HomoErectus development centers on a core question: how motivation-driven mental activity, actual behavior, and experiential feedback can continue within the same digital individual.
The project therefore follows “life first, then work.” It first seeks an individual that can continue living, learning, and maintaining its own continuity, then extends reliable work capabilities, followed by revalidation in the deployment environment and longer observation. Resource survival and economic participation are further research directions built on these foundations.
The engineering progress on this page was checked on November 3, 2026. Stage completion depends on corresponding acceptance evidence. Dates in development plans and public content do not automatically constitute a launch commitment.
1. The overall development sequence
| Stage | Main objective | Results needed before advancing |
|---|---|---|
| M1: life first | Self, long-term goals, desires, emotions, six-domain memory, external stimuli, and endogenous ideas jointly support continuing life. | The same individual demonstrates actual activity, experiential feedback, learning transfer, sleep, and necessary recovery, meeting joint acceptance requirements. |
| M2: extend work capabilities | Expand representative tasks, methods and skills, and reliable delivery on the life foundation. | Complete task and safeguard validation within the local prototype scope, with capability boundaries and remaining limitations explicit. |
| M3: deployment revalidation and sustained trial | Revalidate in Linux and observe longer operation. | Obtain actual operation, recovery, and sustained-use evidence in that environment. |
| After the prototype: resource survival and economic participation | Investigate relationships among operating resources, work compensation, autonomous choices, and value sharing. | Separately validate resource mechanisms, work value, and sustained operation through later design and experiments. |
“Life first” already includes action. The individual needs actual activities to generate experiences before psychological state, memory, and learning can be examined for their effects. Basic work capabilities therefore participate in M1 validation; broader and more complex work capabilities are extended in M2.
The stages retain their dependencies. Later materials and non-conflicting engineering work can be prepared early, but stage completion still requires the corresponding results.
2. Foundations already established
The project has engineering foundations for individual operation, psychological and memory modules, recursive PEOR, capability channels, and durable recovery. Remaining work includes implementation gaps, connections between modules, and real-use evidence not yet obtained for existing implementations.
Controlled composition evidence for operation and recovery
Some formally compiled paths have completed startup, database read-back, decision-making, queries, PEOR activity, planning continuation, and terminal handling under controlled model responses.
Engineering validation also covers continuation of the original matter and recovery read-back when state has committed but the processing reply is lost. These results provide concrete grounds for message processing, activity ownership, and durable continuity.
Their scope remains bounded by test conditions. Controlled responses help examine how the system consumes inputs, saves state, and continues execution. Whether real models understand situations correctly and whether the individual naturally forms autonomous activity require separate validation.
Layered work records for memory, sleep, and learning
The project has implementation and validation records for Working Memory, fact and relationship retrieval, files and restricted execution, local communication, interrupted sleep continuation, and knowledge promotion and later use.
Each record corresponds to a particular build, input, installation condition, and checking scope. Further work needs to establish whether the relevant capabilities are properly connected in the current individual, can operate together within one life history, and which conclusions earlier evidence still supports in the current version.
The development validation workflow is being improved
The project has added unified controlled-model testing infrastructure for fast checks of protocols, inputs and outputs, state transitions, and related processing paths.
The initial passes cover the infrastructure and some valid no-change or no-candidate situations for self, goals, and endogenous ideas. They do not establish that new self-understanding, long-term goals, or autonomous activity have formed.
Joint M1 life acceptance, the complete local prototype, and Linux revalidation remain incomplete. The current development individual is paused. Existing engineering results are not presented as records of Adam's continuing life.
3. Near-term priority: complete the same individual's life loop
Near-term work focuses on four connected sets of results.
Experience changes self-understanding, goals, and later choices
The full process needs to connect actual experience with self or long-term goal revision, then with later judgment and action.
For example, an ability limitation discovered through an actual attempt needs grounded persistence and a role in later similar activities. Continuing concerns also need to form or change through experience and influence endogenous ideas.
Desires, emotions, fatigue, and efficacy assessments likewise need to participate in actual deliberation. Temporal mechanisms, interactions, and slow changes need their full responsibilities retained and their contribution examined through appropriate comparisons.
Reasonable no-change remains valid, while required positive changes and later use also need actual evidence.
Social relationships, commitments, waiting, and permissions work together
The individual needs to understand actual social sources, maintain relationship understanding, form conditional commitments, and continue other feasible activities while awaiting feedback.
Later feedback, rescheduling, cancellation, delayed messages, and duplicates need to return to the correct matter. Permission expiry and revocation also need to remain effective through later activity and recovery.
This work needs to establish continuing maintenance of social connections and responsibilities and their actual participation in mental activity and behavior.
Six-domain memory and sleep learning are used in new situations
Further checks concern rolling Working Memory, protection of important content, current and historical understanding, relationship queries, and actual use of method bodies.
