Recommended Free Tools
AI agents should earn authority in stages, not receive it all at launch. In a Unite.AI interview published October 6, 2026, Cyara CEO Sushil Kumar argues that companies should give each agent a written role, test its competence, supervise its real-world work, and expand its permissions only when it meets defined gates. His central analogy is workforce management: an agent needs a job description, clear limits, an accountable owner, and evidence before promotion.
Who is Sushil Kumar?
Kumar became Cyara’s CEO in December 2025. Cyara’s leadership page lists him in the role and describes more than 25 years of experience across AI, DevOps, cloud infrastructure, product strategy, and software testing. Before Cyara, he co-founded and led RelicX.ai, which was acquired by Harness, and held roles at Broadcom, CA Technologies, and Oracle.
In its December 8, 2025 appointment announcement, Cyara said Kumar would focus on AI-powered customer-experience assurance, partnerships across CCaaS, CPaaS, UCaaS, and AI, and global expansion. The company describes its platform as testing, validating, and monitoring customer journeys across voice, digital, messaging, and conversational AI.
Why does Kumar compare AI agents to employees?
Kumar’s point is that an AI agent’s output can look plausible even when it is wrong. Traditional deterministic software can often be checked against expected results; an agent may make a convincing but mistaken claim without producing an obvious error signal. That makes a successful demo or technically valid response insufficient evidence that the agent is ready to represent a company.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitches#1 Best Overall
His proposed analogy is an employee: assign a defined role, establish what information and authority that role includes, assess performance, supervise work, and add responsibility gradually. He sums up the approach this way: “My read is that autonomy is not a deployment decision. It is a series of promotions.” This is Kumar’s operating perspective from the interview, rather than a claimed universal standard.
What should an AI agent’s role define?
Before an agent interacts with customers or critical systems, Kumar recommends writing down the job it is expected to do. The description should cover:
- Purpose: the customer or business outcome the agent is meant to achieve.
- Authoritative information: which knowledge sources it should rely on when responding.
- Customer data: what information it may access and use.
- Decisions: which choices it can make without approval.
- Responsibility boundary: where its role ends and a person or another process must take over.
Kumar’s test is deliberately practical: “If a company cannot write that down in a paragraph, the agent is not ready for a role. It is ready for a demo.” A written role gives teams something concrete to test and gives employees a basis for recognizing when a request exceeds the agent’s remit.
How should companies set permissions and handoff rules?
A job description alone does not determine what an agent can do in connected systems. Kumar recommends spelling out permissions separately and distinguishing access to information from authority to change it.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →- Read access: identify which systems and records the agent may consult.
- Write access: specify which fields, records, or systems it may alter, if any.
- Commitments: define what the agent may promise on the company’s behalf, such as a resolution, action, or follow-up.
- Handoff triggers: identify situations that require a human, including decisions outside the role or uncertainty the agent is not authorized to resolve.
The degree of proof should rise with the agent’s authority and the potential impact of a mistake. Reading information to answer a routine question is different from changing an account or making a consequential commitment. The organization should set gates appropriate to those differences rather than treating every agent action as equally risky.
What evidence should be required before launch and at scale?
Kumar calls for testing under conditions resembling actual use before deployment, followed by oversight of what the agent does in production. The evidence should address whether it achieves the intended customer or business outcome—not just whether it generates a response that passes a technical check.
He also cautions against treating a high evaluation score as proof that an agent is safe to scale. In the interview, he uses a hypothetical 99% score across a million conversations to illustrate how a seemingly strong percentage can still leave a substantial number of failures. It is an example, not a reported performance result. His summary is: “A pilot can run on an organization’s conviction. Scale requires evidence.”
For voice systems, Kumar highlights another complication: speech-recognition errors can distort downstream evaluation. If the system mishears a customer’s question, the agent may respond to the wrong request; a review of the agent’s answer alone can miss that upstream failure. He presents this as a concern from his experience and interview perspective.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →How should autonomy increase after launch?
Kumar’s staged approach treats expanded authority like a promotion. Organizations should define in advance what an agent must demonstrate before it gets a new permission or handles a more consequential task. Evidence from real use can inform those decisions, but the agent’s business owner remains responsible for setting the gates and deciding whether the evidence is sufficient.
- Start with a bounded role. Limit the agent to a defined task, approved information, and low-impact actions.
- Set promotion gates. Decide what evidence and performance conditions are needed before granting additional access or decision-making authority.
- Monitor production work. Review the agent’s actions and outcomes once real interactions begin, not only its pre-launch test results.
- Turn failures into release tests. When a production problem occurs, add a test that requires the next release to demonstrate that the same issue has been addressed.
- Name an accountable business owner. Assign a person responsible for the agent’s role, oversight, and authority changes.
This framework does not prescribe one universal score or approval threshold. The thresholds depend on the role, impact, and authority being considered; Kumar’s emphasis is on having explicit gates and evidence before expanding autonomy.
What does the interview establish about Cyara?
Cyara positions its platform as customer-experience assurance software for testing, validating, and monitoring customer journeys across channels. In its December 2025 CEO announcement, the company reported that it supported “more than 350 million customer journeys each year” and had validated “over a million AI-generated responses.” Those are figures reported by Cyara in its announcement, not independent measurements. The same release said Cyara had customers in more than 135 countries.
The interview is useful as Kumar’s leadership perspective on managing customer-facing agents; it is not an independent assessment of Cyara’s product or a comparative test of AI governance platforms.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
