How We Define
AI Fluency
AI fluency is, well, fluid.
New models, new standards, and ever-improving tools means that the bar keeps rising. We're constantly evolving Gradient to stay on the frontier, so you and your team can too.
By default, we evaluate candidates on Anthropic's 4Ds, plus one of our own: Deliverable.
Delegation
Setting Goals
Strategically deciding whether, when, and how to engage AI. It involves choosing the right tool for the task and defining the scope of human versus AI responsibility.
Tasked with summarizing 200+ customer feedback responses, a candidate uses AI to categorize and surface patterns but writes the executive summary and recommendations themselves. They know which parts benefit from speed and which need judgment.
Description
Communicating Intent
Clearly articulating goals, context, and desired output formats to the AI. It focuses on effective prompting: setting roles, constraints, and providing examples to get the best results.
Instead of "summarize this feedback," the candidate specifies the format, columns, sorting, and a flagging threshold. The AI delivers something useful on the first pass because the candidate defined what useful looks like.
Discernment
Evaluating Output
Critically assessing AI-generated outputs for accuracy, quality, and alignment with goals. This involves a Description-Discernment loop, where users evaluate and refine outputs rather than accepting the first draft.
The candidate catches the AI flattening two distinct issues into one category and a quote that doesn't match its theme. They push back with specific corrections rather than accepting the first output.
Diligence
Responsible Action
Taking responsibility for the final output, including verifying accuracy, addressing bias, ensuring privacy, and being transparent about the AI's role in the work.
Before sharing the report, the candidate removes a customer's company name from a quote, spot-checks the frequency counts against the source data, and adds a note that the initial categorization was AI-assisted.
Deliverable
Shipping the Work
Producing a polished, tangible output that synthesizes AI-assisted research and generation into something genuinely useful. This is Gradient's own addition: fluency isn't just process, it's results.
The candidate produces a polished feedback report with a clear executive summary, well-organized themes, and actionable recommendations. Ready to present to leadership. The deliverable stands on its own.
Make It Yours
The 5D framework is our default, but you can customize the scoring to match your own AI fluency goals. Here are some frameworks we like for inspiration:
We're not looking for people who know every slash command
The goal isn't to find candidates with hundreds of memorized prompts or a list of AI certifications. Tools change. Interfaces evolve. Models get better.
What matters is judgment and creativity: people who can push your company further in how you think about AI, not just use the tools you already have.