Diskover's Will Hall: Optimizing Unstructured Data for Enterprise AI

In this TechVoices interview, Erkel discusses how enterprises can bring customer feedback into AI development loops, how humans can evaluate sometimes unpredictable AI agents, and why he believes automation will accelerate product development without eliminating the need for human judgment.

At Black Hat 2026, Pentera CEO Amitai Ratzon discussed how artificial intelligence is reshaping offensive security, from AI-native attack technology to agent-driven security operations.

As enterprises race to deploy AI agents, Incorta co-founder and CEO Osama Elkady argues that the real challenge is not simply choosing a more powerful AI model—it is building the data foundation that allows agents to work reliably.

A perfectly secure app can still expose your entire Salesforce instance, depending on what it's connected to.

Guardrails are just instructions an agent might ignore—enterprises need controls that can actually block it.

Expertise and passion aren't enough—even brilliant leaders lose rooms when ideas aren't delivered well.

Defenders currently know more about their own systems than any attacker—but that edge won't last.

An LLM that can't tell Montana from California can turn one confident answer into a legal problem.

Thirty years of identity architecture built for humans is now obsolete for governing thousands of agents.

An agent that only sees your CRM can approve a 'profitable' order it has no way of actually verifying.

AI teams chase accuracy, security teams chase firewalls—neither fully understands AI's new attack surface.

Some customer codebases grew 5x in a year—speed that's now outpacing teams' ability to keep it reliable.

One-shot AI video generation gets you 80% there—the rest still needs a human steering the wheel.

Attackers only need to succeed once; defenders have to succeed every single time—AI just raises the stakes.

One compromised agent with broad access can now unlock reach across an entire organization's systems.

Non-deterministic AI breaks every security assumption built for systems that behaved exactly as expected.

CFOs built careers on precision—now they're being asked to trust systems that are inherently probabilistic.

AI agents need their own identities—including a verifiable kill switch when something goes wrong.

Centralizing your most sensitive data into one vector database quietly builds an attacker's dream target.

Most AI workflow failures aren't the model's fault—they're memory limits, bad GPUs, or flaky networks.

Most companies aren't short on AI enthusiasm—they're stuck untangling siloed, ungoverned data.

AI isn't magic—the goal is to make magicians, not to replace the last mile of human judgment.

In hardware, a hallucination means a product overheats or fails—so this AI can't afford to guess.

A weekend of vibe coding gets you a 0.01 version—real enterprise software takes a lot more than that.

Agentic AI isn't judged by the tool it hands you—it's judged by whether the outcome actually happened.

Attackers don't break in anymore—they log in, using stolen credentials as the front door.

The open internet holds 15 zettabytes of data; enterprises are sitting on 200—mostly untouched by AI.

Five execs agree: the AI honeymoon is over, and 2026 will reward integration over hype.

APIs were built for human developers, not AI agents—that mismatch is quietly killing enterprise ROI.

Every sensor, physical or virtual, speaks time series—making it the native language of AI-era infrastructure.

A CRM's version of an order rarely matches the ERP's—AI agents inherit that confusion unless it's fixed.

A CRM's version of an order rarely matches the ERP's—AI agents inherit that confusion unless it's fixed.

Generative AI is like a two-year-old kitten—full of promise, still learning where not to scratch.

High AI pilot 'failure' rates aren't bad news—they're enterprises learning what actually sticks.

RPA scripts each step; agentic AI just aims for the KPI and figures out how to get there.

Before debating CPU vs. GPU, most enterprises could free up 70% of their space and power just by upgrading.

A calculation costing a year's worth of global electricity on a supercomputer ran for under a dollar.

From EV safety systems to NASA spacecraft, open RISC-V cores are quietly running critical hardware.

Think of MCP as a USB-C port for APIs—and treat every new connection like a new attack surface.

One customer saved $120 million in three weeks—proof agentic AI's ROI is already real.

Real resilience means knowing your 'minimum viable business' before an attack, not after.

AI resolves the issue before you even file the ticket—that shift is what changes IT's cost curve.

AI agents won't replace developers, but they're already reshaping how dev teams are structured.

Writing code is no longer the differentiator—ideas and systems thinking are what still matter.

Managing AI security agents is a lot like managing very fast, very junior employees.

Posts from real employees consistently outperform polished corporate content—by a wide margin.

70% call AI security a shared responsibility, yet only 13% have a dedicated team for it.

Blocking AI to protect data backfires—it just pushes employees toward riskier, unapproved tools.

A model trained on your data can be more sensitive than the data itself—and harder to protect.

Built for HPC before ChatGPT existed, this chip turned out to be exactly what AI networking needed.