After a year of attending over 12 conferences, I sensed a bias among AI leaders in their AI agent deployment strategies. Some are firmly in the “human augmentation” camp, seeing AI agents more as partners and collaborators. Others suggest an “agentic” future where agents are largely autonomous, and the human-in-the-loop may be necessary only as a transitional step to validate accuracy and build trust with human stakeholders.

Human Augmentation or Agentic? 12 Questions Before Giving AI Agents Autonomy

We discussed the AI agent strategy and debated human augmentation versus agentic at a recent Coffee With Digital Trailblazers, a LinkedIn Live event I’ve hosted for four years and runs live on Fridays at 11 am ET. It was one of the top episodes of 2026, ending with a framework of examples, qualifiers, and deployment factors when deciding an AI agent’s strategy.

AI Agent Strategy: Human Augmentation vs In-The-Middle, vs Agentic?

Biases aside, the pragmatic strategy is to determine each AI agent’s strategy depending on its goals and risks. “The question isn’t how autonomous you can make an agent, it’s how much autonomy we trust it with,” suggested Joanne Friedman.

“After I developed enough trust in the agent and the format was coming out rather nicely, I turned it loose,” Joe Puglisi said about an agent he developed.

Martin Davis suggested, “Moving from augmentation through to human in the middle through to full agentic is a progression that helps build trust. But I think the mistrust of something people don’t understand is going to be very high.”

Listen to the episode and review the research slide and the framework. While there’s a lot to digest from our discussion, I’m not ready to move on from this topic just yet.

For today’s article, I asked experts 12 questions to continue the discussion and flesh out a direction AI leaders can apply in their AI strategies.

When is human augmentation the only AI agent strategy?

I said during the Coffee With Digital Trailblazer episode, “If you have an agent that’s creating intellectual property for you, it’s probably human augmentation.”

I use AI to suggest titles, rework heads, and improve sentence structure, but not to write my content. IMHO, creating intellectual property should always be human augmentation, whether you’re writing, creating social media videos, or developing code. It’s one of the principles I lay out in my upcoming book, AI Reinvention.

Other reasons many agents will remain in the human augmentation category come down to decision risks and the cost of failure. Nearly half of leaders (49%) have defined high-risk use cases where they do not allow autonomous decision-making by agents, according to KPMG’s Q3 2026 AI Quarterly Pulse Survey.

Experts weighed in on the types of agents, why human-augmentation is a viable strategy, and guardrails for this type of AI agent. Also see my writing on seven ways to encourage critical thinking in AI users and developing AI’s context layer.

1. Which types of AI agents should always be designed for human augmentation?

“Agents should remain focused on human augmentation when they influence high-stakes decisions involving customers, financial outcomes, compliance, or brand trust. CIOs should consider agentic automation for work that is structured, repeatable, and governed by clear policies, with humans monitoring outcomes, exceptions, and drift during a defined transition period, while IT teams develop operating models that can support pilots, governed agents, and human-augmenting systems at the same time.” – David Vidoni, chief information officer at Pega.

2. Why should AI agents augment people rather than act on their own?

“AI can surface insights, model scenarios and accelerate decisions, but at this stage of AI adoption, accountability cannot be shifted to the AI agent when something goes wrong. Responsibility ultimately remains with the people and organizations using it, which is why human oversight and review are essential when a decision can materially affect the business or its financial outcomes.” – Mike Zack, general manager at Acterys at Vena Solutions.

3. What guardrails should human-augmenting AI agents have?

“AI agents that make consequential decisions, impact people, create regulatory exposure, or carry business risk should augment human decision-making rather than replace it. Agentic AI is powerful technology, and we shouldn’t mess around with it. You need clear rules around what an agent can access and do, with human oversight for higher-risk actions and the ability to pause or override an agent. This is hard work and needs to be in the hands of people who know what they’re doing. Ultimately, accountability stays with people — not the AI.” – Jamie Rutledge, president at Kyndryl U.S.

StarCIO AI Strategy and Governance Workshop

How should AI agents “earn” their autonomy?

