Building in public, Part 1
I've been writing code since 1996.
No degree. Just thirty years of building things. Production Web3 infrastructure, smart contracts across half a dozen chains, full-stack systems that actually shipped to actual users. I've led engineering teams. I've designed protocols. I've been the person who knew how to fix the thing nobody else could figure out.
And right now, if an AI agent tried to find me for a job, it would probably miss me entirely.
That's the problem I want to talk about.
The resume is a document built for humans reading other humans
When it works, it works because a person looks at a list of companies and titles and technologies and constructs a mental model of you. They fill in the gaps with inference. They recognize patterns. They bring context.
They see "20 years of experience, started in 1996, no degree" and they either have the judgment to understand what that means, or they don't. If they do, you get the call. If they don't, you get filtered.
This has always been an imperfect system. But it was a human imperfect system, and humans could be reasoned with, networked around, or occasionally surprised.
That era is ending.
What's actually happening right now
AI agents are entering the hiring pipeline at every layer.
Job descriptions get generated by AI. Applications get screened by AI. Candidate shortlists get assembled by AI. Outreach gets sent by AI. In some cases, entire recruiting workflows run end-to-end without a human touching anything until the final call.
This isn't speculation. I'm building agentic automation inside a legacy media company right now, watching it happen in real time. The infrastructure for agent-driven work is here, and it's being plugged into recruiting because recruiting is expensive, slow, and full of repetitive decisions that pattern-match well.
The problem is that the resume (that PDF, that LinkedIn profile, that structured blob of job titles and bullet points) was never designed to be read by machines with agency.
It was designed to be scanned by humans in under ten seconds.
What an AI agent actually sees when it looks at you
Not what you think.
An AI agent looking for engineering candidates doesn't browse your LinkedIn the way a recruiter does. It's querying for signals. It's looking for structured data it can reason over: technologies, recency, seniority indicators, employment continuity, keyword density.
It doesn't understand that you spent three years in the trenches of a bear market building infrastructure that actually held up. It doesn't know that "lead engineer at [company you've never heard of]" meant you were a team of four and you designed the entire system architecture. It doesn't parse the subtext.
It sees fields. And if the right values aren't in the right fields, you don't exist.
The resume was already a lossy compression of a career. Now we're doing lossy compression on that, and calling it a signal.
The people who get hurt by this first
Not everyone equally.
If you have a clean, linear career at recognizable companies with recognizable titles and a degree from a recognizable school, you probably survive the AI screening layer fine. The system was built around your profile.
If you're like me, with a non-linear path, unconventional background, and deep expertise in domains that are newer than most job description templates, you're invisible to a system that pattern-matches to the median.
And here's the bitter irony: the engineers who are best positioned to build and run agentic AI systems are often the people who got there through unconventional paths. Because the field is new. Because it rewards builders who've operated at the edge of what's possible and figured it out without a roadmap.
Those people are exactly the ones agent-driven screening is most likely to miss.
The deeper issue
The resume doesn't just fail at conveying the what. It completely fails at conveying the how, the why, and the judgment.
It can't tell an agent that you've shipped in production under pressure, that you make good tradeoff decisions, that you're the person who figures out the gnarly thing. It can't communicate that you've been doing agentic AI work before most companies knew what to call it.
It's a static document. It has no interface. It can't answer questions. It can't prove claims. It can't say: here is what I actually know, here is evidence, here is how to verify it.
An AI agent with real capability needs a counterpart that speaks its language. Not a PDF. Not a profile page. Something that can respond to queries, expose structured data, authenticate claims, and communicate value across the protocol layer that agents actually operate on.
That thing doesn't exist yet.
Why I'm building it
I spent the last few months watching this problem from both sides. As someone building agent pipelines at work, and as someone actively navigating the job market with a resume that doesn't adequately capture what I can do.
At some point those two things converged into an obvious question:
What if your professional identity had an API?
Not a profile page that a human reads. An agent-addressable endpoint that speaks MCP, that exposes verified claims, that can respond to structured queries from a hiring agent with real answers, and filter out the noise with something like micropayments so only serious signals get through.
I'm calling it humancard.
I don't have a domain yet. I don't have a product yet. What I have is a clear problem, a protocol stack, and thirty years of knowing how to build things.
This is part one of me building it in public.
If this resonates, follow along. I'll be posting each part of the build as it happens. Next up: why I chose MCP + A2A + x402, and what the architecture actually looks like.