The Agency Multiplier
Show me what you built before anyone asked
There’s a version of the AI hiring conversation that goes: “Does the candidate know how to use the tools?” That’s the wrong question.
The right question is older. It just matters more now.
Agency. The tendency to act without being asked. To see a problem and build toward it rather than wait for a ticket.
AI didn’t create this distinction. It just made the gap impossible to ignore.
What the Gap Looks Like
A high-agency engineer with Claude Code is terrifying in the best way. They’re not waiting for sprint planning. They’re already three experiments in, with notes on what failed.
A low-agency engineer with the same tools is... where they were before. Maybe faster at the assigned work. But still waiting to be told what the assigned work is.
The floor went up. The ceiling for high-agency people went through the roof. That spread is what you’re hiring into now.
The Tell Is in the Portfolio
You can screen for this. It’s not a trick interview question or a personality test.
Ask to see what they’ve built. Not what they shipped at work. What they built because they wanted to. Outside work “work” is a key indicator.
The low-agency pattern is consistent: the portfolio is all assigned work. When you ask what they’d build with a free month, they pause. They describe something sensible and safe. They ask what stack you prefer before they answer.
The high-agency pattern is also consistent. They show up with something. Not a CRUD app, not a todo list. Something with an edge to it: a tool they were personally annoyed didn’t exist, an experiment they wanted to run, a demo they built to prove a point.
The artifact tells you everything. It shows taste, follow-through, and the ability to manufacture their own motivation.
Why This Is the Actual Filter Now
Skill is more available than it’s ever been. You can spin up a competent implementation of almost anything with the right prompts and a few hours. What AI can’t supply is the judgment about what to build, or the initiative to start before anyone asked.
That’s what you’re paying for. The skills are increasingly table stakes. The agency is the multiplier.
High-agency people use AI to accelerate to the next bottleneck. Low-agency people use it to complete tasks slightly faster. Both outcomes are real. Only one of them compounds.
What to Look For
In practice, the screen is simple: find something they built unprompted. Not a side project they mention in passing and never finished. Something they can demo. Something with real edges, where they made choices.
Then ask one question: “What would you build next with this?”
High-agency candidates already have an answer. Usually two.
On the Team Side
Agency isn’t just a hiring filter. It’s something that compounds or decays depending on the environment.
Teams that share principles (not setups) tend to attract and retain high-agency people. Teams that over-specify tend to filter them out. If your onboarding is a checklist of approved tools and workflows, you’re optimizing for the wrong profile in an era where the most valuable thing an engineer can do is decide what to build next.
The codification paradox bites here too: the more you systematize the how, the more you crowd out the space for why.
The question isn’t whether your team can use AI. It’s whether they’d have built something interesting with it before you handed it to them.
That answer was always available. You just had to look at their GitHub.