frameworks, human-in-the-loop intervention points, and escalation protocols. You’ll want to consider a variety of governance questions. How do your agents collaborate without creating circular conversations? How do you maintain accountability when multiple agents contribute to an outcome? How do you ensure that agent teams don’t develop their own communication patterns that become opaque to human oversight?
The patterns for how AI agents interact with each other and people will also evolve over time with some agents interacting directly with people (experience agents), and others interacting solely with other agents and tools (headless agents). As agentic capabilities improve department boundaries may also shift based on end to end process ownership vs. traditional corporate function ownership. These transitions will need to be managed carefully and intentionally to avoid alienating stakeholders and creating confusion.
The reality of adoption: Iteration and user-centricity
Successful agent adoption is messy, iterative, and requires serious change management. Initial implementations won’t be perfect. Early user groups will discover unanticipated edge cases. Governance frameworks need constant refinement. Most importantly, strategy must be transparently communicated and informed by broader stakeholder discussions.
This isn’t a bug – it’s a feature. Organizations that budget for this iterative reality and plan accordingly will outperform those expecting linear progress from pilot to production.
Start with volunteer early adopter teams – with well-understood problems – who are genuinely excited about working with AI agents. These users provide candid feedback and work through initial friction. They become internal champions and help refine agent capabilities before broader rollouts, avoiding organizational resentment and confusion.
Invest heavily in training and support during early phases. Success metrics should include user confidence and adoption rates, not just technical performance indicators. An agent that performs perfectly but users don’t trust or understand won’t deliver business value.
Plan for multiple iteration cycles. Initial implementations reveal integration challenges, hidden workflow dependencies, and organizational dynamics not apparent during design. Build buffer time and budget for these learnings; they’re investments in long-term success, not setbacks.
Identify team members with early successes as potential ambassadors and create a cross-functional Center for Enablement (C4E). This group should educate and enable other teams for broader rollouts, not become gatekeepers or bottlenecks like Centers of Excellence often become.
The path to the adaptive enterprise
The transition from a composable enterprise to an adaptive enterprise will happen when organizations can align autonomous connected agents with existing composable capabilities to dynamically create and evolve business processes adaptively based on business context. This necessitates the strategy, technology, and culture foundations outlined above combined with a programmatic commitment to evolution and improvement.
With all major software vendors bundling agentic capabilities into their platforms, organizations need ways to observe, manage, and govern execution to achieve consistent outcomes across diverse enterprise agents. The key is leveraging composability principles by encapsulating existing enterprise capabilities and data into reusable, governed, observable assets for both traditional process orchestration and agentic transformation, as well as layering on governance and orchestration frameworks with supporting tools and capabilities for conversation and work management across these platforms.
To begin this agentic journey, companies must expose enterprise capabilities for agentic consumption with APIs or emerging protocols like MCP. Then begin authoring and deploying agents with flexible frameworks, and governing agent-to-agent and agent-to-tool communication through emerging protocols using gateways or sidecar processes that function across agentic platforms. As new standards emerge and take hold, flexibility to adopt and sunset capabilities becomes crucial, so technology partners focused on evolution with a pluggable approach are key to success.
Organizations that get this transition right won’t just automate tasks – they’ll create digital workforces that amplify human potential and unlock entirely new ways of creating value. The foundation is treating digital labor as seriously as human labor, with appropriate governance, lifecycle management, and change processes that account for the unique characteristics of AI systems.
The companies that embrace this challenge with proper strategic planning, technical foundation, and organizational readiness will define the next era of business operations.
