When AI skills and rules encode your expertise, who owns the brains?
When AI skills and rules encode your expertise, who owns the brains?
The New IP Problem
Here’s the uncomfortable truth about AI-augmented consulting:
Your CLAUDE.md file is your brain on disk.
Every skill you define, every rule you document, every pattern you encode — it’s the distillation of years of experience. And when you share it with a client, you’re not just delivering a project. You’re transferring your intellectual property.
Traditional Consulting:
├── Knowledge lives in consultant's head
├── Client gets deliverables (code, docs)
├── Consultant leaves, knowledge leaves
└── Next engagement: consultant still valuable
AI-Augmented Consulting:
├── Knowledge encoded in skills and rules
├── Client gets deliverables + the encoding
├── Consultant leaves, knowledge stays
└── Next engagement: client can replicate without you
This is a fundamental shift. And most consultants haven’t figured out how to handle it.
What Exactly Are We Protecting?
Let’s be specific about what constitutes consultant IP in the AI age:
Skills Files
# Skill: temporal-join
## Purpose
Generate a temporal join between two SCD2 tables.
## Inputs
- left_entity: The driving entity
- right_entity: The lookup entity
- join_keys: Business keys to join on
- as_of_date: The point-in-time for the join
## Pattern
SELECT
l.*,
r.{attributes}
FROM {left_entity} l
LEFT JOIN {right_entity} r
ON l.{join_keys} = r.{join_keys}
AND l.{as_of_date} >= r._valid_from
AND l.{as_of_date} < r._valid_to
WHERE l._valid_to = '9999-12-31'
## Validation
- Both entities must be SCD2
- Join keys must exist in both entities
- Result must be deduplicated
This skill encodes:
- Pattern knowledge (half-open intervals)
- Best practices (filtering for current records)
- Edge case handling (deduplication)
- Years of debugging temporal joins
Value: Hours saved per use, errors avoided, consistency guaranteed.
Rules Files (CLAUDE.md)
# Data Engineering Standards
## Temporal Queries
- Use half-open intervals: >= _valid_from AND < _valid_to
- Never use BETWEEN for date ranges
- Standard columns: _valid_from, _valid_to, _loaded_at
## SCD2 Patterns
- Always use MERGE for efficiency
- Handle deletes as soft-deletes (set _valid_to)
- Backdated changes require timeline reconstruction
## Quality Checks
- Every entity: unique PK, not-null PK, referential integrity
- Temporal entities: no gaps, no overlaps
- Freshness SLA per entity type
These rules encode:
- Hard-won lessons from production failures
- Industry best practices
- Domain-specific knowledge
- Methodology that took years to develop
Value: Consistency across team, fewer bugs, faster onboarding.
Pattern Templates
-- scd2_merge.sql.j2
MERGE INTO {{ target }} AS t
USING (
SELECT *,
LEAD(_valid_from) OVER (
PARTITION BY {{ business_keys | join(', ') }}
ORDER BY _valid_from
) AS _next_valid_from
FROM {{ source }}
) AS s
ON t.{{ business_keys | join(' = s.') }}
AND t._valid_from = s._valid_from
WHEN MATCHED AND t._hash != s._hash THEN
UPDATE SET _valid_to = s._valid_from
WHEN NOT MATCHED THEN
INSERT ({{ columns | join(', ') }})
VALUES ({{ columns | map('prefix', 's.') | join(', ') }})
This template encodes:
- The exact SCD2 merge pattern that works
- Edge cases handled
- Performance optimizations
- Compatibility with specific platforms
Value: Reusable across every SCD2 implementation.
