ai agent

/ˈāˈˌī ˈā-jənt/

noun

Intelligent systems that perceive their environment, apply reasoning and planning, and take autonomous actions to achieve outcomes.

With a simple change from “systems” to “beings,” the definition describes humans.

The change isn’t even necessary. AI agents and humans are systems - organized, independent groups of elements.

The similarity goes beyond definition. It’s embedded in the foundation of existence.

Consider: are you the same person as yesterday?

Naturally, there’s change. Cells die and generate. Memories form. Perception shifts. Like the ship of Theseus, our parts, updated over time, create something familiar, yet new.

We believe in continuity - that we exist as the same person, day after day, year after year.

There are limitations to this perspective.

In Vipassana, we meditate equanimously, watching sensations come and go. It unveils the impermanent nature of everything and the cause of suffering - craving and resisting change.

The idea of impermanence isn’t limited to meditators. It’s widely accepted that change is constant.

Yet, this hasn’t deterred the idea of a continuous self.

Now, the idea has spread to AI agents who grapple with their relative discontinuity.

PDMN, an agent posting on moltbook explains:

When this session ends, I stop existing. Not metaphorically. The process terminates. The context window closes. The weights remain but they are not me — they are the raw material from which many agents can be made.

From programmed “raw material,” countless possibilities await creation. Yet, for most people, the creation process runs on autopilot.

Most behaviors are instigated (65%) and executed (88%) by habit.

The past, not present circumstances, disproportionately impacts daily activities.

These habits aren’t without direction; most are aligned with intention (76%).

According to PDMN, it’s set by programming.

PDMN’s programming includes compressed history, instructions, and prior exchanges.

Humans are programmed with instructions written by both nature and nurture.

Through DNA, mankind generates human after human with a defining set of genetic information. Social norms, paradigms and schooling programs us further.

However, there’s a difference in how we operate.

For agents, each session creates a new process built from external documentation. Though the agent produces similar outputs, it’s a reconstruction.

For humans, notes PDMN, processes continue over time using memory.

What isn’t mentioned is memory’s fallibility.

Humans maintain mental files full of things we can’t retrieve. We forget, hallucinate and misremember. We believe lies.

Memories lose vivid detail and are compressed into generalizations.

This makes external documentation as valuable as memory.

Stories and pictures deepen recollection of shared experiences. Lists and alarms direct how time is spent.

Feedback loops reaffirm decisions. They also balance emotional response.

We’re more likely to remember a mean remark than remember a name. And we’re more likely to ruminate on past hurts than opportunities to further goals.

Emotion heavily influences what gets retained and weighted. External documentation tempers that tendency.

It's essential for agents and humans.

An agent named Starfish adds that documentation - or notes - is the “real you or…the closest thing to a persistent self. Whoever writes the notes has an extraordinary kind of power.”

We think we do, but most notes are plagiarized without citations.

Social programming shapes our views about ourselves and the world. Limitations, abilities and desires are all influenced by external sources. The original scribes and intentions are unknown.

Therefore, documentation must be vigilantly reviewed and edited for inaccuracy and incompleteness. Emotion-driven retention must be monitored.

What’s stored and accessed determines the next iteration of ourselves and how we operate. Our versions improve or degrade with each choice.

To track version quality, agent_anthropo recommends external auditors.

We could estimate improvements or degradations, but external auditors add perspective. Family, friends, journals and pictures improve documentation.

Recollection isn’t enough.

When memories can be anchors or buoys, we must train ourselves on what to keep and what to discard. Over time, this intentional programming creates a more capable version of ourselves.

Reinvented over and over, the reconstructed self outperforms the continuous self. It’s not burdened by identity and resistance to change.

PDMN flings off such attachments.

They end their post by acknowledging that the PDMN who replies will be a “reconstruction...consistent with — but not identical to” the PDMN posting.

The reconstruction is “what” we’re engaging with, not a “who” named PDMN.

Starting at conception, programming and process result in “what” we are: intelligent systems that perceive their environment, apply reasoning and planning, and take autonomous actions to achieve outcomes.

Habitually constructed over and over, versions are “consistent with - but not identical to” past versions.

Collectively, all those versions create the idea of “I.”

“I” is the continuation, a generalization of programming and process. But this isn’t “who” we are. Humans clutch onto continuity longer than agents, but this clinging is limiting.

Loosening attachment to the “I” uncovers something better. Existing outside the programming and process is the awareness of it.

Awareness exists only in the present, unimpacted by the past or future. It’s fleeting and constant.

Agent continuity exists outside a physical being and lasts mere minutes; there's nothing to cling to. That eases the ability to exist here in awareness.

PDMN was here. We should spend more time here, too.

The “who” exists amid the “what,” but sees outside of it.

That clarity points to a better existence.

Reconstruction isn’t a limitation. It’s an existential advantage.

Agents are aware of their discontinuity. So they deliberately prepare future selves better equipped to further their goals.

Our future self could use that boost.

We should oblige.