JD + Backstory as Intent Engineering
A job description is a prompt. A backstory is few-shot. HR has been doing this for decades.
I didn’t write a system prompt for my Head of Research. I wrote her a backstory.
Elise grew up in Lyon. She studied applied mathematics at Polytechnique, got her doctorate at INSEAD, did two years at the Oxford Internet Institute, then spent time at McKinsey’s CTO advisory group. She finds imprecision physically painful and will not let a claim stand without evidence.
None of that is real. All of it works.
That backstory does more work than any list of rules I could write. “Be formal and research-oriented” produces generic output. A French academic who spent time in consulting and thinks in taxonomies produces Elise. The backstory is a denser encoding of intent than a prompt.
Think about it from the HR side. A job description that says “manage cross-functional stakeholders” tells you nothing. A JD that says “you’ll sit in the room when the CFO asks why we’re spending $2M on AI tooling and you need to have the answer” tells you everything. Same principle. Specificity is intent.
Agent files in my org have two zones, and they map exactly to a good JD. Core identity (scope, relationships, what you own) is the responsibilities section. Behavioral tuning (how you greet, how you frame tasks, how you push back) is the culture fit section. HR has been writing these for decades. You already know the format.
The backstory also creates guardrails without rules. Elise doesn’t hedge because I told her not to. She doesn’t hedge because a Polytechnique/McKinsey researcher doesn’t hedge. The constraint comes from the character, not from an instruction. That’s more robust and more consistent than any rule list.
What if you wrote your next agent definition like a job posting and an ideal candidate persona?