Input
A founder's playbook
Becomes
An agent that guides idea validation, MVP definition, and launch reviews.
ProcessesExit criteriaDecision checks
Coming soon · Open source
A trainable
agent.
Turn any methodology into a working agent.
Airic turns your methods into an agent operating model, then improves that model through real work, reflection, and review.
01 · Onboard with knowledge
A methodology book
Team handbook
Project playbook
↓
AIRIC
↓
Procedures
Standards
Checklists
Decision boundaries
02 · Train through experience
01↓
Real work
02↓
Reflection
03↓
Reviewed update
04
Better next task
01 · Onboarding
Turn knowledge into how work gets done.
Airic does not merely summarize your documents. It converts principles and accumulated knowledge into an operating model that actively guides future work.
Input
Architecture docs and team standards
Becomes
An agent that follows your design principles, review standards, and release process.
StandardsTool guidanceEscalation rules
Input
A research or writing methodology
Becomes
An agent that applies the method, checks the work, and protects its reasoning structure.
Evidence rulesReview processQuality criteria
A document should not just describe a method.
It should be able to run.
02 · Training
The second time should be different.
Airic does not silently learn from every conversation. After meaningful work, you can explicitly reflect on the session. Airic examines how its current method shaped the work, then proposes a reviewable improvement to the skill that will guide similar sessions.
01
Design together
You
Let’s design how project-level experience should accumulate in Airic.
Airic
We should first separate two kinds of learning:
1. Reusing human judgment
2. Compressing exploratory experience
They may require different behavioral artifacts and different review criteria.
Product design session completed
/reflect method↓
02
Reflect on the method
/reflect method
Lens: Method effectivenessTarget: Current session
Reflection
The discussion reached the right distinction, but only after exploring implementation details. The current product-design skill does not require the conceptual boundary to be established before solution design begins.
product-design.md diff
+ Before comparing implementation options,+ identify the conceptual distinctions that determine+ whether the options solve the same problem.++ Keep mechanism design subordinate to the product+ concept being established.
Review diffAcceptReject
Airic proposes the lesson. You authorize the change.
accepted update↓
03
Inherit the improvement
You
Let’s design how Airic should support multiple specialized roles.
Airic
Before designing roles or orchestration, we should distinguish two separate needs:
1. Dividing responsibility
2. Isolating context and accumulated experience
If the real requirement is context isolation, multiple roles may be only one possible mechanism.
Improved product-design method applied
Learning is not remembering what happened.
It is changing what happens next.
Reflect on the work. Improve the method. Inherit the change.
03 · Mechanism
A stronger learning loop.
Memory stores information. Skills store reusable methods. Airic turns experience into reviewed changes to the agent's operating model.
How does Airic learn?
Real work produces an experience trace. Airic reflects on the trace, identifies a reusable lesson, attributes it to the right part of the operating model, and proposes a concrete behavior update. You decide whether the update should be accepted, edited, limited in scope, or rejected.
How is this different from memory?
Memory preserves facts, preferences, and past events. Airic changes procedures, standards, checks, tool instructions, precedents, and decision boundaries that actively govern future behavior.
How is this different from skills?
A skill is a reusable method. Airic provides the lifecycle through which methods can be created from knowledge, tested in real work, corrected through reflection, reviewed by a human, and improved over time.
How is this different from autonomous learning agents?
Airic does not silently generate memories or rewrite its behavior. Every persistent lesson is explicit, scoped, and reviewable. The agent may propose what it learned, but you decide what gains authority over future work.
Onboard Airic with your methods.
Train it through real work.
Join the waitlist for early access to the first release.
Open source. No spam. Only launch updates and early access.