The Value Chain
Let’s quantify what’s at stake:
┌─────────────────────────────────────────────────────────────────────────────┐
│ CONSULTANT KNOWLEDGE VALUE CHAIN │
├─────────────────────────────────────────────────────────────────────────────┤
│ │
│ Experience (10 years) │
│ ↓ │
│ Patterns (learned from failures) │
│ ↓ │
│ Methodology (systematized approach) │
│ ↓ │
│ Skills & Rules (encoded for AI) │
│ ↓ │
│ Deliverables (generated by AI) │
│ │
│ ───────────────────────────────────────────────────────────────────────── │
│ │
│ Traditional: Client pays for deliverables │
│ AI Age: Client gets deliverables + skills + rules │
│ │
│ Risk: Giving away the source of value, not just the output │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
If you give a client your skills and rules, they can:
- Use AI to generate the same deliverables
- Extend your patterns without you
- Train their team on your methodology
- Never need you again
That might be the goal. Or it might be giving away the farm.
Four Models For Knowledge Protection
Model 1: Full Transfer (Open Source Your Brain)
Approach: Share everything. Skills, rules, templates, methodology. Client owns it all.
Contract Terms:
├── All skills/rules become client property
├── Client can modify, extend, redistribute
├── Consultant retains no exclusivity
└── Higher upfront fee compensates for transfer
When it works:
- One-time engagements with clear end
- Client values independence over relationship
- You’re compensated appropriately (3–5x normal rate)
- You can develop new IP faster than you transfer it
Pricing: €150K-300K for full methodology transfer (not just implementation)
Risk: Client competes with you using your own IP
Model 2: Licensed Access (SaaS Your Brain)
Approach: Client uses your skills/rules but doesn’t own them.
Contract Terms:
├── Skills/rules remain consultant property
├── Client gets usage license (time-limited or perpetual)
├── No modification rights without approval
├── Updates included in license fee
└── Cannot share with third parties
When it works:
- Ongoing relationship expected
- Client wants maintained, updated patterns
- You want recurring revenue
- Clear boundaries enforceable
Pricing: €2–5K/month for access + updates
Risk: Enforcement difficult; client may copy anyway
Model 3: Execution Only (Keep Brain Private)
Approach: You use skills/rules to deliver, but never share the files themselves.
Contract Terms:
├── Deliverables (code, docs) become client property
├── Methodology (skills/rules) remains with consultant
├── Client sees outputs, not inputs
└── Consultant required for future work using methodology
When it works:
- Client values outcomes over capability building
- Complex methodology hard to replicate
- Long-term relationship expected
- You want to remain essential
Pricing: Standard consulting rates (€1–2K/day)
Risk: Client frustrated by dependency; may seek alternatives
Model 4: Tiered Disclosure (Layered Brain)
Approach: Different levels of IP for different relationships.
Tier 1: Public (Blog, Conference Talks)
├── High-level patterns
├── General best practices
├── Conceptual frameworks
└── Builds reputation, attracts clients
Tier 2: Client-Specific (Engagement Deliverables)
├── Applied patterns for their context
├── Customized rules for their platform
├── Implementation of methodology
└── Included in engagement fee
Tier 3: Core IP (Never Transferred)
├── Advanced pattern library
├── Edge case handling
├── Proprietary optimizations
├── Source of competitive advantage
When it works:
- You want to build reputation AND protect IP
- Different clients have different needs
- You can clearly separate tiers
- Enforcement is practical
Pricing: Tier 1 free, Tier 2 included, Tier 3 premium or never
Practical Implementation
What To Share Publicly (Tier 1)
DO share:
- Conceptual explanations (why half-open intervals)
- General patterns (SCD2 overview)
- Best practices (naming conventions)
- Methodology frameworks (phases, maturity models)
Example: This article. The TDS articles. Conference talks.
Value: Establishes expertise, attracts clients, builds community.
What To Deliver To Clients (Tier 2)
DO deliver:
- CLAUDE.md customized for their platform
- Skills for their specific use cases
- Templates applied to their entities
- Rules reflecting their standards
Example:
# ACME Corp Data Engineering Standards
## Platform: Databricks + Unity Catalog
## Temporal Patterns
- Use half-open intervals (applied to their catalog)
- Their specific _valid_from/_valid_to naming
## Quality Rules
- Their SLA requirements
- Their freshness thresholds
- Their referential integrity rules
Value: Client gets working system. You’ve applied your expertise to their context.
