Introducing the MICA Framework

Not every AI is built the same.
Now you can compare them.

The MICA Framework is a 4-factor model for evaluating AI ecosystems on what actually matters: raw intelligence, integration depth, context capacity, and execution power.

Explore the Framework Get Updates

Coined April 27, 2026 — David Offutt, Fort Worth TX

The MICA Framework

Four factors. Every AI comparison.

When someone claims their AI stack is better, this is the scorecard. No hype. Four dimensions that expose what a platform actually delivers.

M
Model Intelligence
Raw cognitive capability

The underlying LLM’s reasoning power, benchmark performance, and depth of training. This is the engine. Everything else is chassis.

I
Integration Depth
How far the AI reaches

How deeply the model connects to your tools, data, workflows, and operating environment. An AI that can’t touch your systems can’t change your outcomes.

C
Canonical Ability
Context window + retrieval capacity

How much the model can hold, reference, and reason across at once. Not just raw context window size — the full retrieval architecture that extends what it “knows” in a session.

A
Agentic Harness
Execution layer around the LLM

The scaffolding that lets the model take action autonomously — tool use, memory, multi-step planning, error recovery. Intelligence without agency is a read-only system.

The “C” Factor

What “Canonical Ability” actually means

Every AI comparison obsesses over benchmark scores. Most practitioners obsess over integrations. Almost nobody asks the question that matters most in real work: how much can this AI actually hold in its head right now?

Canonical Ability is that question made measurable. It combines context window size — how many tokens a model can process at once — with the retrieval architecture layered on top of it. RAG pipelines, memory systems, persistent storage, and multi-turn coherence all live here.

A model with weak Canonical Ability forgets. It loses the thread. It needs re-briefing. A model with strong Canonical Ability carries the full context of your project, your data, and your intent into every response — without you having to repeat yourself.

  • M
    Model Intelligence Raw cognitive capability of the LLM
  • I
    Integration Depth Connectivity to tools, data, workflows
  • C
    Canonical Ability Context window + retrieval capacity
  • A
    Agentic Harness Execution layer — tools, memory, autonomy

MICA Framework — coined April 2026 by David Offutt

The Author

David Offutt
Educator. Agent. Practitioner.
8,000+
Students taught
1
EdTech platform built
TX
Fort Worth based

David Offutt is a Texas entrepreneur who builds things that teach. He founded academytexas.com, a Texas Real Estate Commission-approved continuing education school that has put more than 8,000 agents through coursework.

In his day-to-day, David works as a licensed Farmers Insurance agent in Fort Worth and is retained as an expert witness in Texas insurance litigation cases — which means he reads policy language for a living and testifies about it in court.

He is an active member of Tony’s Skool community, where he translates practitioner-level experience into methods other business owners can actually use.

The MICA Framework came out of a real conversation: David needed a way to explain to other operators why “AI” is not a single thing — and why the ecosystem around a model matters as much as the model itself. He coined “Canonical Ability” to name the gap most comparisons ignore.