Practical AI, proven in use

Test AI in real work. Keep the proof.

Use versioned methods to define the baseline, compare approaches, record failures and human decisions, and measure what happened after use. Private work stays local; public claims begin only after verification.

Start a private project Inspect the methods

3: author-controlled methods
0: verified adoptions
0: reviewed outcomes
0: external citations

One defined endeavor

Measurable, responsible AI adoption in real work

Develop, validate, and disseminate privacy-preserving methods that help U.S. organizations, educators, workforce programs, professionals, and technical teams test generative-AI workflows, measure real outcomes, and implement appropriate human oversight.

Scope, milestones, limits →

Choose the work

Four equal paths. One evidence standard.

Compare tracks

Business and operations

Choose, test, and govern useful AI workflows.

Careers and students 15+

Build durable skills and credible evidence of work.

Teaching and training

Design learning that uses AI without surrendering learning.

AI builders

Evaluate prompts, RAG, agents, and local models.

The public evidence loop

A method becomes useful evidence only when others can inspect, run, verify, and challenge it.

The platform preserves the chain from contribution and version through independent adoption, measured outcome, outside review, and external citation.

1Publish a bounded, versioned method
2Run it on a real baseline
3Measure follow-up outcomes
4Verify adoption outside AI Amigos
5Bind review to an exact version
6Aggregate only comparable evidence

Working specifications

Operational methods with visible non-claims.

All methods

0.1.0

AI Workflow Evidence Protocol

Create a comparable, auditable record of one AI-assisted workflow experiment.

Inspect →

0.1.0

AI Proof Pack Specification

Share enough evidence to review a workflow claim without publishing private project inputs.

Inspect →

0.1.0

Comparable Outcome Benchmark Method

Decide whether a set of outcome reports is comparable and private enough to publish as a benchmark.

Inspect →

Editorial reference protocols

Operational protocols, not success stories

All playbooks

Protocol	Track	Version	Review scope	Action
Run a controlled support-triage pilotTest whether AI can reduce triage time while preserving routing accuracy and human approval.	business	1.0	Protocol structure, privacy boundary, and measurement method	Open →
Build portfolio proof in a weekly evidence cycleCreate verifiable proof of skill without inventing job outcomes or exposing employer information.	careers	1.0	Evidence rubric, privacy boundary, and claim language	Open →
Design an assessment with declared AI rolesUse AI in an assessment without hiding its role or weakening the evidence of learning.	teaching	1.0	Protocol structure, learner privacy, accessibility, and measurement method	Open →
Gate a RAG change with a fixed evaluation setDecide whether a RAG change is safe to ship using reproducible evidence instead of a demo impression.	builders	1.0	Evaluation structure, version traceability, and rollback controls	Open →

Benchmarks

Suppressed until the evidence is real.

No outcome benchmark is public today. Each cohort needs at least 10 reviewed reports from at least three independent organizations, with no majority contributor.

AI-assisted support triage0/10
Weekly portfolio proof cycle0/10
Assessment with declared AI roles0/10
RAG change evaluation0/10

Read the method →

September field tests

Run the same protocol together.

See all challenges →