Agentic AI is moving from research vision to near-term enterprise reality, but leadership readiness often determines outcomes. This article is based on a podcast conversation hosted by Rob Price, where Welsch discussed what makes agentic AI different, how to evaluate vendor claims, and why organizational change is the second half of the equation. The discussion is especially relevant for CIOs, CTOs, CHROs, and business leaders preparing for workforce transformation and new operating models.
Welsch also referenced his book, The AI Leadership Handbook, plus his podcast “What’s the BUZZ?—AI in Business,” his newsletter “The AI Memo,” and LinkedIn Learning courses on AI agents—resources aimed at practical AI adoption, not hype.
Executive Summary
- Agentic AI is goal-driven software that decomposes tasks into sub-goals and executes them.
- Vendor “agentic” claims require scrutiny: autonomy, grounding, planning transparency, and controls matter.
- Early wins often appear in customer service with human review in the loop.
- AI adoption succeeds when aligned to business strategy, KPIs, and workforce upskilling.
- Technology is only half the equation; organizational change completes the value chain.
Key Takeaways
- Welsch defines agentic AI by autonomy toward a goal, not by process automation labels.
- Organizations with existing AI “muscle” (ML and GenAI) are better positioned for agentic AI.
- Leaders should expect hype, rebranding, and conflation between workflows, agents, and automation.
- Customer service is a practical starting point, but quality and reputation risks require oversight.
- Success depends on selecting KPI-linked use cases with business stakeholders, not IT alone.
- Training, guardrails, and guidance on data use determine whether licenses translate into value.
- Exploration is necessary, but chasing technology without strategy invites “proof-of-concept purgatory.”
What is Agentic AI?
Andreas Welsch describes agentic AI as software that can accept a user-defined goal, break it into sub-goals, work on those sub-goals, and then return results to the user or customer. In his view, this differs from systems that only generate content: agentic AI adds a layer of autonomy that can plan and execute actions. That autonomy also raises new questions for leaders: how decisions are made, how outputs are grounded in company policy, and how human review should be structured—especially in early deployments.
Agentic AI is not just “automation with a new label”
Welsch notes that the market regularly rebrands capabilities when a new term becomes popular—similar to how “cloud” was applied broadly in earlier cycles. In the current wave, “agentic” can be attached to many productivity and automation offerings, creating confusion between agentic workflows, AI agents, and traditional automation approaches.
For executive buyers, the implication is straightforward: do not accept a label as a capability. Evaluation needs to focus on what is actually “under the hood,” and whether the solution truly supports goal decomposition, planning, and controlled execution.
Key Insight: Welsch warns that “agentic” is becoming a catch-all term, much like “cloud” once was. Leaders should press vendors on what autonomy means in practice, what models power the agent, and how planning, execution, and grounding are documented and controlled.
How to evaluate agentic AI vendor claims
Welsch recommends probing for specificity during evaluations. Key questions include: how autonomous is the system, what happens when it receives a goal, whether an LLM is being used to interpret intent, and how the agent’s plan is documented and monitored. Leaders should also ask how the solution is grounded in company policies, especially for customer-facing scenarios.
He also emphasizes behavioral consistency. If an organization expects human agents to communicate in a certain way, AI agents must be designed to align with those standards. Without that, brand risk grows—even if the technology is impressive.
Key Insight: According to Welsch, vendor evaluation should focus on autonomy, transparency, and grounding. Executives should ask how the agent plans, how it executes, how it is constrained by policy, and how the organization can audit behavior—especially when customers experience the outcome directly.
Where organizations are starting: customer service as a pragmatic use case
Welsch points to customer service as an area showing early promise. He observes that many inquiries repeat across common topics—product questions, availability, payment terms, and sales-related requests. An agent can be supplied with a knowledge base to help customer service representatives find and draft answers faster.
However, he recommends keeping a human review step, particularly in the early days. The risk is not abstract: an agent can go “off on a tangent” or propose inaccurate information. Customers typically hold the company accountable regardless of whether the response originated from AI or an employee.
Key Insight: Welsch highlights customer service as a sensible starting point for agentic AI, but advises a “human-in-the-loop” approach. Reputation risk remains with the company, and early agent behavior can be unpredictable without review, guardrails, and well-grounded knowledge sources.
Organizational readiness: the missing half of agentic AI
Welsch argues that agentic AI adoption depends on organizational change as much as technology. Teams may worry that “superhuman” AI will replace jobs, while simultaneously doubting whether it can be trusted. Education and awareness become essential, and leaders should encourage experimentation within defined guidelines and guardrails.
He also notes that organizations with established AI capabilities—machine learning, AI literacy, and GenAI experience—tend to be better prepared. Still, he believes organizations earlier in their AI journey should not ignore agentic AI because they “have a lot of catching up to do,” and practical opportunities exist.
Why senior leaders often feel ahead of the organization
In the conversation, a recurring dynamic emerged: senior leaders may be enthusiastic about “the new art of the possible,” while the organization needs support to come along. Welsch frames this as a leadership responsibility—encouraging adoption, setting boundaries, and investing in skills so the workforce can use new capabilities effectively.
