AI does consistent work when an organization writes down how its work gets done. Talent systems are where much of that writing already lives.
TL;DR
Most AI adoption stays at the level of individuals typing prompts. The organizations that get consistent results give AI what a new hire needs: a clear map, written standards, and processes people can read and improve. I build that structure using Interpretable Context Methodology, and talent systems are the natural place to start, because a strong competency framework is already that map.
The problem
AI adoption stalls where most change stalls.
How the work gets done usually lives in people's heads. It sits in old documents, in scattered templates, and in the habits of a few experienced managers. When that knowledge stays unwritten, every AI tool starts from zero, results depend on who is typing, and quality drifts from one team to the next.
That makes it an organizational effectiveness problem. The fix is the same one that works for a new hire: role clarity, clear standards, and a process someone can follow on their first day.
When an organization can explain a process to a new hire, it can explain it to AI.
The method
Interpretable Context Methodology, in plain terms.
Interpretable Context Methodology (ICM) organizes AI work the way a well-run team organizes itself. A short map sits at the top. Each area of work has its own context file. Tasks that repeat become reusable skills. Everything lives in plain text that people can read, question, and edit, and that readability keeps people in charge of what the AI knows and how it works.
ICM describes three layers of working with AI. They also describe how far along an organization is:
i.
Chat. Individuals paste prompts into a chat window. Results depend on the person, and each conversation starts fresh. This is where most organizations are today.
ii.
Skills. Proven processes are packaged so anyone can run them, like a coaching prep sequence or an offer review. Quality starts to travel from person to person.
iii.
Structure. Workspaces hold the organization's context: who we are, how the work runs, and what good looks like. AI does consistent work across the whole team, and the people who own each process own its files.
My work sits at the third layer. The job I see for talent and organizational effectiveness leaders is moving teams up those layers, one process at a time.
ICM is the work of Jake Van Clief, published at arxiv.org/abs/2603.16021. The application to talent systems described here is my own.
Applied to talent management
A competency framework is already the map.
ICM puts one reference point at the center and lets every workspace read from it. Strong talent systems work the same way. When the competency framework defines what good looks like at every level, hiring, performance, pay, development, and succession all speak the same language. That shared language is exactly what AI needs to do consistent work.
The map
The competency framework and the organization's priorities. Every process reads from it, so a hiring rubric, a performance review, and a promotion decision all point to the same standard.
Workspaces
One for each talent process: hiring, performance, compensation, development, and succession. Each holds a short context file that works like a process charter, covering the purpose, the steps, what good output looks like, and the lines that hold.
Skills
Packaged, repeatable work for managers and HR partners: preparing a one-on-one, drafting a development plan, reviewing an offer against the band. Built once, used by everyone, and improved in one place.
Standards
Naming conventions and voice guidelines, so every document is easy to find and every message sounds like the organization.
Governance
The rules live in the files the AI reads every time: which decisions stay with a person, which information gets flagged, and where records belong. HR owns those files and can review every one.
See the structure
A sample talent system, file by file.
Here is the structure for a fictional firm, Alder & Finch. It's a set of short, plain-language files. The map routes each request, the shared core defines what good looks like, and each stage has its own folder and owner. Pick any file to read it.
alder-and-finch · sample talent system · fictional company
the map
_core/shared core
hiring/
skills/
onboarding/
coaching/
skills/
performance/restricted
pay/restricted
skills/
development/
skills/
learning-log/learning loop
alder-and-finch/CLAUDE.md
# Alder & Finch Talent System
Alder & Finch is a fictional professional services firm. This file is the map: it describes how the talent program is organized and sends each request to the right place.
## How to work here
1. Read `_core/` before any task. It holds the one definition of good that every stage uses.
2. Go to the stage that owns the request (routing table below).
3. Read that stage's CONTEXT.md, then run the skill it names.
## Routing table
Request
Go to
Owner
Open a role, plan interviews, build a hiring timeline
`hiring/`
Talent Acquisition Lead
Plan a new hire's first 90 days
`onboarding/`
People Partner
Prepare a 1:1 or feedback conversation
`coaching/`
People Partner
Review cycles, ratings, calibration
`performance/` (restricted)
Head of People
Offers, pay changes, band questions
`pay/` (restricted)
Total Rewards Lead
Manager growth plans, readiness checks
`development/`
L&D Lead
## Rules that apply everywhere
- A person makes every final decision on hiring, pay, ratings, and promotion.
- Follow `_core/people-data-policy.md` before touching any employee information.
- Propose changes to `_core/` through `learning-log/changes.md`. The Head of People approves them.
alder-and-finch/_core/CONTEXT.md
# Shared core
Every stage reads these files. They describe what good looks like at Alder & Finch.
