The race to deploy AI agents is accelerating, but enterprise leaders are learning that speed must be tempered with discipline. At a recent conference hosted by Section, a consultancy led by NYU professor Scott Galloway, two top IT executives shared their experiences: Scott Likens, global chief AI engineer at PwC, and Lasherelle Morgan, senior vice president of AI innovation and acceleration for NBCUniversal. Their advice boils down to four core principles that blend agility with caution.
1. The human is the loop
Likens challenged the common phrase "human in the loop" by asserting that the human is the loop. AI agents should never operate autonomously without human direction, especially in high-stakes environments. Morgan echoed this: "Start with the pain point," she said. "Don't just bring in an AI tool. Ask, 'what are you struggling with?' 'What are you spending five hours of your day on?'" This means starting from the end user and working backward to choose the right tool, ensuring that humans remain central to the decision-making process.
Experts across industries note that AI agents excel at automating repetitive tasks, but they lack context and judgment. For instance, in customer service, an AI agent might handle routine inquiries but escalate complex issues to humans. In finance, an agent could flag anomalies but require a human to approve transactions. The key is to design workflows where humans retain authority over critical actions.
2. Experimentation is important
Likens urged business leaders to embrace rapid experimentation. PwC runs AI-driven experiments in one- or five-day cycles, getting quick feedback. "All this talk of tokens just started a couple of months ago, and now all of a sudden there is a cost focus with AI. That's the wrong way to look at it," he said. Experimentation helps identify use cases that deliver transformative value beyond mere cost savings. However, this requires a mindset shift, especially among mid-level managers accustomed to longer cycles.
Morgan added that experimentation should be targeted: "Start with easily repeatable processes and data. Start with what people hate doing." This approach reduces risk and builds momentum. For example, an internal HR agent that schedules meetings or answers policy questions can quickly demonstrate value. Once successful, similar agents can be deployed in other departments.
3. Blow up a bad process
AI can amplify the flaws in existing workflows. Morgan warned: "You have to have clean data, and a workflow that is clean from start to finish. You need to literally get a pen and paper and write out the process." AI is "really good at blowing up a bad process," she said. Before introducing AI, organizations must map their processes, identify data owners, and ensure data quality.
Likens shared that PwC addressed data issues years before AI became mainstream. The company built a data foundation for regulated areas like accounting and auditing. They tackled the challenge of "tacit knowledge" — the know-how that resides in employees' heads — by extracting it through telemetry and agent behavior analysis. "We focus on architecture first, so it can scale for our people," Likens said. This ensures trust and safety when using AI tools.
4. Governance and guardrails
Not all AI agents carry the same risk. Morgan introduced the concept of "blast radius." For low-risk tasks like scheduling lunch, no human oversight is needed. But for agents that send messages to consumers, strict guardrails are essential. NBCUniversal uses intake forms to track and measure potential impact, scaling governance accordingly.
At PwC, AI responsibility is centralized among a small group of deep AI engineers who set standards and build a trusted "chassis." Then a larger group of hands-on builders across the business tailors solutions for clients. Likens noted that only 1% of the organization is deep AI engineers, while 10% are distributed builders. This structure balances control with flexibility.
Morgan emphasized that governance should be risk-based: "If it's low risk, we don't need a human in the loop. If it's high risk, we have multiple checkpoints." This pragmatic approach allows rapid deployment of low-risk agents while ensuring high-risk ones are carefully managed.
Historical context and broader implications
The debate between speed and caution is not new in technology. During the early days of cloud computing, enterprises faced similar tensions between agility and security. Today, AI agents represent a new frontier. According to a recent survey by Gartner, 96% of IT professionals now use AI, with agentic applications among the top use cases. However, implementation roadblocks remain, particularly around data quality, governance, and talent.
Scott Likens joined PwC in 2020 after a career that included leadership roles at IBM and Deloitte. He holds degrees in computer science and engineering and has authored multiple papers on applied AI. Lasherelle Morgan rose through NBCUniversal's technology ranks, overseeing digital transformation before taking on AI innovation. Her team has deployed agents for content personalization, ad operations, and internal workflows.
Both leaders agree that agentic AI will redefine business processes over the next five years. But the path forward requires a deliberate blend of speed and caution. As Morgan put it: "Meet people where they are. Start with what they hate doing. Then scale." Likens added that the frozen middle — managers resistant to change — is the biggest hurdle. "It's a human challenge," he said.
The insights from these executives provide a blueprint for organizations eager to harness AI agents without losing control. By keeping humans at the center, experimenting rapidly, cleaning up data and processes, and implementing risk-based governance, companies can move fast while staying safe. The future of work will be hybrid, with humans and AI agents collaborating — but only if leaders set the right foundations today.
Source:ZDNET News