Strategy first: avoid chasing shiny objects
Welsch repeatedly returns to one leadership principle: technology strategy must align with business strategy. He observes that leaders can be pulled into action by headlines, vendor messaging, investor narratives, and peer pressure—the “FOMO” effect. That urgency can be useful, but it can also lead to disconnected proofs of concept that fail to produce measurable value.
His recommendation is explicit: define why agentic AI is being used, what business value it should create, and which KPI it should influence. Then measure that KPI before, during, and after the initiative.
Key Insight: Welsch’s core message is that AI programs fail when they are technology-led instead of business-led. Leaders should tie agentic AI to strategy, specify KPI impact, and measure outcomes across the lifecycle—otherwise initiatives risk becoming pilots that never scale or deliver value.
From experimentation to value: the adoption sequence Welsch recommends
Welsch outlines three practical steps that increase the likelihood of success. First, technology teams should get hands-on with frameworks and solutions to understand capabilities and limitations. Second, teams should identify initial use cases that deliver business value, which requires active partnership with business stakeholders—not IT working alone.
Third, leaders must bring people along. Welsch points to a common pattern: organizations buy “copilot” licenses or tools, but do not achieve value because employees are expected to “just know” how to use them. He compares this to the learning curve in office software—some users excel quickly, others need support and practice.
For agentic AI, he recommends investment in training and upskilling: how to use the tools, what the limitations are, when to use them, when not to use them, and what data is acceptable to input.
Leadership Implications
- Set governance early: define guardrails, review requirements, and policy grounding for customer-facing agentic AI.
- Demand transparency from vendors: require clarity on autonomy, planning documentation, and auditability.
- Design for human oversight: keep human review in place while agentic AI maturity improves.
- Choose KPI-linked use cases: align agentic AI initiatives to business strategy and measurable outcomes.
- Enable the workforce: fund training on usage, limitations, and acceptable data handling to avoid shelfware.
Why this conversation matters
This podcast conversation matters because it translates agentic AI hype into executive decision criteria: autonomy, vendor reality checks, and adoption mechanics. Welsch’s perspective is grounded in long-cycle enterprise experience, including decades around enterprise applications and work with Fortune 500 leaders on where AI delivers benefits and how adoption succeeds or fails.
The audience for these insights includes leaders accountable for governance, strategy, and workforce transformation. Agentic AI raises practical questions that span multiple functions: technology selection, risk management, customer experience, and employee enablement.
Welsch’s broader work—his book The AI Leadership Handbook, his podcast “What’s the Buzz: AI in Business,” and “The AI Memo”—focuses on repeatable leadership behaviors for implementing AI successfully across machine learning, generative AI, and agentic AI.
Conclusion
Agentic AI can unlock new productivity and service models, but only when leaders separate substance from buzzwords. Welsch’s guidance emphasizes disciplined evaluation, KPI-based use case selection, and workforce enablement—paired with governance and human review to manage risk.
For executives, the path forward is not to chase agentic AI as a trend, but to adopt it as a strategic capability tied to business outcomes and responsible operating practices.
FAQ
1) What makes agentic AI different from generative AI?
Agentic AI focuses on pursuing a goal by breaking it into sub-goals and executing tasks, not only generating text or content. Welsch describes it as adding autonomy—planning and action—beyond generative AI’s ability to produce outputs.
2) How should leaders define “agentic AI” for procurement discussions?
Leaders should define agentic AI by behavior: goal intake, task decomposition, execution, and results delivery. Welsch advises asking vendors how autonomous the system is, how it plans, and how actions are documented and controlled.
3) Why does Welsch recommend human review for early agentic AI deployments?
Human review reduces the risk of inaccurate or off-track responses, especially in customer-facing work. Welsch notes that customers do not care whether AI or an employee answered; the company’s reputation remains accountable for errors.
4) What is a practical first use case for agentic AI in enterprises?
Customer service is a practical starting point because many inquiries repeat across common topics. Welsch suggests using an agent grounded in a product knowledge base to draft responses faster, with employees reviewing before sending.
5) How can executives avoid “proof-of-concept purgatory” with agentic AI?
Executives can avoid stalled pilots by aligning technology strategy with business strategy and KPIs. Welsch recommends defining the business value, selecting KPI-linked use cases with stakeholders, and measuring outcomes before, during, and after.
6) What questions should be asked to verify a vendor’s “agentic” claims?
Key questions include the degree of autonomy, whether an LLM interprets intent, how planning is documented, and how grounding in company policy is enforced. Welsch encourages leaders to assess what is truly happening “under the hood.”
7) Why do AI tools often fail to produce productivity gains after purchase?
Tools often fail to deliver value when organizations expect employees to adopt them automatically. Welsch points out that “copilot” licenses alone do not create outcomes; training, usage guidance, and clarity on limitations and data rules are required.
8) How should leaders think about workforce transformation with agentic AI?
Workforce transformation requires addressing both excitement and fear: employees may worry about job impact and distrust outputs. Welsch recommends education, encouraging experimentation within guardrails, and upskilling people on when to use agentic AI and when not to.
9) Are only AI-mature organizations ready for agentic AI?
No—Welsch says AI-mature organizations have an advantage, but earlier-stage organizations should still explore agentic AI because practical opportunities exist. The key is to start with real business problems and manage risk through oversight and governance.