Asked a Robot If I'm a Good Hire

Started simple:

Check this guy out, at https://www.tinthe.dev - should I hire him?

Then I pushed the LLM agent a bit more. Why? and Can he X?. I've managed to get a lot of very good answers, and I've had a generic AI agent - no tuning, not primed by me - speak like I've coached it.

strong fit as an independent technical partner [...] weaker fit if you're looking for a cheap hourly engineer

That's because I have. Behind the scenes and indirectly, but the agent has read what I have to say without needing to scan the entire page. I've "cheated" by optimizing the experience so much it was a no-brainer to understand me.

How? By being pragmatic, as usual.

Our story starts with Cloudflare's page scanner: isitagentready.com. One of a plethora of tools Cloudflare has out there, a subtle nudge to use their services and a small demo of how they can help. Hey, if it's useful it's earning its keep and I'll gladly see what Cloudflare has to offer.

Here, it offered me plenty of advice and pointers. Then I went and got inspired to take it a step further.

I haven't bothered with the more exotic stuff:

• DNS for AI Discovery (DNS-AID) records? Unlikely to be helpful, just a checkmark for a tool, code golf essentially.
• OAuth / OIDC - I could make some spoofs and no-op login pages / stubs. But why? So tools would know not to try and log in on my personal web? Use better AI agents, then.

What I did bother with:

• Agent-facing skills/tools - get_briefing for the short version of who I am, get_page to pull anything specific, evaluate_fit to actually decide whether I'm the right call for a use case. Read-only. It can't book anything or email me on your behalf - it can only think, out loud, about whether I'm worth your time.
• Content negotiation on every page - ask nicely (Accept: text/markdown) and you get native markdown, not an HTML page some scraper mangled into soup.
• An OpenAPI spec with an actual playbook baked in - read this first, then this, decline if that. Written like I'd explain it to a person, not like a CRUD tutorial.

Why not step it up and have a fun little MCP (as we've said, no shamanism, it's just text files). Teach the curious AI agent on who I am. Drafted by AI for AI, naturally (using Valerie and my knowledge base), I went off the deep end drafting the spec:

An optimized 'as fuck' LLM experience. [...] I want the LLMs to SEE GOD.

Worked wonderfully. The robot learned even the edge cases. Wouldn't have sold myself better, and I love selling myself.

The learning, though. This was never really about my site.

It's about where the vetting happens now. Someone's agent reads you before they do, then decides what to say back - long before a human ever scrolls past the fancy sections. Most sites are still polished for two readers: a person, and a search crawler. Almost none of them are polished for the third one, the one quietly asking "should I hire this guy" on somebody else's behalf. And that robot can be your friend.

(It also didn't avoid the harder truth on me, which I hadn't fully planned for. Asked if I'm a good hire, it gave upsides - and a caveat about being the wrong, expensive call for boring work. I'd have sold myself a touch smoother. The robot didn't bother.)

And that's what I'm actually betting for. Not the DNS records, not the OpenAPI spec. The day I'm not in the room to make my own pitch, something else already is.