The June 2026 edition of the Techeconomy Business Series went live on Techeconomy TV with a simple goal: to cut through the noise around Agentic AI automation and talk honestly about what is actually happening inside companies trying to adopt Agentic AI.
The theme, “Agentic AI and the Future of Work” faced by a panel of four who see this shift from very different angles: Oluwatomi Obinna Alagbe, a Staff Software Engineer at Malwarebytes; Peter Ndukwo, a Web3 security researcher and ZK contributor at Zipel Labs; Abisola Rachael Aderohunmu, Product Manager Lead at Heala Tech; and Francis Udogu, a digital marketing manager and growth expert. These four powerhouses made a case that using Agentic AI is not as simple as most executives seem to think.
Oluwatomi Obinna Alagbe said: “I haven’t written code in six months”.
This line set the tone early. Oluwatomi is a software engineer by title. His day-to-day work has shifted almost entirely away from writing syntax and toward something less glamorous: managing Agentic AI agents directing Agentic AI agents and checking the output of Agentic AI agents that now do the writing for him. His bigger concern though was about how leadership teams think about Agentic AI tools.
Traditional software is same input, same output every time. Agentic AI agents do not work that way. Feed Agentic AI agents the same prompt twice and you can get two different answers. Because of that unpredictability and because Agentic AI models still make mistakes more than anyone would.
Oluwatomi argued that the real bottleneck in engineering now is not how fast you can code. It is how well you can engineer context: building tight memory boundaries and clean data inputs so the Agentic AI agent stays anchored to what is actually true about your business.
Abisola Rachael Aderohunmu said: “AI doesn’t fix a broken process”.
She picked up that thread. Pushed it further stating she has watched companies rush to deploy Agentic AI out of fear of missing out dropping flashy Agentic AI models into workflows that were already broken.
Her point was blunt: if your internal knowledge base is outdated or inconsistent an Agentic AI agent will not fix that. It will just help you deliver answers to customers faster.
For her the real shift is not about how a task gets done – It is that the value of a professional today is less about execution and more about judgment. Knowing what to automate with Agentic AI what to double-check and what still needs a human call.
This is where the panel got interesting, particularly as Francis Udogu approached Agentic AI agents from the growth side. As tools that could churn out and distribute content across platforms with minimal human input purely to cut costs and save time.
Oluwatomi pushed back on that framing not because it is wrong. Because unsupervised Agentic AI agents with too much access are exactly the kind of thing attackers go looking for.
To square that circle he laid out three things he thinks every company needs before letting Agentic AI agents run loose:
- Least privilege, actually enforced
Most companies over-permission their Agentic AI agents by default giving them access to file systems, databases and backend infrastructure they do not need. Agentic AI agents should ever touch what their specific task requires. Nothing more.
- A human still has to sign off
Anything stakes. Deleting data, changing account settings modifying live systems. Should require a person to approve it. No exceptions, no way around it.
- Test Agentic AI agents like you would test anything that can fail.
Prompts decay Agentic AI models get. Start behaving differently. Oluwatomi’s advice was to run Agentic AI agents against synthetic edge cases so you catch behavioral drift before a customer does.
He closed with a warning that’s easy to miss: AI-specific denial-of-service attacks. Instead of flooding a server with traffic, bad actors are now sending deliberately complex, recursive queries designed to eat up an agent’s processing time, which quietly drives up compute costs while doing real damage to infrastructure.
His overall message was less about hype and more, about discipline: Agentic AI is not a plug-and-play upgrade. It is a system that needs an architect behind it. Someone thinking about boundaries, failure modes and who is accountable when something goes wrong with Agentic AI.
