Agent Skills
› mohitagw15856/pm-claude-skills
› role-redesign-for-ai
role-redesign-for-ai
GitHub针对AI接管部分工作后的岗位重构技能。通过梳理任务清单、重新定义核心职责与考核指标，明确人机分工，解决能力过剩或期望膨胀问题，并规划职业晋升路径。适用于制定新JD或团队产能规划。
触发场景
AI已显著改变岗位职责时
撰写AI采纳后的新版职位描述
团队询问'现在的工作内容是什么'
规划AI引入后的人力容量
安装
npx skills add mohitagw15856/pm-claude-skills --skill role-redesign-for-ai -g -y
SKILL.md
Frontmatter
{
"name": "role-redesign-for-ai",
"description": "Redesign a job role that AI now does a large part of — deliberately, instead of quietly expecting the same headcount to absorb 140% output. Use when AI has changed what a role spends time on, when writing a revised role charter or job description post-AI, when a team asks 'what is my job now', or when planning capacity after AI adoption. Produces a role redesign: the task inventory before\/after, the redefined core of the role, new expectations and metrics, and the growth-path implications. For hiring rubrics use hiring-rubric; for org-wide skills planning use ai-upskilling or career-ladder-map."
}
Role Redesign For AI Skill
When AI absorbs 40% of a role's tasks, orgs default to the worst option: say nothing, and let expectations quietly inflate until the human is doing their old job plus supervising the machine, evaluated by standards from neither. This skill makes the redesign explicit — what the role stops doing, what it now owns, and what "good" means after the shift.
What This Skill Produces
- A task inventory, before/after: what AI took, what it created, what stayed human — with hours
- The redefined core: the role's new centre of gravity, written as a charter
- New expectations & metrics: what performance means now (and which old metrics are dead)
- Level and growth-path implications — including the junior-pipeline problem, faced honestly
Required Inputs
Ask for (if not already provided):
- The role today: title, level, the real task list (or the JD plus what the JD lies about)
- What AI actually absorbed — observed, not vendor-promised: which tasks, how completely, with what verification burden
- The person/team context: one person or a team of eight? tenure mix? current performance framework?
- The org's honest intent: same headcount doing more? fewer people? higher-value work? (The redesign differs; refusing to pick is itself the problem — flag it.)
Redesign Method
- Inventory tasks, not titles. List the role's tasks with weekly hours. Mark each: AI-absorbed (machine does it, human spot-checks) · AI-assisted (human does it faster) · AI-created (new work: prompting, verifying, correcting, supervising agents) · Human-core (judgment, relationships, accountability, taste). The AI-created column is the one orgs forget — verification is work, and it's in this role now.
- Balance the hours honestly. Old role = 40h. Absorbed −12h, assisted −6h, created +8h → 10h of genuine capacity. The redesign decides where those hours go on purpose: deeper human-core work, wider scope, or reduced load. Unallocated capacity becomes silent expectation inflation within a quarter.
- Redefine the core. The role's new centre is what only it can be accountable for. Write the charter in outcomes: what this role owns (decisions, quality bars, relationships), what it supervises (the AI-done work — with the verification standard stated), what it no longer does (named, so nobody performs it out of habit or fear).
- Rewrite the metrics. Kill throughput metrics the machine now drives (tickets closed, words shipped, drafts produced) — a human evaluated on machine output is being evaluated on prompt luck. New metrics live where the human is: judgment quality (error catch rate on AI output, decision outcomes), the human-core outcomes, and supervision health. Pair with
ai-assisted-performance-reviewfor the review conversation itself. - Face the ladder problem. If AI absorbed the tasks juniors learned on, the pipeline to senior judgment is cut. The redesign states how the next cohort develops: deliberate reps on AI-done tasks (inefficient on purpose), verification apprenticeships, or a redesigned junior role — "we'll figure it out" is how professions hollow out.
- Plan the conversation. The redesign lands as a change to someone's identity, not their task list. The rollout: the draft is discussed with the people in the role before it's announced, the "no longer does" list is framed as release not demotion, and comp/level implications are stated in the same meeting they're wondered about.
Output Format
Role Redesign: [title] — post-AI charter
Intent (stated): [more output / fewer people / higher-value work — the org's actual answer]
Task inventory
| Task | Hrs before | Status | Hrs after | Note |
|---|---|---|---|---|
| (with the AI-created verification/supervision rows present) |
Capacity math: [freed hours → where they were deliberately allocated]
The new charter: Owns: […] · Supervises (with verification standard): […] · No longer does: […]
Metrics: [dead metrics, named as dead · new metrics with definitions]
Ladder implications: [how juniors now develop the judgment this role's seniors have]
Rollout: [discussion-before-announcement plan · the comp/level statement · review date for the charter itself]
Quality Checks
- The AI-created work (verification, supervision) appears in the inventory with hours
- Freed capacity is explicitly allocated — no silent 140% expectation
- At least one legacy throughput metric is explicitly killed
- The "no longer does" list is concrete enough that someone could stop doing those things tomorrow
- The junior-pipeline question is answered, not deferred
- The org's intent (headcount vs scope) is stated in the document
Anti-Patterns
- Do not redesign the role without the people in it — a charter discovered in a reorg deck creates the resistance it deserved
- Do not keep old throughput metrics "for continuity" — they now measure the vendor, not the human
- Do not treat verification as slack time — reviewing machine output at quality is skilled work with hours
- Do not write "focus on higher-value work" without naming the work — that phrase is where redesigns go to die
- Do not skip the intent question — a redesign that won't say whether headcount changes will be read as concealing it, correctly
版本历史
- a38bc30 当前 2026-07-05 11:11