File
What it holds
`competency-framework.md`
Six competencies, with observable behaviors by level
`leveling-guide.md`
What changes from Associate to Principal
`values.md`
The three values and how they show up in daily work
`people-data-policy.md`
Who can see what, and what the AI may do with it
Changes here affect every stage, so they go through `learning-log/changes.md` first.
alder-and-finch/_core/competency-framework.md
# Competency framework (excerpt)
Six competencies apply to every role. Each level adds scope.
## Client judgment
- **Associate:** Brings open questions to the right person early.
- **Consultant:** Frames trade-offs for the client and recommends a path.
- **Senior:** Anticipates client risks before they surface.
- **Principal:** Shapes the engagement strategy with client leaders.
## Clear communication
- **Associate:** Writes status updates a teammate can act on without follow-up.
- **Consultant:** Adjusts the message to the audience and the stakes.
- **Senior:** Leads difficult conversations with clients and peers.
- **Principal:** Sets the narrative for the firm on major engagements.
## Also defined in full
Delivery ownership · Developing others · Business acumen · Collaboration
Hiring uses these as interview criteria. Coaching, performance, pay, and development use the same language.
alder-and-finch/_core/leveling-guide.md
# Leveling guide (excerpt)
Associate → Consultant → Senior → Principal. Each level widens scope: own tasks, own workstreams, own client relationships, own firm-level outcomes.
alder-and-finch/_core/values.md
# Values
Clarity first · Own the outcome · Grow together
alder-and-finch/_core/people-data-policy.md
# People data policy
## Access follows role
Data
Who can see it
Interview notes
Hiring team for that role
Performance ratings
The employee, their manager, People team
Pay and offers
Total Rewards, Head of People, the approving leader
Development plans
The employee and their manager
## What the AI may do
- Draft plans, questions, and summaries from the standards in `_core/`.
- Point to the standard behind every recommendation.
- Flag sensitive details (health, protected class, legal matters) and leave them out of drafts.
## What stays with a person
Final decisions on hiring, ratings, pay, and promotion. The AI prepares the work, and a named person decides.
Scores a new manager's answers against the Developing others competency and suggests two focus areas.
alder-and-finch/learning-log/changes.md
# Proposed changes to the shared core
Outcomes from every stage feed back here. A person reviews each proposal before `_core/` changes.
Date
Proposal
Evidence
Status
Mar 3
Add "asks clarifying questions early" to Client judgment, Associate level
90-day reviews: new hires who struggled most rarely asked early
Approved by Head of People
Mar 17
Move Collaboration from stage 3 to stage 2 interviews
Stage 3 ran long in 6 of the last 8 searches
Under review
Apr 2
Retire one Delivery ownership question
Hiring managers rated it the least predictive
Declined: keep one more cycle of data
Three files worth opening
Every name, number, and policy in these files is invented for illustration. The structure follows Jake Van Clief's Interpretable Context Methodology (ICM).
Clarity, Alignment, Trust
The same three ideas, in structure form.
The framework behind my talent work maps directly onto ICM. That fit is why the approach feels natural to me: it asks for the same discipline a good talent system already requires.
Clarity
Context files state what the work is, what good looks like, and where the lines are, so people and AI start from the same understanding.
Alignment
Every workspace reads from one map, so talent systems reinforce each other and stay consistent as the organization changes.
Trust
Every instruction is readable plain text. People can see how the AI works, question it, and correct it, which is what makes the output worth relying on.
What organizations get
Consistency people can see.
i.
Consistent execution across managers. The standard lives in the structure, so every manager gets the same quality of support, whatever their experience with AI.
ii.
Knowledge that stays. Processes are written in plain language, so they stay with the organization when people move on, and new people learn them faster.
iii.
Governance built in. Guardrails sit in the files the AI reads every time, which is where they have the most effect.
iv.
Ownership where the expertise is. HR, learning, and people operations teams own their workspaces directly, because the files are plain English.
v.
A system that improves. When a process changes, one file changes, and every team using it benefits the same day.
How I'd start
One team, one process, then scale.
This is change management with a new kind of team member. It starts small, earns trust with visible results, and grows from there.
1
Pick a repeatable process
Onboarding, interview design, and manager check-ins make strong first candidates: high volume, clear standards, and results people can see.
2
Interview the people who do the work
The goals, constraints, and decisions are already in how people talk about their work. The first job is capturing them.
3
Write the structure
A map, a context file for the process, and one or two skills. Short, plain, and owned by the team that runs the process.
4
Pilot and measure
Set the measures first: time per task, how much of the output gets used as written, consistency across people, and how quickly a new person becomes productive.
5
Refine, then scale
Sharpen whatever was unclear, then use that team's setup as the template for the next one.
In practice
I run my own work this way.
The AI tools on this site were built with this approach, and my day-to-day work runs on it. A single source of truth holds my career history, written standards shape everything I produce, automated checks catch errors before anything goes out, and reusable skills handle the work that repeats. The structure is what keeps the output accurate and consistent.