To all fellow cybersecurity professionals out there,
We’ve been in the trenches of InfoSec for a long time, some of you alongside me for nearly four decades. In that time, the job of the pen tester has shifted from shell scripts and packet captures to cloud-native apps, microservices, and CI/CD pipelines that never sit still. And now, agentic AI is changing pen testing faster than most of us ever expected.
The real question that I believe should be top of mind for pen testers everywhere this New Year is not “Will agentic AI replace us?” but rather, “How will we evolve with it?”
Pen testing as we know it today
Nothing has been able to replace the depth of a strong, context-aware, human-led pen test. We human testers can apply the attacker’s mindset, spot odd behaviors, chain together unrelated quirks, and improvise when facing strange legacy integrations. We understand which vulnerabilities truly matter to the business, not just to the CVSS calculator. We can decide when to strategically stop, how to avoid disrupting critical systems, and how to communicate risk in a way teams can act on.
However, the workflow for many of us looks like this: long scoping calls, rigid rules of engagement, a few intense weeks of manual testing, then a race to finish a PDF before moving on to the next client. That model has real limitations today. We’re spending too much time on repetitive enumeration, regression checks, screenshots, and report writing rather than on creative attack paths. Even elite teams are struggling to keep up with the dozens of web apps, microservices, and rapid releases that happen in today’s enterprises
Enter agentic AI
Agentic AI can now be used to change workflows, not replace experts. The pen tester can evolve from “human doing everything” to “human-in-the-loop”. That means orchestrating agents, curating playbooks, making judgment calls, and translating findings into better defenses. Pairing agentic AI and a human precisely eliminates the cons of traditional pen testing work, while preserving all the pros.
Take Terra Security’s approach, whose platform uses a swarm of specialized AI agents trained on real attack strategies to continuously scan and re‐scan web apps, with human oversight. Instead of a single scanner running a static checklist, dozens of agents track changes to the attack surface and launch context-aware tests. Promising leads (not just raw alerts) are then escalated to human testers, who validate exploitability, refine the attack chain, and frame the impact. Agentic AI changes the physics of pen testing, not the need for pen testers. It is ushering us into a new reality: continuous, context‐aware pen testing that can serve as a strategic layer of your security posture.
What this shift means for key stakeholders
For pen testers
Agentic AI is a force multiplier, not a rival. With agents doing the repetitive lifting, you can focus on creative chaining, lateral thinking, and high‐value investigation. With more time, you can deepen purple‐team work, threat hunting, and architectural reviews, where your attacker mindset makes the biggest difference. The job becomes less about manually repeating the same tests and more about designing, supervising, and extending an agentic system's offensive capabilities. For me, the main reason to become a good attacker was always to become a valuable defender. The dopamine hit of a good exploit is real, but the long‐term satisfaction comes from helping organizations build better defenses around what you learn.
For security leaders
For security leaders, this shift turns pen testing from a project into a program. Leveraging an agentic platform with Human‐in‐the‐Loop oversight enables you to get continuous validation aligned to application changes, with findings getting prioritized by exploitability and mapped to real business impact. Instead of asking, “How many issues did the last pen test find?”, leaders can ask, “How quickly do we detect and fix truly exploitable issues as our systems evolve?”
For dev and DevSecOps teams
For developers and DevSecOps teams, agentic pen testing with human-in-the-loop oversight reduces friction. Continuous testing can plug into staging environments, surfacing exploit‐validated issues closer to when code changes. The focus on business logic and exploit evidence helps teams distinguish critical risks from background noise. And human‐in‐the‐loop governance ensures that risky actions are controlled, reducing the chance that tests break production workflows. The result is not “more findings,” but better, more actionable findings that align with how engineering teams actually work.
Why I’m embracing this evolution
From a veteran pen tester’s perspective, being able to operate a multi‐agent architecture essentially gives us a 24/7 red team that never sleeps. Having this team with a governance layer that allows us to validate high-impact findings, make tough calls, and ensure ethical, safe operations is critical.
This is made possible by proxy tools, such as Terra Portal™, which enable agents to behave more like disciplined penetration testers. That means enforcing expert playbooks inside the traffic itself or ensuring sensible attack patterns are followed. That also includes enabling the agents to pause and escalate potentially disruptive actions to the human-in-the-loop.
Companies pairing agentic AI in this way are not automating pen testers out of the loop. They’re automating pen testers into the parts of the loop that actually require judgment. That combination is what convinces an old‐fashioned pen tester like me that agentic AI is not a gimmick; it is the next logical evolution of our craft.
Promising paths but dangerous dead ends ahead
The innovation happening in our space is remarkable. New agentic models, multi‐agent orchestration, and continuous Pen Test as-a-Service (PTaaS) approaches are moving us away from one‐off projects and establishing pen testing as a core piece of the enterprise security Stack.
That said, I’m wary of some approaches, such as purely autonomous “AI hackers.” Some of these platforms that promise fully unsupervised AI pen testing have little explanation of safety, governance, or human validation, which is especially concerning when done in production environments. Others ignore business logic entirely, failing to reason about workflow contexts, roles, or user journeys.
As attackers adopt their own automation and AI, pen testers have a choice: cling to antiquated workflows or evolve into true humans-in-the-loop, steering powerful agentic systems toward deeper insight and better defense.
This is not the end of the pen tester. For those willing to adapt, it is the most interesting chapter yet.