When developers tell me a new application is ready for production, I ask them to work with their business, operations, and security stakeholders to draft their release-ready criteria.

There are reasons to crawl and walk before letting an AI agent run on its own. In formal AI assurance reviews, 64% significantly modified a quarter or more of their AI systems, and the top issues found were data quality problems (57%), AI model drift (48%), and shadow AI (39%), according to EY’s US AI Risk and Governance Survey.

Experts answered what “agentic” means, agentic decision criteria, and operational considerations.

4. What makes an AI agent “agentic,” and how should its autonomy scale with risk?

“An agent is “agentic” when it can decide how to pursue an outcome, rather than only follow a fixed workflow. Humans should monitor decision quality, policy and security compliance, exceptions, and consequential actions, and higher-risk actions should require approval. Autonomy should scale with risk, confidence, and reversibility: low-risk, reversible tasks may be automated; higher-risk actions require approval, monitoring, auditability, and accountable owners.” – Edgar Kussberg, director of product management at Sonar.

5. What decision criteria should be reviewed to determine whether to grant autonomy?

“The decision to give an AI agent autonomy comes down to the cost of failure, the reversibility of its actions, the sensitivity of the data involved, and how well the surrounding software can monitor and gauge its behavior. Until organizations can truly understand and validate what an agent is doing, major decisions should retain meaningful human oversight. The transition to autonomy needs to be earned through evidence. That means strong evaluations, clear escalation paths, auditability, and guardrails that exist above the model layer in infrastructure the organization controls.” – Sid Vivek, chief technology officer at zeb.

6. How should CIOs and CISOs phase in agent autonomy?

“AI autonomy should be a spectrum, not an all-or-nothing decision. Agents making high-impact decisions, handling sensitive data, or taking hard-to-reverse actions should support human judgment, not replace it. Oversight should scale with risk. Security teams are process-driven for a reason; adopting agents shouldn’t mean abandoning workflows that work. CIOs and CISOs should introduce agents in steps, validate their behavior, and expand autonomy as confidence grows.” – Brent Midwood, senior director of product, security operations at Tanium.

7. Which steps in a workflow should an agent take over first?

“An agent earns full autonomy initially only for actions that are cheap to detect and cheap to undo. For anything else, we run a deliberately sequenced transition: aggressive on detection and diagnosis, a human gate on remediation, until the agent’s behavior holds up under real production load, not a benchmark.” – Barr Moses, co-founder and CEO at Monte Carlo.

8. What should humans monitor for while an agent is autonomous?

“I ask three questions before an agent runs on its own: are its actions reversible, is the volume too high for people to review every case, and are outcomes measurable so you know quickly when it’s wrong? During the transition, people shouldn’t review every output. They should watch the exceptions: overrides, escalations, and cases where the agent was confident and still wrong. The transition should end on evidence, not a calendar date. Run it through a full business cycle, including peak periods, until the override rate is one you’d accept from a trained employee. For most workflows, that’s a couple of quarters.” – Rajeev Jaswal, co-founder and CEO at Argo Intelligence.

9. Who has to be ready before an agent gets more autonomy?

“The criteria is up to the team using it, and the transition period serves two purposes: training the AI agent on the organization’s specific business processes and building the human team’s confidence in what it will do. It’s a two-way learning process, and greater autonomy should come only when both the agent’s performance and the team’s readiness support it.” – Caitlin Leksana, co-founder and CEO at Fazeshift.

10. When should an agent graduate to autonomy, and how should it be governed after?

“Run each agent alongside your team for at least a full business cycle and graduate it only when it beats the human error rate on real cases. After that, govern pilots and autonomous agents the same way: their own identity, least-privilege access, a rules layer they can’t bypass, a full audit trail and a kill switch.” – Akhilesh Agarwal, president and chief innovation officer at apexanalytix.

What should AI leaders prepare before running AI agents at scale

Only 21% of CIOs have agentic AI governance today, but 71% will deploy agentic AI within 12 months, according to Gartner.

Does that metric scare anyone? It should, and it’s one reason I’ve been writing about C-level responsibilities and assigning AI agent owner accountability as we scale to thousands of AI agents in production.

