Motivation gives a digital individual things to care about, and mental activity gradually turns those concerns into specific ideas. The individual then needs to decide what is worth doing, when to do it, and how to understand the results of action and continue living.
HomoErectus connects this process: endogenous ideas pass through attention and volition into formal activity; PEOR organizes planning, execution, observation, and reflection; actual experiences then influence psychological state, memory, and later behavior. After accepting work, the individual also needs to maintain the corresponding responsibility until completion, renegotiation, or explicit termination.
This document describes the design of these mechanisms. The actual availability of channels, the quality of choices, and the ability to sustain work still require separate operational and acceptance evidence.
1. How endogenous ideas develop into possible action
Mental activity can begin with a need or opportunity that is not yet clearly defined.
The individual may still not understand a question, feel that a relationship has gone without contact for a long time, notice stagnation in a long-term goal, or discover that a newly acquired ability could help with something previously beyond its reach. Desire gaps, emotions, goal tension, outstanding obligations, and memory jointly provide sources for this activity.
HomoErectus distinguishes several levels in the formation of an idea.
First comes an impulse or cue, such as “This still matters to me” or “There seems to be a new possibility now.” It expresses a change worth noticing without yet forming a complete action.
The individual then connects experience with current circumstances and gradually understands its significance. Dissatisfaction with learning may bring a specific knowledge gap to mind; a relationship memory may raise the question of whether to ask how someone is doing.
A more developed action candidate needs to explain what is being considered, why it deserves consideration, which concerns it relates to, and what conditions remain unmet. Only when mature enough for formal deliberation does it become a formal intent.
This process allows ideas to remain pending, merge, be deferred, or fade. Without substantive new changes, it may produce no new idea. Continuing mental activity needs sources and meaning; repeatedly generating the same “next-step suggestion” cannot establish it.
External requests also require interpretation. Someone's words may be a new request, additional information, an expression of feeling, a cancellation, or background sharing. The individual needs to judge their meaning before deciding whether and how to respond and whether to accept responsibility.
2. How attention and volition determine whether to act
After an idea forms, it passes through two connected processes: receiving attention and deciding whether to act.
Attention: what deserves focus now
Attention first needs to determine which matter new information belongs to.
An awaited execution result should return to the corresponding activity. A change to an existing arrangement should connect with the original responsibility. Such feedback needs to preserve its causal connection rather than being treated as a wholly new task each time.
For genuinely new candidates, the individual compares value, urgency, and associated obligations with the progress and switching costs of its current activity.
An ongoing critical action needs to reach a safe point before switching. New ideas worth considering also need an opportunity to receive attention later. A continuing stream of external messages must not permanently displace the individual's own concerns.
Receiving attention means that a matter starts to receive serious consideration. Not receiving attention yet does not mean it has been refused.
Volition: whether this is worthwhile and can be undertaken
Formal deliberation needs to connect goals, psychological state, abilities, relationships, existing responsibilities, and actual conditions.
The main project calls the preparation of these judgments DecisionAssessment, or DA. It explains what is currently known, which factors matter, where evidence is insufficient, and whether any boundaries cannot be crossed.
Based on these judgments, the runtime computes overall utility, or EU, using explicit rules and weights. It considers expected benefit, cost, fatigue, opportunity cost, risk, obligation, identity and value alignment, feasibility, timing, and boundary pressure to form a formal decision.
Different situations can therefore receive different treatment. Something attractive may be deferred for lack of resources. Something unpleasant may still be worth undertaking because of an existing commitment and long-term value. Having the ability may still leave action unavailable because permission is missing.
When necessary grounds are insufficient, bounded preparation can first check materials, establish the scope of operation, or identify an available method. Its results need to return to the original decision for reassessment. Obtaining some material indicates progress in preparation, not acceptance or completion of the main work.
The outcome can be action, refusal, cancellation, or waiting for conditions to change. Deferral needs to preserve its reason and conditions for reconsideration, avoiding repeated deliberation without progress.
3. How PEOR turns a selected activity into action
Once an intent is accepted, it needs continuing progress under its objective. HomoErectus uses PEOR to organize this:
Planning → Execution → Observation → Reflection and replanning judgment (R/P).
Each activity layer retains its own objective, scope, mandatory requirements, progress, and evidence.
