Using the right tool for the job is always better.
Anyone who does DIY projects around the home knows how using the right tool can dramatically make the job you are doing far easier. Use the wrong tool, and that task suddenly becomes a burdensome nightmare.
And after over 38 years in cybersecurity, I know that applies to cyber defense strategies, but I add one more axiom: Use the dumber, faster thing first for best results. Dumber things are usually faster at blocking a large number of things. Smarter tools are better at the details, but slower. So, start defending and blocking with the faster, dumber tools before moving onto the slower, smarter tools.
For example, when setting up a network security boundary. I am a big believer in using the dumber, faster tools first. This means, if you can set up something physical to block a lot of bad traffic, do that first. Then use a router with defined paths to block as much of the bad traffic as you can. You use the router to define what is or is not inside the internal security domain and construct other domains as you need them.
Then, and only then, use a firewall with a deny-by-default rule set. It is only inspecting and blocking traffic that gets past the router. Anything that gets past the router should then be inspected by an application-level proxy and/or firewall. That device tries to block any anomalous application-level commands or data. Only at the end should your involved application inspect the incoming traffic and commands, root out improper inputs, and use isolated identity accounts and ACLs to secure the application further. After that, you have logging and humans to finish out the pathway.
During each phase of the incoming traffic, the dumber, faster device filters out as much of the nonsense as possible. You want your smarter, slower devices to handle as little of the workload as possible.
This applies to AI as well.
I was talking to some of the KnowBe4 engineers and developers, who are heads down, working on our agentic AI products, and they shared with me their strategy for making AI use and responses as fast as possible. After hearing what they said, I figured it couldn’t hurt to share their strategy.
Efficiently Applying AI
Traditional applications with IF-THEN statements and deterministic logic are dumber and faster. AI is non-deterministic, smarter but slower. Use each where it makes sense. In many cases, it makes sense to intercept incoming requests first with a slower, traditional program and only pass along what the traditional program cannot adequately handle to the AI. And then pass along what the AI cannot do to the human (if involved).
I am a big believer in giving a human an opportunity to interact with another human during these days of early AI and hallucinations, especially in customer support scenarios. That way, if an AI is not able to resolve a customer’s problem or request to their satisfaction, it can be heard by a human. Final appeals should always be evaluated by humans (at least for now).
Back to the discussion on AI efficiencies.
If you have multiple AI stacks, meaning you have your own local AI or AI agent(s), use them first, if you can. Then use a vendor’s stack (e.g., Microsoft, Apple, etc.), and then, and only then, push up work to a large AI cloud vendor (like Anthropic, OpenAI, etc.). Use the fast local AI first and only move up the line to further away AIs as you need them.
KnowBe4’s programming teams have been using this approach since the very beginning: Use the right tool where it is needed. Use AI where it makes sense, but don’t use AI everywhere just to say you are using AI. Use the right tool.
KnowBe4’s AI
KnowBe4 has been leading with AI for over 10 years now. Businesses that utilize KnowBe4’s AI-powered features and agents significantly lower their cybersecurity risk compared to those who do not. For example, customers who use our Artificial Intelligence Defense Agents (AIDA) to manage their simulated phishing tests are able to provide more education to users who “fail” those tests at a rate two to three times greater than if the simulated phishing templates were selected by human admins. That means more anti-social engineering education, which is real cyber risk reduction.
AIDA also helps customers to manage and operate our product, including more easily generating new phishing templates more personalized for their company and targeted employees, the ability to more easily push remedial training, AIDA-generated Knowledge Refreshers, and AIDA-generated policy quizzes, based on your company’s own policies.
Every product and service we have is either being supplemented by an AI agent or being invented from the ground up as an AI agent. Our goal is to help you better manage human risk by managing both the human and their AI agents to best protect your organization.
At the same time, we realize that you don’t always need AI for every feature and improvement. Our customers tell us the same thing. In a recent customer survey, our customers told us that having AI for the sake of just having AI ranks very low on their list of priorities. But if AI can deliver improved and new feature sets that really reduce cyber risk, then they are all-in.
KnowBe4 knows this. We are using AI where it makes sense to use AI and traditional products where it makes sense as well. We will combine the best of both worlds to provide the best in products, customer satisfaction and to best reduce cybersecurity risk.
We have and are creating more and more AI agents to protect your entire enterprise, humans and agents, to protect your environment. We’ve protected humans since our founding, and now we will protect the agents those humans are using. And we will do it in efficient ways that provide the best accuracy and speed to best lower your cybersecurity risk.
No need to reinvent the wheel, use the dumbest, fastest tools first before moving on to the faster, smarter tools. Works every time.
