Naming an agent “Researcher” can quickly produce an assistant that writes research reports. What makes a user return to it is more specific: it knows which claims require checking, how to handle missing material, which conclusions need qualifications, and where its responsibility ends.
Teloa organizes an AI employee’s capabilities around these maintainable working agreements. A model supplies reasoning and expression; a role organizes them into recurring work that can be delegated. The user keeps a responsible participant with defined duties, methods, and action scope, beyond a conversational persona with a particular voice.
Why We Are Building Teloa describes this role through seven elements: Role, Identity, Knowledge, Skills, Tools, Memory, and Permissions. Their significance becomes clear when they work together on one task.
Professional capability shows up in how a problem is handled
Suppose a research employee checks a product update and prepares evidence for an external announcement. The material introduces a new usage condition, while an older description still promises broader applicability. Fluent summaries of both documents do not resolve the problem.
The employee needs to identify the conflict, determine which evidence applies to the current version, and explain where the old promise needs review. With insufficient evidence, it should preserve unresolved points and deliver material the owner can use to make a decision.
Some of this professionalism comes from the model, and some from the working agreement. The same model, given different material, methods, and checking standards, performs different work. Teloa makes those requirements explicit so users can inspect, correct, and reuse them.
How seven elements form a capability together
In this update review, the role first defines the extent of the investigation: check changes and limitations, provide evidence, and leave final external commitments to the owner. Knowledge supplies the versioned material; skills can preserve established methods for comparison and checking conflicts. Without a ready-made skill, the employee can still plan checks within its responsibility and authorization, with the result assessed against delivery requirements.
Tools make the method actionable. External material requires an available connection; files require the corresponding reading capability. Permissions define where those tools may operate. A tool name in a role description identifies a need; availability and authorization still have to be in place.
Identity and memory connect this task to longer collaboration. The user can recognize the original responsible role and keep sourced, applicable experience with it. Which material and experience reaches the next task still depends on actual preparation and reading. An employee name does not put its entire history into every input.
| Element | Question it addresses in this work |
|---|---|
| Role | How far should the review go, and what standard must delivery meet? |
| Identity | Who continues to hold this responsibility, and whose history is this? |
| Knowledge | Which versions of the material support the judgment? |
| Skills | How should differences, evidence, and conditions be checked? |
| Tools | Which actual capabilities obtain material or carry out operations? |
| Memory | Which sourced experience is worth using again within an appropriate scope? |
| Permissions | What may be read or changed, and which decisions return to the owner? |
Their value comes from combination. A known procedure without access to material leaves a prerequisite missing. Complete material without a comparison method can become a summary. Methods and tools without clear responsibility can turn a research conclusion into an action nobody decided to take.
Lasting standards and current requirements need to work together
The role keeps long-term standards such as traceable sources and a distinction between fact and inference. The task specifies which products and release materials to examine and when delivery is due. The user can change the current goal without rebuilding a research employee for every release.
The two can also create tension. An owner may ask for “just one sentence” while the research standard requires evidence. A reasonable response may shorten the main text while preserving inspectable sources and limitations. A professional role gives that trade-off a stable standard to discuss.
Writing every customer condition into the role makes it increasingly specific to the previous task. Leaving all methods in individual tasks loses the benefit of repeated collaboration. Teloa connects the lasting definition with the current delivery while allowing each to change.
This suggests a practical order for expanding capability. Let a delivery expose a gap, then decide whether to add material, improve a method, connect a tool, or narrow responsibility. The number of items in a capability catalog matters less than whether they solve the problem together.
A role can evolve while history keeps its original basis
Repeatedly missing usage conditions may call for a revised skill. A new release may call for updated knowledge. A judgment outside the role’s remit may call for a narrower scope. Independent maintenance locations let users improve a method without copying the entire employee.
Adopting a new version remains a specific choice. Saving a skill update does not switch every existing role to it. Temporarily changing the model in main chat does not silently overwrite the role’s selected configuration either. Changes to methods, models, and responsibilities have different meanings and should not be bundled into one setting.
Past work also needs its original conditions. Adding a usage-limitation check today does not make last month’s report one that performed it. The association between role versions and actual runs provides a basis for review, allowing improvement while explaining how earlier work was completed.
An AI employee earns its lasting place when the next task can begin from standards already developed. In a similar conflict, what to check, how to deliver, and where the owner must decide each have a place for further improvement.