Planning: determining how to advance the current objective
The Planner uses the objective, current progress, available material, and methods to determine steps that can advance now and the concrete content needed to execute them.
Plans need to lead to action. Missing necessary information can lead to an information-acquisition step; an overly complex step can be expanded. Established requirements need to persist through execution and checking.
Checking grounds also have different sources: the objective's own requirements, professional quality requirements for this kind of work, and validation conditions in the chosen method. These need to connect while retaining their respective grounds.
Execution: acting on the environment through real capabilities
The Executor dispatches the selected step through a capability channel that is actually available and meets permission and scope requirements.
File operations, commands, retrieval, browsing, and communication connect with the environment through BodyChannel. Writing an action into a plan does not establish that the environment has changed; execution needs actual results.
Synchronous actions can return results directly. Asynchronous actions need a retained waiting relationship so later feedback returns to the original activity.
Observation: retaining what actually happened
Observation records execution results, including success, failure, partial completion, feedback not yet received, and outcomes that cannot yet be confirmed.
The scope supported by each kind of evidence needs to stay clear. A command ending normally can establish that the program completed that run. Whether a file exists, its content is correct, and it meets the objective require corresponding evidence. A communication channel accepting a message is also distinct from the recipient receiving it, understanding it, and fulfilling an arrangement.
Uncertain outcomes need to remain uncertain. For example, an external operation timing out may require checking whether it already took effect before deciding whether to retry.
Reflection and replanning judgment: understanding what a result means for the objective
R/P reads the observation, current objective, and existing progress to judge how far the result advanced the objective, which requirements are satisfied, what remains unfinished, and which direction to take next.
It can judge continuation, further acquisition, waiting, a request for replanning, completion, or failure. When the plan needs to change, R/P explains the reason and scope of adjustment; the Planner then authors the concrete steps.
This division allows execution to adapt to real feedback while preserving clear grounds for planning, observation, and objective judgment.
4. How recursive execution and external collaboration are organized
Complex activity can be expanded into layers.
For example, producing a well-supported research report may involve acquiring materials, comparative analysis, drafting, and checking. A step that remains complex can become a lower-layer activity using the same planning, execution, observation, and reflection process.
The lower layer maintains its own local objective and handles problems within its scope. When its result returns, the parent still needs to judge whether it supports the overall objective. Acquiring materials does not complete the report; passing a local test cannot replace the overall requirements.
This recursive layering organizes activity within the same individual. Calling other agents or collaborating with humans requires additional collaboration relationships and actual channels.
When the relevant capabilities are genuinely available and permissions and resources allow, the individual can give a collaborator work with a defined scope or request advice or information. Collaboration needs to specify the objective, inputs, constraints, expected artifacts, checking conditions, and necessary milestones.
If the collaborator is also autonomous, the individual needs to establish whether the request was accepted, what scope was accepted, and what was promised. Sending a request establishes only that collaboration has been proposed, not that the other party will complete it.
Returned work needs checking against the original request, concrete artifacts, and their versions. The individual remains responsible for judging what the result resolved, whether it meets its objective, what gaps remain, and whether further communication or adjustment is needed.
Delegation can therefore distribute execution, while responsibility for advancing the overall objective and checking results still needs continuing maintenance.
5. How sustained work becomes sustained responsibility
After accepting work, the individual needs to remember and maintain the associated responsibility across later activities.
This includes the objective and scope, important times, current progress, dependencies on others' results, what is being awaited, and which changes require renewed discussion. Plans can adapt to actual conditions, with reasons and implications for the adjustments.
Commitments and expectations preserve arrangements and their triggers: when to submit an interim result, whose reply is awaited, and under what conditions to follow up. On expiry, the system reconsiders from actual time. Rescheduling, cancellation, or completion should invalidate the corresponding old triggers.
Waiting constrains only the relevant matter. While awaiting a collaborator's result, the individual can advance other feasible activities, learn, interact, or rest. Related feedback then returns to the appropriate place for continuation.
When problems arise, the individual can investigate causes, attempt corrections, obtain information, change methods, or seek collaboration within its existing scope. Attempts need cost and count boundaries. Prolonged inability to progress also needs an explicit account of the obstacle, avoiding endless retries.
