The Week AI Got an Org Chart Problem

When Your Best Employee Is a Process

There's a piece in Fortune from April 13 that nails something I've been chewing on for months: AI agents are acting like employees, but company structures still treat them like software. Read that sentence again. It's not a metaphor problem. It's an organizational design problem, and it's the kind that doesn't announce itself with a bang but with a slow accumulation of awkward workarounds until one day someone realizes the org chart is fiction.

Think about what it actually means for an AI agent to "act like an employee." It means it's making decisions, routing information, interacting with other systems (and people) on behalf of the company. It has a scope of responsibility. It might even have something resembling judgment, or at least a convincing facsimile. But when something goes wrong, who gets the call? Not the agent. The engineer who deployed it? The manager who approved the workflow? The vendor? Nobody has a good answer yet, and that's the point.

The instinct, naturally, is to shove agents into existing IT governance frameworks. Treat them like any other SaaS tool. Provision access, set permissions, run audits. But that's like managing a new hire through your procurement system. The mismatch isn't just bureaucratic, it's conceptual. Software doesn't negotiate priorities between departments. Software doesn't decide which customer email is urgent. Agents increasingly do.

Governance Is the New Product

This is exactly why Databricks' announcement on April 15 about expanding agent governance with their Unity AI Gateway feels less like a product update and more like a thesis statement. If agents are proliferating inside enterprises (and they are), then the layer that monitors, controls, and audits those agents becomes critical infrastructure. Not optional. Not nice-to-have. The plumbing.

And Cloudflare is making a similar bet from a different angle. Their "Project Think" blog post, also from April 15, describes building AI agents directly on Cloudflare's infrastructure. So now you've got one company saying "we'll help you govern your agents" and another saying "we'll help you run them at the edge." The governance layer and the execution layer are being built simultaneously by different players, which is either a sign of a healthy ecosystem or a recipe for fragmentation. Probably both.

Look, the uncomfortable truth is that most companies deploying agents right now are winging it. They're bolting autonomous decision-making onto organizational structures designed for humans who can be fired, retrained, or at minimum asked "what were you thinking?" The Fortune piece captures this tension perfectly, even if it doesn't resolve it. Nobody resolves it yet. We're in the "figure it out as we go" phase, which (and yes, I know how that sounds) is actually fine as long as people admit that's what's happening.

NVIDIA's Quantum Bet Is Sneakier Than It Looks

OK, let's talk about NVIDIA. On April 14, they launched something called Ising, described as the world's first open AI models designed to accelerate the path to useful quantum computers. This got covered by basically everyone: NVIDIA Newsroom, The Quantum Insider, The Next Platform, Quantum Computing Report. Four sources on the same announcement. That alone tells you something about how starved the quantum space is for tangible progress.

Here's what's sneaky about it. NVIDIA isn't building a quantum computer. They're building the AI models that make quantum computers less useless. That's a very NVIDIA move. They did the same thing with GPUs and deep learning a decade ago: don't build the thing, build the thing that makes the thing work better, then become indispensable. The Ising models are open source, which means NVIDIA is essentially seeding the quantum development ecosystem with tools that run on (surprise) NVIDIA hardware. It's generous in the way that giving away razors is generous.

The "open" part matters here, though, and not just as marketing. As The Next Platform framed it, NVIDIA is bringing open-source AI models to quantum computing. That's a deliberate philosophical choice at a moment when the open-source debate in AI is genuinely heated. Which brings us to the counterpoint.

Open Source's Identity Crisis

ZDNET published a piece on April 15 with a headline that stopped me cold: "Like handing out the blueprint to a bank vault." That's how one company described why AI led them to abandon open source entirely. The logic is straightforward, if debatable: when your AI models encode deep knowledge about your systems, your data, your vulnerabilities, open-sourcing them is a security risk that no amount of community goodwill justifies.

So here's the tension. NVIDIA open-sources Ising to accelerate quantum development. Another company locks everything down because openness feels like exposure. Both positions are rational. Both are responding to the same underlying reality: AI models are not just code, they're compressed knowledge, and knowledge is power, and power is dangerous in the wrong hands. The question isn't whether open source is good or bad (that's a boring question). The question is: open source of what, exactly? A model that helps simulate quantum error correction is a very different artifact from a model trained on your company's internal security architecture.

NVIDIA can afford to be open because Ising's value accrues back to their hardware ecosystem regardless. The company in the ZDNET piece can't afford to be open because their model's value IS the proprietary knowledge embedded in it. Same word, "open," completely different risk profiles. Context is everything, and anyone who tells you there's a universal answer to the open-source-in-AI debate is selling something.

The Infrastructure Layer Hardens

Step back and look at the week as a whole. Cloudflare is building agent execution infrastructure. Databricks is building agent governance infrastructure. NVIDIA is building quantum-AI bridge infrastructure. These aren't consumer-facing announcements. Nobody's going to tweet excitedly about Unity AI Gateway. But this is where the real game is: the picks-and-shovels layer that determines what agents can do, where they run, and who watches them.

The Fortune piece about agents-as-employees is the human story. The infrastructure announcements are the engineering story. And they're the same story, really. You can't have agents acting like employees at scale without governance frameworks, execution platforms, and (eventually) something resembling HR for non-human workers. That last part sounds absurd until you remember that we already have SLAs, uptime guarantees, and incident response protocols for software. Agent governance is just the next iteration, with higher stakes because the "software" is making judgment calls.

What sticks with me is that image from Fortune: agents acting like employees inside structures built for software. It reminds me of those photos of the first cars on roads designed for horses. The roads eventually changed. The org charts will too. The only question is how many fender-benders happen in the meantime.