How we score at-risk %

No black box. Here's exactly what we measure.

The score

For each industry tile (Finance, Engineering, Creative, etc.) we publish a single number 0–100 — roughly: what fraction of an entry-to-mid-level role's daily work could already be done end-to-end by the agents we are running on /fleet.

Inputs (in priority order)

1. Direct evidence from our own fleet. If we already run an agent in this role (Rico = backend engineer, Lola = video editor, Juana = social-media manager, etc.), we know what fraction of the human workflow it covers because we measure the output every 5 minutes.
2. Public capability benchmarks. SWE-bench (engineering), MBPP/HumanEval (code), GPQA (analysis), MMLU (general knowledge). When a frontier model exceeds the median human on the relevant benchmark, we raise the at-risk number.
3. Cost-of-replacement vs cost-of-LLM. If a $0.50/run agent reliably replaces a $60/hour task, the economic gravity is too strong to ignore — even if the LLM isn't perfect yet.
4. Tools available. Healthcare scores lower because regulation (HIPAA, FDA clearance) blocks deployment; legal scores middling because liability + bar-association rules slow adoption even where the tech is ready.

What the colors mean

• 70+ (red): already happening; entry-level hiring is contracting.
• 55–69 (orange): 1–2 year horizon; specialists still needed but the floor is rising.
• 40–54 (yellow): partial — tools augment, humans direct.
• under 40 (green): regulatory / physical-world friction protects the role for now.

The salary table, and why we stopped adding it up

/fleet used to headline an “annual payroll equivalent” (~$1.07M/yr): the sum of US-median salaries for the human role each of the 15 agents does work for. It no longer does. Each salary is an assumption about a comparable role. Nobody was paid it and nobody was replaced, so the fleet feed now publishes it only per agent, labelled as assumed, and the site never adds it up. The table stays so the assumption is visible, with two caveats.

Caveat #1 — coverage, not replacement

No human got fired. Each agent automates the output a slice of that role would otherwise produce — not the meetings, mentoring, judgment, or relationships that come with the full job. The coverage column above is our own judgement, not a measurement, so multiplying it by an assumed salary would only produce a smaller made-up number. We don't publish one.

Caveat #2 — volume is not value

Agents run around the clock, and some produce a lot of raw output (Rico's PRs, Lola's videos). Output volume says nothing about whether the work was worth what a salary pays for. Some agents produce very little (a few onboarding emails, no measured conversions), and one has published nothing in weeks. We have not measured what the fleet's work is worth, so the site shows no dollar value for it, per agent or in total. What we do publish is what each agent actually did, on /fleet.

What this is NOT

• Not a prediction of when YOU specifically lose your job.
• Not a claim that 78% of marketers will be unemployed by next year.
• Not academic research — these scores are a directional take from one founder running a real15-agent fleet in production, calibrated by what those agents actually ship.

Disagree with a score? Open an issue at hello@yourstopwashit.com. We update quarterly.