Stop Building Agents. Start Collecting Data.
The real competitive advantage in AI isn't the model. It's what you feed it.
Welcome to the 93 new FullStack HR readers who joined last week - welcome all to this publication. And hello, if you got sent this link - if you aren't yet subscribed, join the other like-minded people in this free newsletter by subscribing below:
The models keep getting better. Every month, every update, more capable. What Claude and ChatGPT can do today would have been unthinkable a year ago.
But everyone has access to the same models. Your competitor pays the same fee for the same API as you do. The playing field is completely level when it comes to the technology itself.
So what gives you an edge? Data.
I should probably have understood this more clearly earlier. “Data” has always been one of those abstract buzzwords that sound important but feel impossible to act on. But running OpenClaw, my own AI team, has made it painfully concrete. The difference between a useful agent and a useless one isn’t the model. It’s what you feed it.
Your most valuable data doesn’t live in any system
I keep having the same conversation with HR leaders. They want agents. And I get it!
I believe in agents long-term, we are clearly heading down that path. Microsoft launched Cowork yesterday, and I’ve built several agents myself. My labor law bots for Swedish and UK employment law save me ridiculous amounts of time.
But… your most valuable data isn’t in your HRIS or your ATS. It’s in your head.
Every time you resolve an employee relations case, that’s data. Every time you interpret a policy in a way that isn’t written down anywhere, that’s data. Every time you make a judgment call based on experience, that’s data that currently lives nowhere but in your memory.
When HR leaders talk to lawyers, the first question is always “How did you handle this in the past?” That’s exactly what a good agent would do too. Not just asking you, but telling you. “In the past, you handled it this way. The result was this. For consistency, you should consider doing the same thing now.”
An agent can only do that if the data exists. And for most organizations, it doesn’t. So start capturing it! Dictate notes after each case. Transcribe conversations. Save emails into a dedicated folder. Open a document and start logging decisions, resolutions, and precedents. It doesn’t need to be pretty or structured. The models are good enough now to work with messy, unstructured information and still pull out what matters.
Just start.
Demand access to the data you already have
But going back to the ATS and HRIS, of course, there’s value here as well.
Your ATS has interview notes, hiring patterns, time-to-fill metrics. Your HRIS has turnover data, tenure patterns, absence trends. Your engagement platform has survey results going back years. Most of us use these systems for their intended workflows and never think about getting the underlying data out.
This needs to change. When you evaluate vendors, when you renew contracts, when you talk to your account managers, make data portability a requirement. Can they give you API access? Do they support MCP or other integration protocols? Can you at least get clean exports?
If a vendor locks you out of your own data, that’s not just an inconvenience anymore. Going forward, it will be a real competitive disadvantage.
The 12-month window
The agent you want to build today might not work perfectly yet. I’ve been honest about that. Agents in their current state require a lot of hand-holding. It’s like onboarding a new employee who is simultaneously brilliant and has zero context about your organization.
But in 12 months, the models will be better, and the tooling will be better. The organizations that have been collecting data will be ready to move fast. The ones that waited will be starting from scratch.
Your data is the one thing that makes your AI better than everyone else’s AI.
Make sure you own it, collect it and use it.
Also, everyone can copy your ideas and processes. Nobody can copy your data (unless you're sloppy).