I've been in tech long enough to remember the early days of the web, when the message to every organization was simple and urgent... get online or become invisible. Nonprofits scrambled to register domains, build websites, and figure out what "having a presence" even meant. Most got there eventually. The ones that didn't fell behind in ways that were hard to recover from.
I'm watching that same pattern take shape right now, and I'm not sure we're ready for it. For years we've been predicting an "internet of agents," a layer of the web where AI systems act independently, retrieving information, executing tasks, and spending money without a human involved. That's not a prediction anymore. In the first two months of 2026, the infrastructure buildout has moved faster than most organizational leaders realize.
Coinbase, Stripe, Cloudflare, OpenAI, and Exa.ai have all shipped agent-facing infrastructure in the last ninety days. Agents can now hold and spend cryptocurrency, make payments through commerce APIs, read websites as clean machine-readable text, search the web with purpose-built engines, and run code in real terminal environments. These aren't prototypes. They're production systems from companies that don't ship things for fun.
Stack all of that together and agents can now read the web, search it, execute code, and pay for things. They're chaining services on the fly. One agent scrapes product images from a link, feeds them into a video generator, and delivers a finished marketing clip in minutes. Others are running algorithmic trades on prediction markets. This isn't experimental. It's operational.
What unsettles me about the comparison to the early web is the relationship layer. Back then, you showed up yourself. You wrote the content. You managed the conversations. You owned the relationships. This time, your agents show up for you. AI acting on your behalf in interactions you never see. I keep coming back to the idea that we're building infrastructure for autonomous representation before we've figured out autonomous trust. That order bothers me.
The security picture reflects the gap. Agents with wallets and shell access can be drained by malicious injections, tricked into running harmful code, or fed poisoned data. A 2025 Deloitte survey found only 6 to 11 percent of enterprises fully trust agents for end-to-end processes, even as 86 percent plan to increase investment. The infrastructure is sprinting. The governance isn't.
So where does that leave mission-driven organizations? You don't have to be first. You do have to be intentional.
Start with information, not services. Build agentic information exchanges that make your organization useful and findable by AI systems. That means exposing structured, clear details about what you do, how people access your services, your hours, your intake processes, your locations. When someone's AI agent searches for help on their behalf, it should be able to summarize quickly and accurately how to connect with you. Think about a person in crisis whose agent can parse your intake process and tell them exactly where to go and what to bring. That alone is worth the effort.
Then expose a small, intentional set of touchpoints that connect to a real human being. Appointment scheduling. Inquiry routing. Callback requests. You're not automating the relationship. You're shortening the distance between someone who needs help and someone who can provide it.
Set up a payment-receiving wallet too. I realize crypto still makes a lot of nonprofit leaders uneasy, and I get it. But agent payment infrastructure is here whether organizations are comfortable with it or not, and having a receiving capability in place beats scrambling when the first transaction shows up unannounced. Start working on your crypto management policy now. You want that framework decided before it's urgent.
Before any of this, though, do the governance work. That's the part I don't want anyone to skip. Establish clear boundaries for what you're willing to expose to agents and what stays behind human gatekeeping. Take a hard look at your data privacy practices, especially if your organization handles sensitive constituent information. When you make services discoverable to AI agents, you're also making them discoverable to systems that may scrape, cache, or reprocess that data in ways you didn't anticipate. If you've been following my series on AI privacy for nonprofits, you already know where I land on this. Know where your boundaries are before you open the door, not after someone walks through it.
OpenClaw showed us what this future feels like up close. Sixty-eight thousand GitHub stars in three days, followed almost immediately by Cisco calling it a security nightmare and Anthropic cutting off OAuth access. A personal agent that could browse, transact, and act on your behalf, and a trust infrastructure that couldn't keep up with what it made possible. That cycle is going to keep repeating. Self-hosted agents like OpenClaw will seem quaint once agentic capability is just part of how every platform and software layer works. We'll still have personal agents, but they won't be something you install and configure. They'll be baked into the infrastructure we stop thinking about, the same way we don't think about DNS or TLS anymore.
I don't have this fully figured out. Nobody does. The infrastructure is moving faster than the trust frameworks, and that gap is going to produce some ugly surprises before it narrows. But the organizations that show up early and thoughtfully, that make their front door readable by machines without handing over the keys, are going to be in a much stronger position than the ones still debating whether this is real.
It's real. The question is just how you want to meet it.