Artefact CTO Victor Coimbra: Why AI Cannot Think About the Future

Artefact Americas CTO Victor Coimbra and Nathan Kirk on NR Labs Podcast Episode 2: AI Cannot Think About the Future

NR Labs Podcast, Episode 2 · Host: Nathan Kirk · Watch on YouTube

On Episode 2 of the NR Labs podcast, our cybersecurity podcast for technology and security leaders, host Nathan Kirk talks with Victor Coimbra, CTO of Artefact in the Americas. They cover how Artefact governs AI without a central bottleneck, why critical thinking is the skill most at risk, and why, in Victor's words, "AI cannot think about the future."

Watch and follow

Who is Victor Coimbra?

Victor Coimbra is the CTO of Artefact in the Americas and has worked with AI for almost a decade, "before it was cool," as he put it. He was previously a high school physics teacher, which shapes his view that critical thinking, not memorization, is what makes someone good at the work.

How should enterprises govern AI use?

Artefact uses decentralized AI governance with strong guardrails. Teams choose their own AI tools, but within three controls. Victor's goal is to "control the chaos" without a central team that slows everyone down.

  1. Strong guardrails. There are many ways to build something, but "there's only one bridge that you can cross": a single deployment gateway with the same tests and rules from dev to production.
  2. Individual budgets. Each person gets their own pool of money for APIs and subscriptions and is responsible for it.
  3. Non-negotiable rules. Usage is audited continuously. Break the rules and you lose your access to the AI tools.

Building governance enforced through gateways and monitoring instead of paperwork is the focus of our Advanced GRC services.

What is AI skill atrophy?

AI skill atrophy is the gradual loss of critical thinking when people hand too much of their thinking to AI. Victor sees it as an organizational problem: the pressure to produce more, faster, leaves no room to think.

His two fixes:

  • Protected thinking time. Block an hour or two on the calendar for reflection that isn't tied to a task.
  • Teaching spaces. "The best way for you to learn something is by teaching."

How do you avoid AI brain fog?

Decide up front how much of each task to hand to AI. Victor sorts work into four buckets:

  1. Delegate fully: low-stakes tasks like meeting summaries and follow-ups.
  2. Collaborate: presentations or interview prep, where you keep much of the thinking.
  3. Understand: new clients or industries. Don't offload; use the agent to ask better questions.
  4. Create: prototypes and mockups, where you can offload more of the build.

Can AI make decisions?

AI is reliable for decisions based on past data, and unreliable for decisions about the future. Victor trusts models to find patterns and forecast from historical data, but not to solve new problems.

"By definition, AI cannot think about the future," he said. "They can only predict what was in the past, and I think it's our job as humans to actually think about the future." His rule: "If the solution of what you're trying to solve lives in the future, try to solve that first before you go the AI route."

How do engineering teams keep their skills when AI writes the code?

Move the human effort from reviewing code to designing systems. Victor's teams spend less time debugging AI-written code and more time stress-testing architecture before anything is built. "AI can execute extremely well, but you need to be really thinking about what you're building before you build," he said.

How do you secure agentic AI?

Not with yesterday's controls alone. AI is very good at learning patterns from the past, so if AI systems are protected only with familiar methods like sandboxes and VMs, AI will eventually find a way around them.

"We cannot be naive into thinking that if we handle this in the same way that we handled it in the past, after millions of training rounds AI is not going to figure out a way to break it," he said. Security teams need new protocols and architectures, not better versions of old ones. Building those defenses is what our Cyber Innovation services work on.

Key takeaways

  • Let teams choose their AI tools, but give them one approved path to production.
  • Give people their own AI budgets and audit usage continuously.
  • Protect time for thinking and teaching so critical thinking doesn't erode.
  • Trust AI with the past; keep humans in charge of the future.
  • Secure agentic AI with new approaches, not just legacy sandboxing.

Watch the full episode

Hear the full conversation between Nathan Kirk and Victor Coimbra: Watch Episode 2 on YouTube. For more AI and cybersecurity podcast conversations, subscribe to the NR Labs YouTube channel and follow NR Labs on LinkedIn.

FAQ

How does Artefact govern AI use?
With decentralized governance and three guardrails: one approved deployment path to production, individual AI budgets, and non-negotiable rules that are audited continuously.

What is AI skill atrophy?
It's the loss of critical thinking and problem-solving skills when people offload too much thinking to AI. Victor Coimbra recommends protected reflection time and teaching others to counter it.

Can AI make decisions about the future?
Victor Coimbra says no. AI is reliable for decisions based on past data, but new problems need human creativity because AI models are trained on the past.