Seed, Share, and Harden Skill Lifecycle
Start with a standard skill, invite field improvements, then promote the proven version
- Difficulty
- Advanced
- Time to result
- ~ongoing to results
- Steps
- 7
- Confidence
- 98%
The Seed, Share, and Harden Skill Lifecycle begins with a centralized baseline that gives practitioners a credible starting point. Team members remain free to adapt it, build specialized variants, and share useful discoveries with colleagues. This decentralized experimentation generates improvements from people closest to the work. Promising variants are then evaluated, strengthened with clearer instructions and controls, and promoted into an organization-wide library once they reach an acceptable standard. Personal skills, team sharing, and trusted organizational skills form distinct maturity levels rather than competing philosophies. The cycle supports rapid AI adoption while creating a path toward consistency and observability. Standardization follows demonstrated usefulness instead of preceding all experimentation, and the central library continues to evolve as field users discover better techniques.
Origin
Nate Folin explained how Perplexity's sales and RevOps teams create personal skills, share them, harden effective examples, and elevate them to organization skills on Marketing Against The Grain.
Core principles
- 01Give teams a useful starting point instead of a blank page
- 02Allow practitioners to adapt skills to real work
- 03Share improvements so learning compounds across the team
- 04Harden successful variants before organization-wide promotion
- 05Maintain a trusted central layer without suppressing experimentation
How to run it
- 1
Seed a baseline
Create a competent starting skill for a recurring team use case, such as pipeline hygiene or lead leakage.
Pro tip Make the baseline easy to copy and adapt.
Watch out Mandating an immature baseline can freeze a weak process.
- 2
Enable local experimentation
Allow practitioners to modify the baseline or build their own versions around real needs.
Pro tip Ask employees to dogfood the product in their daily work.
Watch out Experimentation needs clear boundaries for sensitive data and external actions.
- 3
Share discoveries
Provide a team space where useful personal skills and techniques can be exchanged.
Pro tip Require a short description of the use case and observed outcome.
Watch out A large uncurated repository can become another form of AI sprawl.
- 4
Evaluate evidence
Assess promising skills for output quality, consistency, safety, and connection to actual results.
Pro tip Test with representative cases and fresh context.
Watch out Popularity is not evidence that a skill is reliable.
- 5
Harden the winner
Improve instructions, edge-case handling, approval gates, and observability for the best-performing variant.
Pro tip Merge useful discoveries without combining genuinely different mechanisms.
Watch out Hardening should not erase specialized variants that solve distinct problems.
- 6
Promote organization-wide
Elevate the proven skill into the trusted organizational library and communicate when it should be used.
Pro tip Version the skill so teams can understand future changes.
Watch out Promotion without ownership leads to stale shared assets.
- 7
Repeat the cycle
Continue gathering improvements from field usage and periodically reevaluate the standard.
Pro tip Track whether the shared skill correlates with desired operational outcomes.
Watch out A best-in-class skill is temporary when models, tools, and processes keep changing.
In the wild
A RevOps leader publishes a prospecting starting point. Sales representatives adapt it, exchange improvements, and create specialized approaches from their calls and accounts. The team evaluates the strongest version, adds controls and clearer instructions, and promotes it for broader use.
→ The organization gains a trusted standard while preserving practitioner-led innovation.
Central skills detect pipeline hygiene issues, lead leakage, stale deals, and emails needing follow-up. Meanwhile, representatives build creative skills for transcript-derived gifts and testimonials, with mature examples later shared across the organization.
→ Core controls remain centralized while novel use cases emerge from the field.
Common mistakes
Mandating too early
Declaring an unevaluated first version to be the universal standard suppresses better field discoveries.
Allowing unmanaged sprawl
Uncurated personal skills become difficult to discover, compare, trust, and maintain.
Promoting without evaluation
A useful anecdote does not prove consistent quality or business impact.
Is it for you?
Best for
Organizations that want reusable best-practice agents without preventing teams from discovering better methods.
Not ideal for
It is poorly suited to tightly regulated workflows where unapproved personal variants cannot safely be used.
From the transcript
“My take on this is I do build something to as a starting point, and oftentimes we have improvements every single week.”
“So people are sharing and going wild, uh, so to speak, and and building their own skills, sharing them with other team members.”
“everybody on the team is developing cool skills and sharing, and then we usually harden them and share them across the entire org once they…”
From the episode
How 1 Human + AI Replaced a 15-Person RevOps Team