Agentic AI is here. Now, CIOs must orchestrate

Agentic AI is about to change how companies create value. Morgan Stanley estimates AI could add $13 to $16 trillion in value to the S&P 500, with nearly $500 billion estimated coming from agentic AI alone. Yet, most enterprises aren’t ready. According to new research from Asana’s Work Innovation Lab, only 20% of organizations have successfully scaled AI agents, while nearly a third don’t even have a plan. Workers, meanwhile, already expect to delegate 43% of their workload to AI agents within three years.

The problem isn’t the technology — it’s the planning and execution. Too many pilots stall out because CIOs haven’t built the systems, guardrails and culture to move beyond experiments. The expectation has shifted: We’re no longer asked to “try AI.” We’re responsible for scaling it safely and showing clear business value.

Here’s my practical playbook for making that shift.

From individual tools to collaborative agents

We’ve entered the era of human-agent collaboration, where multiple AI agents can coordinate across entire teams and functions. This is a fundamental shift from the “one person, one AI tool” model. Yet most organizations remain stuck in pilot mode, where agents are limited to narrow tasks like summarizing notes or drafting emails.

The real opportunity lies in designing systems where agents don’t just assist individuals but work collectively — sharing information across teams, coordinating dependencies and producing results that leaders can trust.

Imagine one agent serving the needs of multiple stakeholders at once, rather than dozens of isolated assistants duplicating effort, and learning on the run from human feedback.

So how do we get there? Our research shows the gap comes down to structure. In organizations that provide formal training, nearly 80% of employees feel confident using AI, compared to just 40% where training is absent. And governance matters just as much: Only 19% of workers say their company has clear rules for when to rely on AI versus people. The takeaway is to treat AI like infrastructure, not one-off experiments.

What CIOs should do:

• Start with outcomes, not governance. The first question to ask isn’t about policies or structures — it’s “Where can AI help us solve our hardest business problems?” CIOs who anchor AI in business outcomes create a clear mandate for adoption and ensure governance serves strategy, not the other way around.
• Build the structure around it. Once the priorities are clear, create the guardrails to scale responsibly. That often means a cross-functional AI council that brings IT, legal, security, HR and business leaders into one room. Assign owners for each agent — like product owners — who are accountable for safety, updates and results. This stops shadow projects before they spread and gives leaders visibility into experiments while they’re still manageable.
• Empower people with both rules and tools. Governance alone won’t change how work gets done. Define what agents can and can’t do, but also give employees the training and freedom to use AI on their most repetitive work and their hardest problems. Encourage shared learning across teams so successful use cases compound. This grassroots innovation, paired with a top-down strategy, is what takes AI beyond scattered pilots and makes it part of how the business runs.

You can’t automate chaos. Redesign workflows first

Too many companies bolt AI onto messy processes and expect transformation. The real opportunity is to reimagine workflows with AI as a collaborator from the start. That means asking: “What should this process look like if agents are helping us run it?”

Start with high-volume, low-risk processes that already have structure — like IT ticket triage, expense approvals or marketing campaign intake. Redesign them so they work smoothly with humans and agents together. Otherwise, AI just scales the chaos faster than people ever could.

Steps to take:

• Redefine workflows for AI-human collaboration. Map ownership, inputs and outputs and escalation paths — not just to tidy up, but to decide where agents add the most value and where humans need to stay in control.
• Build an approval and deployment process. Require intake forms, sandboxing, security reviews, privacy checks and audit logs. Keep a registry of all models and agents with clear service levels and troubleshooting guides.
• Create a knowledge base. Publish a central library of available agents with usage instructions and real-world examples.
• Find champions. Identify early adopters to co-design new workflows and teach others best practices.

Done right, even simple rule-based agents become the foundation for higher-autonomy systems — because the workflows they’re embedded in were designed for them, not just retrofitted after the fact.

