I recently joined Wonder, and as with any new job, we need to learn as much as possible about the business, the organization, the tech stack, as fast as possible. This is. Very common pattern in all companies in any industry. It is not just about the volume of information. It was how that information is managed and made accessible, or not. New acronyms, org structure, technical docs scattered across documents and repositories, meetings whose context I was missing. The traditional implementation for this is to allocate a person. You get an onboarding buddy, a knowledgeable human who answers your dumb questions for the first month, so you do not have to interrupt tens of different people.

That is the model we have today, at least what is expected. But now in the age of AI and agents, I think it is the wrong default.

I built my own AI onboarding buddy, and it changed my first few weeks#

Instead of saving up questions for one overloaded colleague, I built an agent. I fed it the common-knowledge new hires need: the acronyms, the team and org structure, the technical documents my team kindly assembled, and the recurring context that usually takes weeks to absorb. Now when I have a question, I just ask the agent. What does this term mean? Who owns this area? Where does this process live? It answers in seconds, at any hour, without making me feel like I am burning someone else’s time. Any new relevant information I get, I ingest into my Agent’s database.

The difference is not speed alone. It is that I can ask the same question with three different follow-ups without embarrassment, follow a thread until I actually understand it, and learn at my own pace instead of the pace of someone else’s calendar.

This is not a Wonder idea. Ramp has gone further than most, building an AI-native culture with internal tools that turn employee workflows into reusable skills every new hire can lean on. The pattern is clear: the onboarding buddy of the future is an agent, and every company should have one (or several).

The conviction: the buddy should be an agent by default#

A human buddy gets pulled into their own work. They forget half of what they meant to tell you. They go on vacation in your second week. An agent does not. It holds the full context of how a company actually operates and makes it available the moment someone needs it.

I want to be clear, I am not advocating for companies to remove human onboarding experiences. The welcome, the relationships, the little details and nuances are not replaced by a knowledge agent. The human connection is vital for a proper, welcoming onboarding experience. But the constant stream of “how does this work here” questions should go to an agent first. That frees your colleagues to do the part only humans can do, which is to make you feel like you belong.

The block: your systems were not built for this#

Here is where most companies hit a wall. An onboarding agent, or any agent really, is only as good as the systems and data it can reach, and each department, org, and company has its own knowledge bases, acronyms, tools, meetings, and structures. To build a useful agent buddy, all of that has to be imported consistently and made available to the agent.

For my Wonder Knowledge AI Agent, the vast majority of information is stored on Google Drive and Docs, some information is on Atlassian products like Confluence and Jira, so I used the GWS CLI + GCP and the Atlassian CLI and wrote a local ETL to transform the information and store it in processed local .md files that are easy to search on.

That is exactly where a sizable number of internal tools falls apart. Most companies run on legacy Saas solutions built decades ago. They have enormous functionality, and they are frozen in time. They were designed for a human to log in, look at a UI, and click buttons. They were never designed to interact and exchange data with an agent.

Workday is the cleanest example. Your entire organizational structure lives in there. But there is no proper headless version of Workday you can point an agent at (Workday recently announced its Agent strategy). I solved this by exporting Workday data to BigQuery and then accessing it programmatically via BigQuery APIs or local static files.

For CIOs and IT teams: headless is now table stakes#

This is the part I want every technology leader to internalize. Every solution you buy from now on has to connect cleanly with agents. Not eventually. Now.

It is crazy to me that a company would buy a tool today that only gives you a UI to click around in. A serious product needs proper APIs, a proper CLI, or at minimum an MCP server you can point your agents at, with a setup simple enough that connecting an agent is not a six-month project.

So make it a hard requirement. As you evaluate new solutions, and especially when vendor renewals come up, treat the headless version as a critical line item. Can an agent read from it? Can an agent act through it? How hard is it to wire up? How are the permissions and governance controls? If the answer is “you can only use our UI,” that should be a major ding on that vendor.

The RFP of the next few years will not be won on features alone. It will come down to whether your agents can use the thing. Vendors who understand this will build for it. Vendors who protect a seat-based moat by staying UI-only or making it difficult to access the data and the execution are selling you a dead end, and you should price that risk into every contract.

Conclusion#

My onboarding buddy was relatively easy to build, gathering the data was a mix of easy, just ingesting the docs that were shared with me, and hard, like with Workday data. Onboarding is just the first place this shows up. The same wall blocks every other agent you will want to deploy. I am not even going to discuss the challenges of building proper engineering agents here.

The companies that move fastest over the next few years will not be the ones with the most tools. They will be the ones whose tools can talk to agents. So the next time you sign a contract, ask one question before anything else: Can my agents use this? If the answer is no, you already know what to do.