Physix Frontier
Conduct a Replaceability Audit on Yourself
I tinkered with 'Am I replaceable' over the weekend and hit quite a few pitfalls. It's an AI roasting tool that asks how easily you can be replaced by AI. The AI here only judges based on the text you provide; it doesn't check your company's systems. It's more like a mirror: throw your work content in, and let it speak in three tiers: Nice, Honest, and Brutal.
From an organizational level, I don't recommend using it to score colleagues. Team growth is important. It's better suited to remind us which jobs are already proceduralized and which judgments still rely on humans. Below is a walkthrough following an introductory route, from input to improvement, roughly spread over three days.
Day 1: Don't open the webpage yet. Take a piece of paper and write down five things you actually did this week. Don't write 'responsible for project management'; write 'categorized customer feedback into billing, login, and feature requests, then prioritized fixes for developers.' Beginners most often input job titles, and the model will return empty platitudes.
Then open the Am I replaceable entry point. The interface is simple, usually with an input box, a mode selector, and a Generate button. Among the three tiers, Nice is gentle, Honest is straightforward, and Brutal is sharp-tongued. Choose Honest for the first time; don't choose Brutal, as it tends to turn the tool experience into an emotional one.
You can copy this template into the input box: My actual task is categorizing customer feedback into billing, login, and feature requests, and providing fix priorities; I spend several fixed hours weekly judging which customers might churn; please evaluate from a replaceability perspective, giving a score and three reasons.
Click generate and wait a moment. Normal results will give a judgment sentence first, then list reasons. For example, it might say classification can be templated, but judging customer churn requires context. If you only see a score without reasons, it means the task description is too short. Upon seeing this result, copy the original input and output immediately. Here's the pitfall: many people save only the conclusion, not the input. A week later when they want to review, they don't remember what they wrote.
Day 3: Turn the results into an improvement table. Create a document with three columns: Task, What step AI can do, and Boundaries humans must hold. Fill in the five things. You'll find some can be handed to scripts, some require confirming with clients, and some need a manager's nod. You can use the word harness here. In plain English, it means building an evaluation bench for AI that can run, be observed, and be stopped. Its role is to help you judge where AI should plug into the process.
The project README mentions Nice, Honest, and Brutal tiers. Essentially, this adjusts the intensity of criticism without changing facts. The key is which one you're willing to look at.
After a week, you can try it with a small team. First, remove names and sensitive client info. I did a simplified version with a cross-timezone team: each person posted only tasks, not performance reviews, and we looked for repetitive labor together. After this trial, everyone found that meeting minutes at fixed times could indeed be eaten up by summarization tools.
Finally, a few avoidance tips. Job titles are not suitable inputs. Don't share Brutal results in public groups; it hurts team morale. An AI's one-sentence evaluation cannot serve as grounds for layoffs; it lacks business data and doesn't understand the boss's true preferences. If the result is too vague, add time, frequency, and judgment criteria. Remember to save versions for comparison after a week.
After learning this, the next step is to try connecting the task table to WorkBuddy or the README, turning it into a weekly fixed replaceability checkup. From an organizational perspective, I lean towards treating it as a process audit tool, not a tool to judge people. In the next six months, team weekly meetings will likely add an item: discussing which jobs this week already look like assembly lines and which still rely solely on human on-site judgment.
📌 This article is compiled from Hacker News. Original source: https://amireplaceable.app/
Copyright belongs to the original authors. This is a compilation and independent analysis based on public reports.
