Community · Vendor Neutral · Working Standard

AI Engineering Standard

Scaffold a production AI harness in five minutes — vendor-neutral working standard for AIDLC

Corpus shipped. What remains Draft is industry adoption — external pilots, shared governance, and inter-org critique — not inventing the practices.

bash — aies harness setup
$npx create-aies-kit
Scaffold AIES rules, skills & prompts in your projectv1.0.0

Maintained in the open. Attribution: project steward · open standard specification.

🧮 Interactive FinOps Calculator

Token Waste & Cost Savings Calculator

Calculate how much your team saves by applying AIES context allowlists, progressive disclosure & model routing.

Read Token Waste Handbook →
$5,000/mo
$200/mo$25,000/mo$50,000/mo
60 KB / prompt
10 KB (compact)100 KB (monorepo dump)200 KB
8 engineers
1 dev25 devs50 devs
Estimated Monthly Savings Under AIES
$3,400 / month
🎉 $40,800 projected annual API savings
Context Reduction
60 KB → 11 KB
82% less token noise
Dev Time Saved
~144 hrs / mo
Fewer iteration loops
Share:X / TwitterLinkedInReddit
🎯 Interactive Diagnostic

AI Engineering Maturity Quiz

Assess where your engineering team lands on the 10-Level AI Maturity Scale in 60 seconds.

1How do you structure context for AI agents or prompts?

2How do you manage agent skills and reusable instructions?

3How do you bound agent iteration loops and API budget spend?

4How do you test prompt & agent changes before deployment?

5What governance & security controls protect your AI harness?

⚡ Core Mission & Justification

Why AI engineering needs a standard

Teams invent private dialects for prompts, agents, evals, and approvals. The result is unreproducible delivery, silent security gaps, and tools that cannot interoperate. AIES exists so organizations share one vocabulary, one set of gates, and one way to evidence production readiness — without locking to a single model vendor.

“Trust comes from documentation, versioning, and community critique — not from declaring ‘the standard’ on day one.”

💰 Quantified Enterprise Value & Cost Savings

Why adopting a standard harness pays for itself at every AIDLC step

Standardizing prompts, context windows, MCP tools, and eval gates eliminates trial-and-error engineering, slashes token bills, and guarantees audit readiness.

📉 42% Lower
Defect Density & Regressions
Automated CI/CD eval gates block prompt hallucinations and breaking model upgrades before merge.
💰 35%+ Saved
LLM API Token Bills
50/50 context budget rules eliminate window bloating and prevent costly token waste.
⚡ 2.5x Faster
Time-to-Market Delivery
Instant scaffolded IDE harnesses eliminate custom setup scripts for Cursor, Copilot & Claude Code.
🛡️ 4 Days
Audit Readiness Time
Machine-readable evidence files provide immediate compliance proof for NIST AI RMF & EU AI Act.

Step-by-Step AIDLC ROI Breakdown

Standard vs Ad-Hoc
AIDLC StageAd-Hoc Engineering RiskAIES Standardized PracticeQuantified Impact
1. Idea & ArchitectureTeams invent private prompt/agent dialectsStandard harness profile (`AIES.md` + `AGENTS.md`)50% faster setup
2. Design & ContextBloated context windows cause lost-in-the-middle decay50/50 context layout & token budget rules35% token savings
3. Build & MCP ToolsUnrestricted tool execution (indirect prompt injection)Bounded MCP tool contracts & authorization schemasZero OWASP LLM01 gaps
4. Test & EvaluationManual spot testing before production deployCI/CD quality gates (`aies audit` + Promptfoo)42% fewer regressions
5. Deploy & GovernanceManual 6-week security and regulatory binder gatheringNIST AI RMF & EU AI Act machine evidence bindingAudit readiness in 4 days

Engineering principles

The constitution every AIES artifact should obey.

Full principles →
Human firstAI assistedReproducibleObservableTestableDeterministic where it mattersVendor neutralSecure by defaultContext rich, noise poorDocumentation driven

Community & Contributing

AIES improves through issues, PRs, pilot evidence, and disagreement on the record.