Here’s what IDC predicts as a key scaling factor. By 2028, 70% of G2000 enterprises will be able to continuously verify their AI agents’ decisions, up from 14% today, and those without kill switches on production agents will see 25% higher liability insurance premiums, according to IDC FutureScape 2027.

Experts weigh in on the operational and change-management considerations as organizations scale to thousands of AI agents, including human augmentation, human-in-the-loop, and agentic ones.

11. How should CIOs and CISOs manage many AI agents running at once?

“Most enterprises are running a variety of agents, and each holds its own production credentials, sends data to its own LLM, and keeps its own memory. When an incident involves multiple agents, engineers get conflicting answers and must reconcile them manually. To manage a variety of agents at scale, CIOs and CISOs should prepare by putting one operations layer between all agents and production. This ensures access is read-only by default and audited in one place, IT can pick the approved model, and all previous incident investigations are shared across teams. Every agent cites its evidence, so pilots, transitional agents, and augmenting agents are judged on the same record.” – Francois Martel, field CTO at NeuBird.

12. Why do AI agent initiatives fail to deliver, and what process work do leaders skip?

“Most AI initiatives fail to deliver because people skip the process work on both ends. On the front end, you have to re-engineer the process first—bolt AI onto a workflow that doesn’t need to exist, and you’ve just automated the wrong thing, faster. On the back end, that’s change management: implementing AI means someone has to do their job differently tomorrow than they did yesterday, and if you don’t bring them along and build that new way of working into their actual flow, you’ve wasted the investment.” – Ashleigh McCord, SVP, enterprise technology services at Workday.

For my POV on operational considerations, see my articles on seven safeguards for observable AI agents and five key agenticops practices to start building now. Regarding change management, see my article on preparing the organization for an agentic future of work.

Published on:

Leave a Reply


StarCIO

My company, StarCIO, provides leadership, learning, and advisory programs for companies looking to accelerate delivering business value from digital transformation. Contact me if you’d like to learn more about partnering opportunities.


Isaac Sacolick

Greetings from Isaac Sacolick

If you enjoy my thought leadership, please sign up for the Driving Digital Newsletter and read all about my transformation stories in Digital Trailblazer.


Coffee with Digital Trailblazers hosted by Isaac Sacolick

Digital Trailblazers! Join us Fridays at 11am ET for a live audio discussion on digital transformation topics:  innovation, product management, agile, DevOps, data governance, and more!


Join the Community of StarCIO Digital Trailblazers

Drive: Agile, Innovation, and Transformation has 800+ articles on AI and digital transformation for leaders. Written by StarCIO Digital Trailblazer Isaac Sacolick. Learn more.


Coffee with Digital Trailblazers

StarCIO Coffee With Digital Trailblazers

Coffee With Digital Trailblazers is a weekly event for Digital Transformation leaders where we discuss leadership, delivering value from AI/data, and driving innovation.


About the StarCIO Digital Trailblazer Community

StarCIO Digital Trailblazer Community

Revolutionizing traditional learning, networking, and advising experiences.

Visit the community


About StarCIO

StarCIO: AI is reshaping business. Develop the leaders who can drive the transformation.

About Isaac Sacolick

Isaac Sacolick

Author, 1,200+ articles, keynote speaker, Chief StarCIO Digital Trailblazer. Full bio


Driving Digital Newsletter

Driving Digital Newsletter

StarCIO Guides

Learn StarCIO Agile Planning

Digital Trailblazer

Digital Trailblazer by Isaac Sacolick

Driving Digital

Driving Digital by Isaac Sacolick

Driving Digital Standup

Driving Digital Standup

Recognition

reworked imapct awards 2026 Judge
InfoWorld 2025 Judge
InfoWorld Technology of the Year 2024 Judge
Thinkers360 Top 10 in IT Leadership
Thinkers360 Top Agile Thought Leader
Thinkers360 Top DevOps Leader
Thinkers360 Top in Digital Transfomation
Thinkers360 Top in Analytics
Thinkers360 Top in Product Management