Enterprise AI needs more than a model. DataKnobs separates three responsibilities that must work together: create the capability, govern what it is allowed to do, and expose the decisions that should remain measurable and adjustable.
From context to action
Why the trio matters
In agentic systems, durable differentiation increasingly comes from enterprise context, data, workflow integration, controls, evaluation, and the ability to change behavior safely. The trio makes those responsibilities explicit.
The platform trio
The source page described KreateData, KreateExperience, and KreateAgents. Those remain useful creation patterns inside KREATE, while KONTROLS and KNOBS now receive equal architectural weight.
KREATE creates the assets users and agents consume: structured data products, knowledge, applications, websites, portals, assistants, agents, and workflows.
KONTROLS defines who can use what context, which actions are permitted, what evidence must be retained, and when people must review or approve a decision.
KNOBS turns design and runtime choices into parameters that can be measured, compared, governed, and changed without rebuilding the whole system.
What is a knob?
A knob can be technical, data-centric, policy-centric, operational, or business-facing. The point is not the UI control; the point is making the choice explicit enough to test, audit, and change.
Sources, lookback windows, filters, ranking, entity scope, features, freshness, and retrieval policies.
Model, prompt, examples, tool strategy, workflow graph, memory, confidence, and evaluation thresholds.
Tool access, autonomy, spend limits, routing, approvals, escalation, safe fallback, and rollback behavior.
Lifecycle
A production AI system should not become less governable after launch. The same knobs and controls used during experimentation should remain observable and adjustable in operation.
Define the user decision, inputs, outputs, owners, risk, and business acceptance criteria.
Create data products, knowledge, experience, agents, and workflow integrations.
Apply access, policy, evidence, human review, audit, and compliance requirements.
Evaluate alternatives, thresholds, prompts, models, tools, cost, latency, and autonomy.
Track drift, quality, business outcomes, and operational burden; adjust approved knobs safely.
Examples
The same responsibilities show up whether the product is a regulated enterprise workflow, a public intelligence product, or a website/agent system.
KREATE: classification and workflow. KONTROLS: privacy, evidence, human review. KNOBS: thresholds, routing, model/prompt variants.
KREATE: earnings and market data products. KONTROLS: source/lineage boundaries. KNOBS: lookbacks, scoring, prompts, strategies, alerts.
KREATE: pages, content, assistants. KONTROLS: claims, SEO, accessibility, privacy. KNOBS: content variants, models, prompts, publishing rules.
KREATE: obligations, evidence, exposure. KONTROLS: policy and approval workflow. KNOBS: source priority, thresholds, judge prompts, exception routing.
Where ABExperiment fits
ABExperiment can compare prompt variants, models, retrieval strategies, UI treatments, thresholds, or other controlled choices. That makes it an important implementation capability for KNOBS and continuous evaluation, while KREATE, KONTROLS, and KNOBS remain the three core responsibilities.
FAQ
They are three platform responsibilities: KREATE builds the capability, KONTROLS governs its execution, and KNOBS exposes important choices so they can be measured, tested, and adjusted.
No. Knobs can represent data, context, prompts, models, retrieval, thresholds, tool access, autonomy, policy, routing, cost, latency, or other consequential behavior.
Experimentation and evaluation are mechanisms for measuring and comparing knobs. ABExperiment can support those workflows, but it is not presented here as a fourth peer pillar.
Use KREATE to create the capability, KONTROLS to define trustworthy execution, and KNOBS to keep quality, risk, cost, speed, and autonomy measurable over time.