Back to the blog

AI Strategy

Fast, Medium, Slow: A Memory Model for Enterprise Agent Teams.

By John Utley

Published

Three streams of task records pass through a human approval gate before becoming lasting memory.

When an operator says give the agent memory, they usually mean one of three different things: keep the thread of this task, remember what happened this week, or change the rules the team will use next quarter. Those should not live in the same store, update at the same speed, or have the same write permissions.

Fast memory is this task. It holds working context, recent signals, and draft lessons. It should be cheap to update and easy to discard.

Medium memory is the daily record. Reviewed notes, pipeline movement, and useful exceptions can carry context between sessions without becoming policy. This layer should be versioned and inspectable.

Slow memory is approved policy, playbooks, and definitions. It should be hard to write to, easy to roll back, and owned by a named person. A lesson should not arrive here because a prompt happened to work once.

The promotion rule is the operating system. A lesson moves down a layer only after human review, with an owner and a log. That is the same idea as Sophizo's R1 to R4 action rules, applied to memory: a person still decides what becomes durable.

ForecastIQ is the reference build, not a separate forecasting product. Weekly pipeline changes stay in the fast layer. Quarter-to-date patterns live in the medium layer. Approved forecasting rules and definitions stay slow, and a person approves changes to that layer. Every forecast can record what was retrieved and why, so a later review can answer which version made the call.

This works with the assistants, retrieval systems, and agent platforms companies already run. You do not need a new foundation model. You need clear layers, write rights, versioning, and a promotion path.

If you cannot list the stores that hold prompts, memory, or retrieved documents, start with the learning-loop scorecard or a Learning-Loop Audit. If you already know the work and need the architecture, the Agent Memory Architecture Sprint designs the layers and the review workflow.

Keep reading

Related articles

A presenter explains three memory layers to a leadership team in a meeting room.
AI Governance

What Google's Nested Learning Paper Means for Mid-Market Teams.

Nested Learning is a research paper, not a product. The practical takeaway for operators is that governance must keep up if AI starts learning after launch, and you can install the controls now.

An operator maps the connected stores used by several AI agents.
AI Governance

Log the Memory, Not Just the Model.

If you only version the model, you cannot explain last month's decision. Log which memory was in force, what was retrieved, and who was allowed to change it.

An operator arranges three layers of agent memory with a visible workflow between them.
AI Strategy

Lesson Promotion: The Review Step Most Agent Teams Skip.

A useful correction is not the same thing as a new policy. Name the person who decides when a lesson may move from a session into durable memory.

Three streams of task records pass through a human approval gate before becoming lasting memory.
AI Governance

Validate Once Is Not Enough: Continual Learning, NIST, and ISO 42001.

Controls written for a frozen model need a second life after launch: versioning, decision logs, and re-validation triggers mapped to NIST and ISO 42001.

Related reading