Open role
AgentWrangler
Space (sp-ce.co) · Remote · Full-time
TL;DR
- —Space, an AI-native brand agency, is hiring an Agent Wrangler to own the roster of 127 AI agents: what gets built, what earns its slot, and what gets retired.
- —The number on Space's homepage is real and public, and this role is accountable for its integrity, which means it goes down some quarters.
- —The work is evals, drift-hunting, cost-per-outcome accounting, and pairing every agent with a named human owner.
- —Level 5 on Space's published AI Fluency Ladder is the floor for this seat, not the ceiling.
- —Applying is three questions in a short form: one piece you shipped and now hate, the best idea you ever cut, and one thing in culture that makes you jealous.
01
About Space
Space is an AI-native brand agency. Three always-on leaders, 82 on-demand specialists, zero juniors. 127 agents, 362 agentic workflows, 14 tools we built ourselves. We ship brand identities, websites, campaigns, and products for funded companies in weeks: Validere got strategy, a full rebrand, and a new site in 8 weeks, lifted inbound 31%, and closed a $43M Series B on the story. We grow entirely by word of mouth, and we publish our AI rules, our hiring bar, and our pricing logic at sp-ce.co/notes, because an agency that won't tell you where AI stops is telling you it doesn't know.
02
The job
None of our 127 agents has ever been retired, and that ends with you. Some of the 127 carry this company daily. Some are museum pieces from March that nobody audited, because until now nobody owned the roster. The number on our homepage is real, and a real number needs a manager with the nerve to make it smaller.
You'll run the strangest team in the industry: 127 employees who never sleep, never complain, never improve on their own, and never tell you when they've quietly started failing. You hire them by building them. You review them with evals instead of one-on-ones. You promote the ones that earn more responsibility, and you retire the ones that don't, in public, because the count is on the homepage.
Every agent here answers to a named human under our published AI rules. You're the human the whole system answers to: the org chart, the census, and the standard for what deserves to exist as software at all.
03
What you'll do
Run performance reviews for software: evals, error rates, drift, cost per outcome, and the only question that matters, whether anyone would miss it.
Build the next agent only after the case survives you. Most proposed agents should end as proposals.
Retire agents on the record. The homepage number moves when the truth does, in both directions.
Pair every agent with a named human owner, and keep the pairing honest as people and workloads shift.
Hunt drift before clients feel it: the agent that was excellent in April and is confidently mediocre by August.
Publish the census. What we run, what we retired, and why, in the notes, under your name.
04
What you won't do
Worship the count. A quarter where 127 becomes 119 and everything ships faster is a great quarter.
Keep a zombie agent because it demos well. Demo value is not production value, and you'll be the one who says so.
Manage juniors. There are none, and the agents don't count: software doesn't have a career.
Sit in status meetings, log timesheets, or build spec work for pitches. The machines and the model retired all three.
Wait for an agent to fail in front of a client to learn it was failing.
05
Who you are
Level 5 on our published AI Fluency Ladder. This seat builds the ladder's top rung, so the floor here is the other postings' ceiling.
A builder with a full ledger: automations you shipped, and automations you retired with a written reason.
Eval-brained. You'd rather have one honest benchmark than ten impressive demos.
An accountant about it. You can say what an agent costs per outcome and whether a human at that task would be cheaper, and you'll say it either way.
A writer. The census only works if people outside your head can read it.
06
Who you aren't
A collector. Someone who measures progress in agent count built the first 127; defending that number honestly is now the harder job.
A demo artist. Impressive and load-bearing are different careers.
Level 2 on the ladder: someone who ships whatever the model returns. That's worse than not using AI at all, and we said so in print.
Anonymous. Your name goes on the census, and we treat that as the perk it is.
07 · No resume required
How to apply
No cover letter. No 90-page portfolio PDF. No resume upload, even. The application is three questions:
One piece of work you shipped and now hate, with the reason. Two hundred words max.
The best idea you ever cut, and why it was right to cut it.
One thing in culture right now that makes you jealous.
Use AI to answer if you like; we do. We'll be able to tell whether you reviewed it.
Comeargue.
Yes, this job description was drafted with AI. A named human rewrote it, cut a third of it, and will defend every line on a call. That's the entire model, and it's hiring.
Frequently asked questions
Yes. An agent that costs more than it returns, drifts without recovering, or duplicates a better agent gets shut down, and the reason gets written down.
Yes, and it changes. Keeping it honest is the point of this job.
You own the workers; they own the assembly lines. You decide which agents exist and whether they're any good. The Architect decides where agents and humans stand in the flow of work.
Because the seat is sized to the person in it. You'll hear a real range in the first conversation, before you've invested anything beyond the three answers.
A named human reads all three answers. If they're real, we call you.