Jensen Huang, Nvidia's famed CEO, rarely issues praise. After Nvidia had just "blown the doors off the quarter," Jensen stood up in front of his team and said: "I look in the mirror every morning and say: You Suck."

It matters then that two weeks ago at Nvidia's annual GTC conference he said that: "Every company in the world today needs to have an OpenClaw strategy, an agentic systems strategy. This is the new computer."

GTC keynotes are almost exclusively about Nvidia's own products: new GPUs, new platforms, partner announcements where Nvidia is at the center. Jensen mentions external companies regularly, but always in the context of "they're building on Nvidia." OpenClaw is structurally different. He gave two hours of airtime to an open-source project built by an Austrian indie developer, called it the most important software release in history, and said every CEO on earth needs a strategy around it.

So why is Jensen "Claw Crazy", and how did we get here?

What makes OpenClaw unique

OpenClaw is the first AI that behaves like a person you hired, not a tool you use. It runs on your own hardware, 24 hours a day, seven days a week. It doesn't wait for you to show up. It watches for things that need attention, acts on them, and tells you what it did.

The best analogy: the difference between a calculator and an accountant. A calculator does nothing until you punch the buttons. An accountant watches your accounts, notices a problem at 2am, and texts you about it in the morning.

Let's unbundle three critical components of OpenClaw:

First: it can start on its own. A scheduled time fires, a message arrives, a new email lands. The agent wakes up and works without you touching it. No consumer chatbots before did this natively.

Second: it remembers. Every AI model is stateless by nature, meaning it starts from zero every single time. OpenClaw solves this by storing memory outside the model, in plain readable files sitting on your own machine. Actual files you can open in any text editor. There's one called SOUL.md, which defines the agent's personality, its rules, how it should behave. There's MEMORY.md, which is what it has learned about you, your preferences, your past tasks, your working patterns. And there's HEARTBEAT.md, which is a list of things it should proactively check on its own, on a schedule, without being asked. When the agent wakes up, it reads these files back in and picks up exactly where it left off. You can edit them directly, back them up, put them in Git. Open the file, read it, change it. Done.

Third: it lives where you work. This is what makes the first two matter. OpenClaw has persistent, live connections to the surfaces where your work actually happens: WhatsApp, Slack, email, Telegram, iMessage, your calendar. OpenClaw is always on, accumulating context about what's going on and ready to act wherever you need it to. Because it's permanently embedded in your actual work environment, when it acts, it acts with a full picture of what's been happening, not a cold start.

What this means for how we build at Gigi

We didn't need Jensen's keynote to tell us this. There are "OpenClaw" primitives in aspects of our product today, and will continue to become more "Claw-like" as we evolve.

The primary way our customers interact with our product is through "tasks." Our customers assign Gigi a task on a scheduled basis or based on a triggering event, Gigi completes the work of the task and our customers review and accept Gigi's work. Event based tasks provide an excellent example of Gigi's "Claw-like" capabilities. For example, a common event based task are domain exclusions. Our customers often ask Gigi to review any domains or apps that have received a lot of impressions with limited to no outcomes (conversions or detailed page views). To do this, Gigi will, every day, download the inventory report in the Amazon DSP, review the report to see if any domains or apps met the defined conditions, and allow our customers to block those domains across every single line item. An agency can create this task once across every single client, and Gigi will perform this task without provocation every single day. It's a perfect example of an always-on agentic interaction that could never be replicated by human touch at scale.

Next, Gigi is going to start meeting our customers where they work. Gigi will begin to appear (with permission, of course) in Slack, email, and project management software so Gigi can benefit from improved context across an agency or brand wherever they work and, most importantly, take directions for actions from whichever surface area or mode of communication is most convenient to our customers.

We're also building Gigi a better memory. Not memory in the abstract sense, but a running picture of every advertiser's world: where their campaigns stand, how they're pacing, what's been tried, what worked. That picture updates continuously. When something happens, Gigi doesn't treat every signal the same way. A minor impression dip gets noted and filed. A budget about to exhaust three days before the end of the month gets acted on immediately. Gigi learns the difference.

Every action Gigi takes gets recorded in full: what triggered it, what options were on the table, what it decided, and why. Over time, that record becomes the most valuable thing we build. When Gigi adjusts bids on a campaign that's overpacing, it knows what it did last month in the same situation, what the outcome was, and whether to do it the same way again.

This last capability is where we believe we'll be able to demonstrate true differentiation. Workflow automation is the floor. Any well-resourced team can replicate it. What compounds into a real moat is decision and context capture: what the agent chose, under what circumstances, and what happened as a result. That's our OpenClaw strategy. Hopefully Jensen approves.

Note: This was originally published as part of our Cherry Picked newsletter. Subscribe here ➡️ https://cherry-picked-gigi.beehiiv.com/subscribe