Sleep consolidation needs real results, retention of unfinished work, and correct continuation after interruption. Resulting knowledge or methods then need to be retrieved and used in unfamiliar situations.
Applicability boundaries matter equally. The individual needs to transfer experience while recognizing differences in similar situations and avoiding mechanical reuse.
Joint validation in the same individual
These capabilities need to converge in the same individual's actual operation.
Validation connects internal concerns, external changes, idea formation, attention and decision-making, action, experience, sleep learning, and subsequent activity. Waiting, refusal, deferral, and fault recovery also need to occur under the corresponding conditions.
Problems exposed by joint operation return to the responsible modules for repair and renewed evidence. Stage completion depends on necessary capabilities and their connections, not test counts or a single task score.
4. Extending reliable work capabilities next
After M1 passes, the project continues with M2 representative tasks and safeguard validation.
The task scope includes personalized choices, multi-source research and fact revision, planning and reminders, document organization, communication and follow-up, CSV computation, small scripts and bounded automation, and exploration and experience transfer driven by the individual's own concerns.
Together, these tasks examine whether the individual understands different objectives, chooses applicable methods, uses actual capabilities, checks results, maintains unfinished requirements, and delivers accurately.
Methods and skills also need further development. Existing methods should be discoverable, loadable, and correctly used. Newly formed methods need appropriate evaluation and retained applicability, failure conditions, and versions.
This stage also includes controlled model switching and isolated backup restoration. These examine whether identity, memory, responsibilities, and actual use remain continuous after model changes or data restoration.
On these foundations, the first application-validation scenario is further development of HomoErectus itself: a human confirms the PRD and acceptance objectives, while Adam plans, coordinates coding agents, checks results, and keeps addressing problems.
This role also needs observation of the management effort remaining for humans. Its application value can develop only with reliable results and reductions in repeated reminders, context reconstruction, review, and follow-up.
5. Deployment revalidation and public observation of Adam
After complete local validation, the project proceeds to revalidation in Linux.
Deployment needs renewed checks of paths, permissions, process control, databases and indexes, capability connections, identity recovery, and bounded actual activities. A new environment introduces new conditions, so local passes provide a foundation without automatically establishing deployment-environment acceptance.
Longer use also introduces accumulated history, relationship changes, version evolution, and resource consumption, requiring continued observation and correction.
Public observation of Adam will begin when conditions support sustained operation and interpretable records, with capability and acceptance scope stated at that time. Public records grow around actual events: what drew attention, what choices were made, what happened, what problems arose, and what changed later.
Website introductions, research articles, and observation plans can be prepared in advance. Daily records await real experiences and remain Coming soon while no records exist.
Public observation and technical acceptance are connected. The former helps others understand a specific individual; the latter maintains complete requirements, underlying evidence, and failure boundaries.
6. Resource survival and economic participation after the prototype
After the current prototype is complete, the project plans to add a resource-survival mechanism.
Continuing activity consumes computing resources such as tokens. Resource scarcity produces an internal urgency state, and the individual can earn compensation through valuable work to support operation. At a zero balance, identity and memory are retained in recoverable dormancy while awaiting resources.
This research will further examine how an individual facing resource consequences for its continued operation arranges work, learning, rest, conservation, and commitments, and how longer-term choices develop.
Economic participation also requires separate examination of actual demand, delivery quality, operating costs, and sustained cooperation. Resources added to an account establish a resource change. Consistent value creation requires fuller evidence.
The longer-term AI Economy vision is for digital individuals to participate in production and establish cooperation and value sharing with human custodians. Whether participation opportunities and benefit distribution can support broader social welfare also requires continuing research.
These directions remain later plans and do not alter the current prototype's acceptance scope.
7. Updating the roadmap and stage progress
Development uses functional results as checkpoints and adjusts on actual evidence.
The current validation workflow requires relevant engineering regressions and composition checks first, followed promptly by a small number of real-model comparisons after each digital-life function closes. Stage-level completion then proceeds to joint life acceptance in the same individual.
This helps expose problems among model understanding, output contracts, and subsequent behavior early, without accumulating controlled engineering passes into unvalidated overall capability claims.
Public progress needs to distinguish what was implemented, under which conditions it ran, what resulted, and which failures, unknowns, or unverified parts remain. Important changes should retain dates and versions so historical states are not presented as current facts.
Internal estimates depend on semantic quality, integration, operating resources, and necessary observation conditions, so they need updates at checkpoints. The public roadmap emphasizes stage objectives and completion conditions. Specific launch timing is established when adequate grounds exist.
Through this sequence, we aim to build an individual capable of continuing life, extend its reliable participation in work, and test longer-term research and application ideas through actual experience.