Vicki Reyzelman, senior solutions engineer at Akamai, used a single metric on the This Week in AI podcast to illustrate a broader trend: there are roughly 144 AI agents for every person on the internet. That ratio helps frame how the rapid spread of agents is reshaping cybersecurity, infrastructure, education and AI governance.
Security operations must match agent speed
Automated agents and their management systems can operate continuously and respond in seconds. Reyzelman cited reports of attackers moving within minutes and vulnerabilities being exploited soon after public disclosure. She also described an incident involving one of her customers where an attacker returned, changed tactics, and tried again.
Traditional security workflows — investigating alerts, understanding a vulnerability, deploying a patch and monitoring the outcome — assume there is time for human response. As automated reconnaissance and adaptive attacks accelerate, that assumption becomes less reliable. Reyzelman argued for multiple defensive layers across APIs, applications and networks so that a single missed signal does not become a catastrophic failure.
This Week in AI has tracked agent security issues previously, and the current conversation focuses on how enterprise security must change around more autonomous software. That shift increases reliance on automated defenses, stricter permissioning, and monitoring systems capable of constraining machine activity at comparable speed.
AI capacity depends on physical infrastructure
AI capacity requires more than models: it needs electricity, cooling, water, data center space and the supply infrastructure that provides them. Reyzelman linked major hyperscaler investments with projections for sharply higher data center energy and water use by 2030.
An audience comment illustrated the point: a university data center that can reuse waste heat in colder months must shed that heat during warmer weather. That example shows how environmental and climate factors limit deployment choices.
These constraints directly affect decisions. Organizations must weigh power availability, cooling systems, water access, latency, security and local infrastructure capacity alongside model performance and cost.
Government policy already shapes those choices: Reyzelman paired growing investment in AI infrastructure with increasing regulatory requirements in Europe. AI infrastructure now spans engineering, economics, compliance and public policy, so deployment decisions increasingly involve multiple systems at once.
Human judgment remains essential when AI acts physically
Rapid AI adoption increases the value of foundational knowledge. Reyzelman discussed AI use in education and research: students may gain easier access to explanations and answers, but those who lack subject understanding may be unable to spot incorrect results. The same issue appears in scientific work, where reliable AI output still depends on reliable data and reproducible processes.
The evaluation problem becomes more consequential when AI controls physical systems. Reyzelman described systems that perceive their surroundings, pass environment data to a model, and use the model’s output to guide physical actions. Failures in these systems can have impacts beyond a wrong answer on a screen.
Practitioners therefore still need to evaluate evidence, recognize weak assumptions, and decide where automated action should stop. Better models can reduce some manual work, but they also increase the value of people who understand the domain well enough to know when a system’s output does not fit the situation.
Takeaways and next steps
AI systems can now operate faster and more independently than many supporting processes. Consequences include:
- Security teams must defend at machine speed.
- Infrastructure planners must account for limited physical resources (power, cooling, water and data center capacity).
- Researchers, students and practitioners must prioritize domain knowledge and critical evaluation of increasingly capable systems.
The 144‑to‑one ratio makes this shift concrete: agent adoption is already testing whether organizations can govern these systems, provide the required infrastructure, and preserve informed human oversight.
Further episodes
This Week in AI will continue to explore news, issues and developments shaping the AI era in its next episode, with weekly releases across platforms including YouTube, Spotify and Apple.
