Elena Verna published a sharp observation this month, and it lands harder than most AI strategy commentary because it does not talk about tools at all. Her line: “Agents don’t have agency. They wait to be told what to do.”
Set that line next to how most companies are running their AI transformation right now. They are buying agents by the dozen. They skip the one thing that makes any of them useful: a system where people are trusted to act.
The failure is organizational design, and underneath that, it is trust.
The Control Systems Already Tell the Truth
Look at what a typical large company does before any agent shows up. Information is gated by title, so the people closest to the work lose access to the data they need to decide. Approval cycles run through three layers, so any action waits days for a signature. Employees are treated as risk vectors to constrain, rather than operators to enable.
Verna names the result precisely: “All of this pretty much screams at employees: We don’t trust you, stay in your lane.”
That is the diagnosis the AI conversation keeps skipping. A company that has spent a decade engineering its people into narrow lanes has built a machine for withholding agency. Bolting autonomous agents onto that machine keeps autonomy out of reach. The machine was designed to prevent it.
So the agents arrive, and they behave exactly like the org chart taught everyone to behave. They wait. They escalate. They ask permission. They produce output that sits idle, waiting for someone authorized to act on it. The transformation stalls, and leadership blames the model.
Agents Inherit the Trust You Already Have
Here is the part that should reframe the whole spend. Agents act as an amplifier of whatever trust already exists in the organization.
If your people are allowed to see the full picture and act on it, an agent slots into that flow and extends their reach. If your people are gated, throttled, and second-guessed, an agent inherits the same constraints, because the same humans are still the only ones permitted to approve anything. The bottleneck stays exactly where it was. You just added a faster thing in front of a slow gate.
Verna points to companies that built differently. Anthropic staffs flat across technical and non-technical work, so decisions sit close to the people doing the work rather than climbing a hierarchy. Lovable assigns each agent a domain-expert “parent,” a human accountable for the agent’s behavior and empowered to direct it. Notice what both have in common. Both are trust architectures rather than tooling choices. They decide who is allowed to act, and they push that authority outward instead of hoarding it at the top.
That is the unlock: a structure where acting is permitted, more than a better agent.
Governance Is Control as Enablement
This is where the word “governance” usually goes wrong. Most leaders hear governance and picture a brake: more review, more approval, more lanes painted on the floor. That instinct is exactly the machine Verna is describing, and it is the reason both human agency and agent autonomy die in these companies.
Governance done right is the opposite: control as enablement. The job is to decide, deliberately, where trust can be extended safely so that people and agents can act without waiting.
Three moves make this concrete.
Push decision rights to where the work happens. For each recurring decision, name the closest competent person and give them authority to act inside clear boundaries. The boundary is the governance. The authority is the enablement. A boundary without authority is just a cage; authority without a boundary is chaos. You need both, and most companies have built only the first half.
Make the boundary explicit, then trust inside it. An agent with a clear domain, a named human owner, and defined limits is safe to run autonomously inside those limits. A human with the same clarity is safe to act without a third signature. The discipline is in drawing the boundary well and trusting whoever operates inside it. When you trust inside a well-drawn boundary, speed returns.
Treat surveillance as a tax rather than a strategy. Every control that exists to catch a rare bad actor charges every good operator a daily fee in latency and discouragement. Some controls are worth that fee. Most are inherited, carried forward without re-justification, and quietly screaming the distrust Verna described. Audit them. Kill the ones that cost more agency than they protect.
The Same Scoreboard for People and Agents
The deepest implication is that governing agents and people demands the same logic. The trust you extend to a human operator and the autonomy you grant an agent are the same design decision, measured on the same scoreboard: how close to the work does authority to act actually sit?
A company that answers “far away” will fail at both. Its people will stay in their lanes, and its agents will wait to be told what to do, because the surrounding system was built to make both things happen. A company that answers “close” gets compounding returns. People act, agents extend that action, and the organization moves at the speed of trust rather than the speed of approval.
Verna’s title is the strategy in four words. What your company needs is agency, and agency is something an organization either grants or withholds long before any software is involved.
Do This Now
Pick one decision your team makes every week that currently waits for an approval two levels up. Ask why that approval exists. If the honest answer is “we still hesitate to trust the person closest to it,” you have found the real blocker, and it runs deeper than any agent can solve. Move the decision down, draw the boundary, and watch how fast both your people and your agents start to act.
Trust is the load-bearing structure of every AI transformation that actually works.
This analysis synthesizes Your Company Needs Agency, Not Agents (Elena Verna, June 2026).
Victorino Group helps leaders redesign control systems so trust enables both people and agents. Let’s talk.
All articles on The Thinking Wire are written with the assistance of Anthropic's Opus LLM. Each piece goes through multi-agent research to verify facts and surface contradictions, followed by human review and approval before publication. If you find any inaccurate information or wish to contact our editorial team, please reach out at editorial@victorinollc.com . About The Thinking Wire →
If this resonates, let's talk
We help companies implement AI without losing control.
Schedule a Conversation