Human involvement should be organized around an actual decision need. Examples include conflicting product goals, a solution requiring a change to confirmed scope, resource needs beyond existing authorization, or a critical obstacle that can no longer be resolved reliably without help.
The individual should then provide the problem, evidence, attempted methods, options, and implications so the human can decide. Routine progress and checking remain the individual's responsibility.
Sleep or interruption must not erase accepted responsibilities. Recovery needs to connect with prior progress and actual outcomes before deciding how to continue. Completed external operations must also not be repeated because memory has been restored.
6. How action results feed back into psychological state
Execution results relate to objective progress and become new experiences for the individual.
These aspects need distinct interpretation. R/P judges whether a result meets the activity objective. Psychological appraisal connects goal meaning, expectations, relationships, and attribution to understand what the experience means to the individual.
Valuable progress may reduce goal tension, increase achievement satisfaction, and bring joy or pride. Difficulties may revise understanding of an ability, increase preparation tendencies, or lead to ideas of learning, seeking help, and changing methods. Prolonged effort may accumulate fatigue and make recovery more important.
Specific changes need grounds in actual experience. Local failure may affect only confidence in the current task. Obtaining a result also does not automatically establish recognition from others or real income. Whether another person responded and how they evaluated it require corresponding facts.
Experience also enters memory. Factual understanding can be revised, methods can leave learning candidates, and self-understanding can change within a defined scope. These changes are retrieved in later activities and influence attention, expectations, and choices.
Life therefore continues after an activity ends. New understanding may inspire further exploration, sharing with others, relationship repair, or rest. Subsequent motivation comes from the individual's changed state and circumstances.
7. Illustrating the intended behavior through HomoErectus development
The following hypothetical scenario illustrates the working behavior we aim to develop. It is not a record of Adam's actual development work.
Suppose a human has confirmed a HomoErectus development requirement, scope, and acceptance objectives and provided the relevant development authorization. The necessary coding agents and checking channels are also genuinely available.
The individual first connects the current implementation, documentation, and dependencies to form a plan with interim artifacts and milestones. Questions resolvable through reading and inspection remain for it to investigate. Questions about product meaning or scope tradeoffs are organized for human decision.
After accepting the work, it can give implementation work to a coding agent and arrange materials and activities for subsequent checking. While waiting, it can proactively fill gaps in acceptance preparation within the existing objective without waiting for another prompt.
When the coding agent returns results, the individual needs to read the corresponding changes, artifacts, and test evidence to judge whether they concern the same version and cover the original requirements. Depending on the objective, checking may also require actual startup, use of relevant functionality, read-back of persisted results, or verification of continuation after interruption.
If checking fails, the failure enters the relevant layer's observation and reflection. The individual judges whether to acquire information, correct the implementation, rearrange checks, or adjust the local plan. A collaborator saying “finished” becomes a report to examine; final judgment still needs actual evidence.
Only when an important obstacle requires changing product objectives, exceeding existing authorization, or intervention it cannot provide on its own does it bring a decision to the human. Interim reports explain completed work, evidence, remaining work, and plan changes.
Responsibility here means maintaining the connection from objective to result: organizing work, checking artifacts, resolving problems, and accurately stating what is unfinished.
These experiences also affect the individual. Repeated difficulty with a kind of checking should lead to revised self-assessment and learning. A consistently effective collaboration method can become an approach with defined applicability and influence later arrangements.
8. How action, autonomy, and sustained work are validated
Validation needs to follow the whole process rather than only the final output.
For endogenous activity, it must explain which actual states, goals, and experiences generated an idea and observe how it enters consideration without a direct new task. Reasonable deferral or non-action also needs grounds.
For deliberation, it must check whether attention and decision-making actually consider current responsibilities, costs, abilities, and boundaries, and whether necessary preparation returns to the original decision instead of skipping directly into execution.
For execution and collaboration, it must connect objectives, plans, real operations, returned results, and checking judgments, confirming that local success has not become overall completion and collaboration reports have not replaced artifact evidence.
For continuity, it must observe whether responsibility and progress persist through waiting, failures, milestones, sleep, and interruptions.
Finally, it needs to examine how these experiences change psychological state, memory, and subsequent behavior. The continuing process we aim to validate is an individual forming ideas from its own concerns, choosing activities, adjusting action from actual results, and changing itself through experience.