What To Never Share (Tier 3)
DON’T share:
- Your complete skills library
- Advanced edge case handling
- Cross-client learnings
- Optimization techniques developed over years
- The “secret sauce” prompts that make AI work well
Example:
# NEVER IN CLIENT DELIVERABLES
## Advanced Temporal Patterns
- Bi-temporal correction handling
- Retroactive change propagation
- Timeline merge algorithms
## Performance Optimizations
- Partition elimination techniques
- Join optimization patterns
- Incremental processing strategies
## AI Prompt Engineering
- The exact prompts that work
- Context structuring techniques
- Error correction patterns
Value: This is what makes you valuable. Keep it.
Contractual Protection
Whatever model you choose, protect it contractually:
IP Ownership Clause
Intellectual Property:
(a) Pre-Existing IP. Consultant's pre-existing intellectual property,
including but not limited to methodologies, patterns, templates, skills
definitions, and AI prompts ("Consultant IP"), shall remain the sole
property of Consultant.
(b) Deliverables. Work product created specifically for Client during
this engagement ("Deliverables") shall be owned by Client upon payment.
(c) Derivative Works. Any modifications or extensions to Consultant IP
created during the engagement shall remain Consultant IP, with Client
receiving a perpetual, non-exclusive license to use such modifications
in connection with the Deliverables.
(d) No Transfer of Methodology. Nothing in this agreement transfers
ownership of Consultant's methodology, including AI skills, rules files,
or prompt libraries, unless explicitly stated in a separate IP Transfer
Agreement.
Confidentiality Clause
Confidentiality:
Client agrees that Consultant's methodology, skills definitions, rules
files, and AI prompts constitute Confidential Information and shall not
be disclosed to third parties or used for purposes other than the
engagement without prior written consent.
Non-Compete Consideration
Be careful here. You want to protect IP without preventing clients from:
- Hiring other consultants
- Developing their own capabilities
- Using general industry knowledge
Focus on protecting specific artifacts, not general concepts.
The Bigger Picture
Here’s the uncomfortable truth:Perfect protection is impossible.
If you teach someone your patterns, they can recreate them. If you show them how you use AI, they can learn it. Knowledge, once shared, spreads.
The question isn’t “how do I prevent all copying?” It’s “how do I stay valuable?”
Stay Ahead
Your value = Current IP + Rate of new IP creation
If you share IP today but create new IP faster than clients can absorb it,
you remain valuable.
Strategy: Continuously develop new patterns, skills, and techniques. Share older IP, keep newer IP private. The frontier is always yours.
Build Relationships
Client loyalty = Technical value + Relationship value + Trust
Even if client could replicate your IP, they may choose not to because
working with you is easier than maintaining it themselves.
Strategy: Be easy to work with. Responsive. Trusted advisor. The relationship itself is IP that can’t be copied.
Create Switching Costs
Switching cost = Learning curve + Integration depth + Customization
If your skills and rules are deeply integrated into client systems,
switching to alternatives is expensive.
Strategy: Integrate deeply. Customize extensively. Make your patterns the foundation.
Key Takeaways
- Skills and rules ARE your IP— they encode years of expertise in files that can be copied
- Choose your model deliberately— full transfer, licensed access, execution only, or tiered disclosure
- Protect contractually— IP ownership, confidentiality, and derivative works clauses
- Layer your disclosure— public concepts, client-specific applications, private core IP
- Stay ahead— create new IP faster than you transfer old IP
- Build relationships— trust and ease of working together can’t be copied
- Perfect protection is impossible— focus on staying valuable, not preventing all copying
The Bottom Line
In the AI age, your brain is on disk. Your skills files are the distillation of everything you know. Sharing them freely is giving away what makes you valuable.
But hoarding everything means no one knows your value.
The answer is deliberate, tiered disclosure:
- Share concepts to build reputation
- Deliver customized applications to clients
- Keep the secret sauce private
And always, always, keep developing new IP. The best protection isn’t a contract — it’s staying ahead.
This article is part of the series on AI-augmented consulting. See also:The End Of Offshore Data TeamsandCoaching Over Contracting.