Decision-making is shifting (fast) — govern it

Agents are moving from just executing tasks to supporting decisions — and increasingly influencing them. CIOs need to decide: Where should agents have autonomy and what oversight is required?

A tiered approach works best:

• Simulate: Agents suggest actions, but humans decide (e.g., recommended contract changes).
• Shadow: Agents act in parallel and humans verify (e.g., budget allocation suggestions).
• Partial automation: Low-risk autonomy with human escalation (e.g., approving expenses under $500).
• Supervised autonomy: Higher autonomy with regular human checkpoints (e.g., routing and resolving tier 1 support tickets).

This framework clarifies your risk tolerance while meeting compliance and audit requirements. As CIO, you own the matrix of what’s delegated, what’s supervised and what gets logged.

Make ‘agent manager’ a core workforce skill

AI changes jobs more than it eliminates them. The key skill now is managing agents: directing their work, checking their output and improving their instructions.

Our research shows workers are cautiously optimistic — 52% think AI will help their work, but 29% worry it might replace them. CIOs have a critical role to play in helping employees see AI as a teammate, not a threat — through guidance, training and clear examples of where it adds value.

A practical approach:

• Role-based training tracks: AI knowledge isn’t one-size-fits-all. Builders (often IT or operations) need the technical depth to create and maintain agents. Reviewers (risk, governance and quality teams) need to know how to evaluate and approve outputs. Operators (daily business users) need practical guidance on how to work with agents and provide feedback. Structuring training this way avoids generic “AI 101” sessions and ensures people learn what’s directly relevant to their role.
• Make the “agent manager” role explicit: Don’t treat agent supervision as informal or extra work. Update job descriptions, performance metrics and incentives so employees know that supervising, correcting and improving agent workflows is part of their core job.
• Partner with HR on sentiment: Training only works if employees feel supported. Work with HR to regularly check trust levels, readiness and adoption feelings. Catching concerns about job security or workload early lets you adjust training and communication before skepticism turns into resistance.

This approach doesn’t just drive adoption — it builds resilience in a mixed human-AI workforce where employees see themselves as partners with agents, not competitors.

The new ROI: Return on intelligence

Traditional ROI metrics — efficiency and cost — don’t capture AI’s full impact. CIOs should measure return on intelligence: the combined improvement in decision quality, throughput and innovation when humans and agents work as one system.

Key metrics to track:

• Efficiency: cycle time, cost-to-serve, throughput per employee
• Decision quality: error rates, rework avoided, approval lead times
• Innovation: time-to-insight, experiments launched, feature velocity
• Trust & adoption: weekly active users, human-in-the-loop acceptance rate, audit pass rates
• Resilience: incident response times, rollback speed, change failure rates

Build these metrics into your deployment from day one. The goal isn’t “more AI” — it’s better outcomes and more engaged teams.

The CIO playbook: 4 moves to start this quarter

1. Anchor AI in business outcomes. Begin by asking: “Where can AI help us solve our hardest business problems?” This frames adoption around value, not technology.
2. Form a cross-functional AI council. Once priorities are clear, bring IT, legal, HR, risk and business leaders together to set risk levels, decision rights and intake processes. Assign owners accountable for each agent’s performance.
3. Pilot rule-based agents in key workflows. Choose three high-volume processes and move deliberately through the path from simulate to shadow, then partial automation and supervised autonomy. This builds trust and a path to higher autonomy without surprises.
4. Track ROI on two levels. Measure efficiency gains, but also human outcomes like engagement and burnout. Pair this with role-based training so employees have the skills and confidence to make AI part of their daily work.

Scale systems, not just tools

The companies that scale responsibly now won’t just get more efficient — they’ll set the tone for how employees trust and adopt AI over the next decade. For CIOs, this is bigger than deploying technology: it’s shaping the future of work itself.

Do this right and you won’t just modernize your company. You’ll show what intelligent, resilient, human-centered organizations look like in the age of AI agents.

This article is published as part of the Foundry Expert Contributor Network.
Want to join?